1 /*
2 * Copyright (c) 1999, 2021, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/unsafe.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/stubRoutines.hpp"
58 #include "utilities/macros.hpp"
59 #include "utilities/powerOfTwo.hpp"
60
61 #if INCLUDE_JFR
62 #include "jfr/jfr.hpp"
63 #endif
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return NULL;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
85 !C->directive()->is_intrinsic_disabled(mh) &&
86 !vmIntrinsics::is_disabled_by_flags(mh);
87
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return NULL;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
124 if (C->print_intrinsics() || C->print_inlining()) {
125 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
126 }
127 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
128 if (C->log()) {
129 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
130 vmIntrinsics::name_at(intrinsic_id()),
131 (is_virtual() ? " virtual='1'" : ""),
132 C->unique() - nodes);
133 }
134 // Push the result from the inlined method onto the stack.
135 kit.push_result();
136 C->print_inlining_update(this);
137 return kit.transfer_exceptions_into_jvms();
138 }
139
140 // The intrinsic bailed out
141 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
142 if (jvms->has_method()) {
143 // Not a root compile.
144 const char* msg;
145 if (callee->intrinsic_candidate()) {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
147 } else {
148 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
149 : "failed to inline (intrinsic), method not annotated";
150 }
151 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
152 if (C->print_intrinsics() || C->print_inlining()) {
153 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
154 }
155 } else {
156 // Root compile
157 ResourceMark rm;
158 stringStream msg_stream;
159 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
160 vmIntrinsics::name_at(intrinsic_id()),
161 is_virtual() ? " (virtual)" : "", bci);
162 const char *msg = msg_stream.as_string();
163 log_debug(jit, inlining)("%s", msg);
164 if (C->print_intrinsics() || C->print_inlining()) {
165 tty->print("%s", msg);
166 }
167 }
168 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
169 C->print_inlining_update(this);
170
171 return NULL;
172 }
173
174 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
175 LibraryCallKit kit(jvms, this);
176 Compile* C = kit.C;
177 int nodes = C->unique();
178 _last_predicate = predicate;
179 #ifndef PRODUCT
180 assert(is_predicated() && predicate < predicates_count(), "sanity");
181 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
182 char buf[1000];
183 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
184 tty->print_cr("Predicate for intrinsic %s", str);
185 }
186 #endif
187 ciMethod* callee = kit.callee();
188 const int bci = kit.bci();
189
190 Node* slow_ctl = kit.try_to_predicate(predicate);
191 if (!kit.failing()) {
192 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
193 : "(intrinsic, predicate)";
194 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
195 if (C->print_intrinsics() || C->print_inlining()) {
196 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
197 }
198 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
199 if (C->log()) {
200 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
201 vmIntrinsics::name_at(intrinsic_id()),
202 (is_virtual() ? " virtual='1'" : ""),
203 C->unique() - nodes);
204 }
205 return slow_ctl; // Could be NULL if the check folds.
206 }
207
208 // The intrinsic bailed out
209 if (jvms->has_method()) {
210 // Not a root compile.
211 const char* msg = "failed to generate predicate for intrinsic";
212 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
213 if (C->print_intrinsics() || C->print_inlining()) {
214 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
215 }
216 } else {
217 // Root compile
218 ResourceMark rm;
219 stringStream msg_stream;
220 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
221 vmIntrinsics::name_at(intrinsic_id()),
222 is_virtual() ? " (virtual)" : "", bci);
223 const char *msg = msg_stream.as_string();
224 log_debug(jit, inlining)("%s", msg);
225 if (C->print_intrinsics() || C->print_inlining()) {
226 C->print_inlining_stream()->print("%s", msg);
227 }
228 }
229 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
230 return NULL;
231 }
232
233 bool LibraryCallKit::try_to_inline(int predicate) {
234 // Handle symbolic names for otherwise undistinguished boolean switches:
235 const bool is_store = true;
236 const bool is_compress = true;
237 const bool is_static = true;
238 const bool is_volatile = true;
239
240 if (!jvms()->has_method()) {
241 // Root JVMState has a null method.
242 assert(map()->memory()->Opcode() == Op_Parm, "");
243 // Insert the memory aliasing node
244 set_all_memory(reset_memory());
245 }
246 assert(merged_memory(), "");
247
248 switch (intrinsic_id()) {
249 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
250 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
251 case vmIntrinsics::_getClass: return inline_native_getClass();
252
253 case vmIntrinsics::_ceil:
254 case vmIntrinsics::_floor:
255 case vmIntrinsics::_rint:
256 case vmIntrinsics::_dsin:
257 case vmIntrinsics::_dcos:
258 case vmIntrinsics::_dtan:
259 case vmIntrinsics::_dabs:
260 case vmIntrinsics::_fabs:
261 case vmIntrinsics::_iabs:
262 case vmIntrinsics::_labs:
263 case vmIntrinsics::_datan2:
264 case vmIntrinsics::_dsqrt:
265 case vmIntrinsics::_dsqrt_strict:
266 case vmIntrinsics::_dexp:
267 case vmIntrinsics::_dlog:
268 case vmIntrinsics::_dlog10:
269 case vmIntrinsics::_dpow:
270 case vmIntrinsics::_dcopySign:
271 case vmIntrinsics::_fcopySign:
272 case vmIntrinsics::_dsignum:
273 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
274
275 case vmIntrinsics::_notify:
276 case vmIntrinsics::_notifyAll:
277 return inline_notify(intrinsic_id());
278
279 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
280 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
281 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
282 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
283 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
284 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
285 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
286 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
287 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
288 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
289 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
290 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
291 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
292 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
293
294 case vmIntrinsics::_arraycopy: return inline_arraycopy();
295
296 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
297 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
298 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
299 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
300
301 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
308 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
309
310 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
311 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
312
313 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
314 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
315 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
316 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
317
318 case vmIntrinsics::_compressStringC:
319 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
320 case vmIntrinsics::_inflateStringC:
321 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
322
323 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
324 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
325 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
326 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
327 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
328 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
329 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
330 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
331 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
332 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
333 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
334 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_PRIMITIVE_OBJECT,Relaxed, false);
335
336 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
337 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
338 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
339 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
340 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
341 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
342 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
343 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
344 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
345 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_PRIMITIVE_OBJECT,Relaxed, false);
346
347 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
348 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
349 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
350 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
351 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
352 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
353 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
354 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
355 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
356
357 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
358 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
359 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
360 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
361 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
362 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
363 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
364 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
365 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
366
367 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
373 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
374 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
375 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
376
377 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
378 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
379 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
380 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
381 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
382 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
383 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
384 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
385 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
386
387 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
388 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
389 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
390 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
391 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
392 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
393 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
394 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
395 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
396
397 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
398 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
399 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
400 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
401 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
402 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
403 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
404 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
405 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
406
407 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
408 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
409 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
410 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
411 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
412 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
413 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
414 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
415 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
416
417 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
418 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
422
423 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
435 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
439 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
440 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
441 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
442 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
443
444 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
450 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
459
460 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
461 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
464
465 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
466 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
470
471 case vmIntrinsics::_loadFence:
472 case vmIntrinsics::_storeFence:
473 case vmIntrinsics::_storeStoreFence:
474 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
475
476 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
477
478 case vmIntrinsics::_currentThread: return inline_native_currentThread();
479
480 #ifdef JFR_HAVE_INTRINSICS
481 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
482 case vmIntrinsics::_getClassId: return inline_native_classID();
483 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
484 #endif
485 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
486 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
487 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
488 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
489 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
490 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
491 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
492 case vmIntrinsics::_getLength: return inline_native_getLength();
493 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
494 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
495 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
496 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
497 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
498 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
499 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
500
501 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
502 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
503
504 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
505
506 case vmIntrinsics::_isInstance:
507 case vmIntrinsics::_getModifiers:
508 case vmIntrinsics::_isInterface:
509 case vmIntrinsics::_isArray:
510 case vmIntrinsics::_isPrimitive:
511 case vmIntrinsics::_isHidden:
512 case vmIntrinsics::_getSuperclass:
513 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
514
515 case vmIntrinsics::_asPrimaryType:
516 case vmIntrinsics::_asValueType: return inline_primitive_Class_conversion(intrinsic_id());
517
518 case vmIntrinsics::_floatToRawIntBits:
519 case vmIntrinsics::_floatToIntBits:
520 case vmIntrinsics::_intBitsToFloat:
521 case vmIntrinsics::_doubleToRawLongBits:
522 case vmIntrinsics::_doubleToLongBits:
523 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
524
525 case vmIntrinsics::_numberOfLeadingZeros_i:
526 case vmIntrinsics::_numberOfLeadingZeros_l:
527 case vmIntrinsics::_numberOfTrailingZeros_i:
528 case vmIntrinsics::_numberOfTrailingZeros_l:
529 case vmIntrinsics::_bitCount_i:
530 case vmIntrinsics::_bitCount_l:
531 case vmIntrinsics::_reverseBytes_i:
532 case vmIntrinsics::_reverseBytes_l:
533 case vmIntrinsics::_reverseBytes_s:
534 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
535
536 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
537
538 case vmIntrinsics::_Reference_get: return inline_reference_get();
539 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
540 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
541
542 case vmIntrinsics::_Class_cast: return inline_Class_cast();
543
544 case vmIntrinsics::_aescrypt_encryptBlock:
545 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
546
547 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
548 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
549 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
550
551 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
552 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
553 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
554
555 case vmIntrinsics::_counterMode_AESCrypt:
556 return inline_counterMode_AESCrypt(intrinsic_id());
557
558 case vmIntrinsics::_galoisCounterMode_AESCrypt:
559 return inline_galoisCounterMode_AESCrypt();
560
561 case vmIntrinsics::_md5_implCompress:
562 case vmIntrinsics::_sha_implCompress:
563 case vmIntrinsics::_sha2_implCompress:
564 case vmIntrinsics::_sha5_implCompress:
565 case vmIntrinsics::_sha3_implCompress:
566 return inline_digestBase_implCompress(intrinsic_id());
567
568 case vmIntrinsics::_digestBase_implCompressMB:
569 return inline_digestBase_implCompressMB(predicate);
570
571 case vmIntrinsics::_multiplyToLen:
572 return inline_multiplyToLen();
573
574 case vmIntrinsics::_squareToLen:
575 return inline_squareToLen();
576
577 case vmIntrinsics::_mulAdd:
578 return inline_mulAdd();
579
580 case vmIntrinsics::_montgomeryMultiply:
581 return inline_montgomeryMultiply();
582 case vmIntrinsics::_montgomerySquare:
583 return inline_montgomerySquare();
584
585 case vmIntrinsics::_bigIntegerRightShiftWorker:
586 return inline_bigIntegerShift(true);
587 case vmIntrinsics::_bigIntegerLeftShiftWorker:
588 return inline_bigIntegerShift(false);
589
590 case vmIntrinsics::_vectorizedMismatch:
591 return inline_vectorizedMismatch();
592
593 case vmIntrinsics::_ghash_processBlocks:
594 return inline_ghash_processBlocks();
595 case vmIntrinsics::_base64_encodeBlock:
596 return inline_base64_encodeBlock();
597 case vmIntrinsics::_base64_decodeBlock:
598 return inline_base64_decodeBlock();
599
600 case vmIntrinsics::_encodeISOArray:
601 case vmIntrinsics::_encodeByteISOArray:
602 return inline_encodeISOArray(false);
603 case vmIntrinsics::_encodeAsciiArray:
604 return inline_encodeISOArray(true);
605
606 case vmIntrinsics::_updateCRC32:
607 return inline_updateCRC32();
608 case vmIntrinsics::_updateBytesCRC32:
609 return inline_updateBytesCRC32();
610 case vmIntrinsics::_updateByteBufferCRC32:
611 return inline_updateByteBufferCRC32();
612
613 case vmIntrinsics::_updateBytesCRC32C:
614 return inline_updateBytesCRC32C();
615 case vmIntrinsics::_updateDirectByteBufferCRC32C:
616 return inline_updateDirectByteBufferCRC32C();
617
618 case vmIntrinsics::_updateBytesAdler32:
619 return inline_updateBytesAdler32();
620 case vmIntrinsics::_updateByteBufferAdler32:
621 return inline_updateByteBufferAdler32();
622
623 case vmIntrinsics::_profileBoolean:
624 return inline_profileBoolean();
625 case vmIntrinsics::_isCompileConstant:
626 return inline_isCompileConstant();
627
628 case vmIntrinsics::_countPositives:
629 return inline_countPositives();
630
631 case vmIntrinsics::_fmaD:
632 case vmIntrinsics::_fmaF:
633 return inline_fma(intrinsic_id());
634
635 case vmIntrinsics::_isDigit:
636 case vmIntrinsics::_isLowerCase:
637 case vmIntrinsics::_isUpperCase:
638 case vmIntrinsics::_isWhitespace:
639 return inline_character_compare(intrinsic_id());
640
641 case vmIntrinsics::_min:
642 case vmIntrinsics::_max:
643 case vmIntrinsics::_min_strict:
644 case vmIntrinsics::_max_strict:
645 return inline_min_max(intrinsic_id());
646
647 case vmIntrinsics::_maxF:
648 case vmIntrinsics::_minF:
649 case vmIntrinsics::_maxD:
650 case vmIntrinsics::_minD:
651 case vmIntrinsics::_maxF_strict:
652 case vmIntrinsics::_minF_strict:
653 case vmIntrinsics::_maxD_strict:
654 case vmIntrinsics::_minD_strict:
655 return inline_fp_min_max(intrinsic_id());
656
657 case vmIntrinsics::_VectorUnaryOp:
658 return inline_vector_nary_operation(1);
659 case vmIntrinsics::_VectorBinaryOp:
660 return inline_vector_nary_operation(2);
661 case vmIntrinsics::_VectorTernaryOp:
662 return inline_vector_nary_operation(3);
663 case vmIntrinsics::_VectorFromBitsCoerced:
664 return inline_vector_frombits_coerced();
665 case vmIntrinsics::_VectorShuffleIota:
666 return inline_vector_shuffle_iota();
667 case vmIntrinsics::_VectorMaskOp:
668 return inline_vector_mask_operation();
669 case vmIntrinsics::_VectorShuffleToVector:
670 return inline_vector_shuffle_to_vector();
671 case vmIntrinsics::_VectorLoadOp:
672 return inline_vector_mem_operation(/*is_store=*/false);
673 case vmIntrinsics::_VectorLoadMaskedOp:
674 return inline_vector_mem_masked_operation(/*is_store*/false);
675 case vmIntrinsics::_VectorStoreOp:
676 return inline_vector_mem_operation(/*is_store=*/true);
677 case vmIntrinsics::_VectorStoreMaskedOp:
678 return inline_vector_mem_masked_operation(/*is_store=*/true);
679 case vmIntrinsics::_VectorGatherOp:
680 return inline_vector_gather_scatter(/*is_scatter*/ false);
681 case vmIntrinsics::_VectorScatterOp:
682 return inline_vector_gather_scatter(/*is_scatter*/ true);
683 case vmIntrinsics::_VectorReductionCoerced:
684 return inline_vector_reduction();
685 case vmIntrinsics::_VectorTest:
686 return inline_vector_test();
687 case vmIntrinsics::_VectorBlend:
688 return inline_vector_blend();
689 case vmIntrinsics::_VectorRearrange:
690 return inline_vector_rearrange();
691 case vmIntrinsics::_VectorCompare:
692 return inline_vector_compare();
693 case vmIntrinsics::_VectorBroadcastInt:
694 return inline_vector_broadcast_int();
695 case vmIntrinsics::_VectorConvert:
696 return inline_vector_convert();
697 case vmIntrinsics::_VectorInsert:
698 return inline_vector_insert();
699 case vmIntrinsics::_VectorExtract:
700 return inline_vector_extract();
701
702 case vmIntrinsics::_getObjectSize:
703 return inline_getObjectSize();
704
705 case vmIntrinsics::_blackhole:
706 return inline_blackhole();
707
708 default:
709 // If you get here, it may be that someone has added a new intrinsic
710 // to the list in vmIntrinsics.hpp without implementing it here.
711 #ifndef PRODUCT
712 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
713 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
714 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
715 }
716 #endif
717 return false;
718 }
719 }
720
721 Node* LibraryCallKit::try_to_predicate(int predicate) {
722 if (!jvms()->has_method()) {
723 // Root JVMState has a null method.
724 assert(map()->memory()->Opcode() == Op_Parm, "");
725 // Insert the memory aliasing node
726 set_all_memory(reset_memory());
727 }
728 assert(merged_memory(), "");
729
730 switch (intrinsic_id()) {
731 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
732 return inline_cipherBlockChaining_AESCrypt_predicate(false);
733 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
734 return inline_cipherBlockChaining_AESCrypt_predicate(true);
735 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
736 return inline_electronicCodeBook_AESCrypt_predicate(false);
737 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
738 return inline_electronicCodeBook_AESCrypt_predicate(true);
739 case vmIntrinsics::_counterMode_AESCrypt:
740 return inline_counterMode_AESCrypt_predicate();
741 case vmIntrinsics::_digestBase_implCompressMB:
742 return inline_digestBase_implCompressMB_predicate(predicate);
743 case vmIntrinsics::_galoisCounterMode_AESCrypt:
744 return inline_galoisCounterMode_AESCrypt_predicate();
745
746 default:
747 // If you get here, it may be that someone has added a new intrinsic
748 // to the list in vmIntrinsics.hpp without implementing it here.
749 #ifndef PRODUCT
750 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
751 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
752 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
753 }
754 #endif
755 Node* slow_ctl = control();
756 set_control(top()); // No fast path instrinsic
757 return slow_ctl;
758 }
759 }
760
761 //------------------------------set_result-------------------------------
762 // Helper function for finishing intrinsics.
763 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
764 record_for_igvn(region);
765 set_control(_gvn.transform(region));
766 set_result( _gvn.transform(value));
767 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
768 }
769
770 //------------------------------generate_guard---------------------------
771 // Helper function for generating guarded fast-slow graph structures.
772 // The given 'test', if true, guards a slow path. If the test fails
773 // then a fast path can be taken. (We generally hope it fails.)
774 // In all cases, GraphKit::control() is updated to the fast path.
775 // The returned value represents the control for the slow path.
776 // The return value is never 'top'; it is either a valid control
777 // or NULL if it is obvious that the slow path can never be taken.
778 // Also, if region and the slow control are not NULL, the slow edge
779 // is appended to the region.
780 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
781 if (stopped()) {
782 // Already short circuited.
783 return NULL;
784 }
785
786 // Build an if node and its projections.
787 // If test is true we take the slow path, which we assume is uncommon.
788 if (_gvn.type(test) == TypeInt::ZERO) {
789 // The slow branch is never taken. No need to build this guard.
790 return NULL;
791 }
792
793 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
794
795 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
796 if (if_slow == top()) {
797 // The slow branch is never taken. No need to build this guard.
798 return NULL;
799 }
800
801 if (region != NULL)
802 region->add_req(if_slow);
803
804 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
805 set_control(if_fast);
806
807 return if_slow;
808 }
809
810 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
811 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
812 }
813 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
814 return generate_guard(test, region, PROB_FAIR);
815 }
816
817 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
818 Node* *pos_index) {
819 if (stopped())
820 return NULL; // already stopped
821 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
822 return NULL; // index is already adequately typed
823 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
824 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
825 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
826 if (is_neg != NULL && pos_index != NULL) {
827 // Emulate effect of Parse::adjust_map_after_if.
828 Node* ccast = new CastIINode(index, TypeInt::POS);
829 ccast->set_req(0, control());
830 (*pos_index) = _gvn.transform(ccast);
831 }
832 return is_neg;
833 }
834
835 // Make sure that 'position' is a valid limit index, in [0..length].
836 // There are two equivalent plans for checking this:
837 // A. (offset + copyLength) unsigned<= arrayLength
838 // B. offset <= (arrayLength - copyLength)
839 // We require that all of the values above, except for the sum and
840 // difference, are already known to be non-negative.
841 // Plan A is robust in the face of overflow, if offset and copyLength
842 // are both hugely positive.
843 //
844 // Plan B is less direct and intuitive, but it does not overflow at
845 // all, since the difference of two non-negatives is always
846 // representable. Whenever Java methods must perform the equivalent
847 // check they generally use Plan B instead of Plan A.
848 // For the moment we use Plan A.
849 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
850 Node* subseq_length,
851 Node* array_length,
852 RegionNode* region) {
853 if (stopped())
854 return NULL; // already stopped
855 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
856 if (zero_offset && subseq_length->eqv_uncast(array_length))
857 return NULL; // common case of whole-array copy
858 Node* last = subseq_length;
859 if (!zero_offset) // last += offset
860 last = _gvn.transform(new AddINode(last, offset));
861 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
862 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
863 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
864 return is_over;
865 }
866
867 // Emit range checks for the given String.value byte array
868 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
869 if (stopped()) {
870 return; // already stopped
871 }
872 RegionNode* bailout = new RegionNode(1);
873 record_for_igvn(bailout);
874 if (char_count) {
875 // Convert char count to byte count
876 count = _gvn.transform(new LShiftINode(count, intcon(1)));
877 }
878
879 // Offset and count must not be negative
880 generate_negative_guard(offset, bailout);
881 generate_negative_guard(count, bailout);
882 // Offset + count must not exceed length of array
883 generate_limit_guard(offset, count, load_array_length(array), bailout);
884
885 if (bailout->req() > 1) {
886 PreserveJVMState pjvms(this);
887 set_control(_gvn.transform(bailout));
888 uncommon_trap(Deoptimization::Reason_intrinsic,
889 Deoptimization::Action_maybe_recompile);
890 }
891 }
892
893 //--------------------------generate_current_thread--------------------
894 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
895 ciKlass* thread_klass = env()->Thread_klass();
896 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
897 Node* thread = _gvn.transform(new ThreadLocalNode());
898 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
899 tls_output = thread;
900 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
901 thread_obj_handle = _gvn.transform(thread_obj_handle);
902 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
903 }
904
905
906 //------------------------------make_string_method_node------------------------
907 // Helper method for String intrinsic functions. This version is called with
908 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
909 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
910 // containing the lengths of str1 and str2.
911 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
912 Node* result = NULL;
913 switch (opcode) {
914 case Op_StrIndexOf:
915 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
916 str1_start, cnt1, str2_start, cnt2, ae);
917 break;
918 case Op_StrComp:
919 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
920 str1_start, cnt1, str2_start, cnt2, ae);
921 break;
922 case Op_StrEquals:
923 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
924 // Use the constant length if there is one because optimized match rule may exist.
925 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
926 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
927 break;
928 default:
929 ShouldNotReachHere();
930 return NULL;
931 }
932
933 // All these intrinsics have checks.
934 C->set_has_split_ifs(true); // Has chance for split-if optimization
935 clear_upper_avx();
936
937 return _gvn.transform(result);
938 }
939
940 //------------------------------inline_string_compareTo------------------------
941 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
942 Node* arg1 = argument(0);
943 Node* arg2 = argument(1);
944
945 arg1 = must_be_not_null(arg1, true);
946 arg2 = must_be_not_null(arg2, true);
947
948 // Get start addr and length of first argument
949 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
950 Node* arg1_cnt = load_array_length(arg1);
951
952 // Get start addr and length of second argument
953 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
954 Node* arg2_cnt = load_array_length(arg2);
955
956 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
957 set_result(result);
958 return true;
959 }
960
961 //------------------------------inline_string_equals------------------------
962 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
963 Node* arg1 = argument(0);
964 Node* arg2 = argument(1);
965
966 // paths (plus control) merge
967 RegionNode* region = new RegionNode(3);
968 Node* phi = new PhiNode(region, TypeInt::BOOL);
969
970 if (!stopped()) {
971
972 arg1 = must_be_not_null(arg1, true);
973 arg2 = must_be_not_null(arg2, true);
974
975 // Get start addr and length of first argument
976 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
977 Node* arg1_cnt = load_array_length(arg1);
978
979 // Get start addr and length of second argument
980 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
981 Node* arg2_cnt = load_array_length(arg2);
982
983 // Check for arg1_cnt != arg2_cnt
984 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
985 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
986 Node* if_ne = generate_slow_guard(bol, NULL);
987 if (if_ne != NULL) {
988 phi->init_req(2, intcon(0));
989 region->init_req(2, if_ne);
990 }
991
992 // Check for count == 0 is done by assembler code for StrEquals.
993
994 if (!stopped()) {
995 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
996 phi->init_req(1, equals);
997 region->init_req(1, control());
998 }
999 }
1000
1001 // post merge
1002 set_control(_gvn.transform(region));
1003 record_for_igvn(region);
1004
1005 set_result(_gvn.transform(phi));
1006 return true;
1007 }
1008
1009 //------------------------------inline_array_equals----------------------------
1010 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1011 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1012 Node* arg1 = argument(0);
1013 Node* arg2 = argument(1);
1014
1015 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1016 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1017 clear_upper_avx();
1018
1019 return true;
1020 }
1021
1022 //------------------------------inline_countPositives------------------------------
1023 bool LibraryCallKit::inline_countPositives() {
1024 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1025 return false;
1026 }
1027
1028 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1029 // no receiver since it is static method
1030 Node* ba = argument(0);
1031 Node* offset = argument(1);
1032 Node* len = argument(2);
1033
1034 ba = must_be_not_null(ba, true);
1035
1036 // Range checks
1037 generate_string_range_check(ba, offset, len, false);
1038 if (stopped()) {
1039 return true;
1040 }
1041 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1042 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1043 set_result(_gvn.transform(result));
1044 return true;
1045 }
1046
1047 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1048 Node* index = argument(0);
1049 Node* length = bt == T_INT ? argument(1) : argument(2);
1050 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1051 return false;
1052 }
1053
1054 // check that length is positive
1055 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1056 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1057
1058 {
1059 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1060 uncommon_trap(Deoptimization::Reason_intrinsic,
1061 Deoptimization::Action_make_not_entrant);
1062 }
1063
1064 if (stopped()) {
1065 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1066 return true;
1067 }
1068
1069 // length is now known postive, add a cast node to make this explicit
1070 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1071 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1072 casted_length = _gvn.transform(casted_length);
1073 replace_in_map(length, casted_length);
1074 length = casted_length;
1075
1076 // Use an unsigned comparison for the range check itself
1077 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1078 BoolTest::mask btest = BoolTest::lt;
1079 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1080 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1081 _gvn.set_type(rc, rc->Value(&_gvn));
1082 if (!rc_bool->is_Con()) {
1083 record_for_igvn(rc);
1084 }
1085 set_control(_gvn.transform(new IfTrueNode(rc)));
1086 {
1087 PreserveJVMState pjvms(this);
1088 set_control(_gvn.transform(new IfFalseNode(rc)));
1089 uncommon_trap(Deoptimization::Reason_range_check,
1090 Deoptimization::Action_make_not_entrant);
1091 }
1092
1093 if (stopped()) {
1094 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1095 return true;
1096 }
1097
1098 // index is now known to be >= 0 and < length, cast it
1099 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1100 result = _gvn.transform(result);
1101 set_result(result);
1102 replace_in_map(index, result);
1103 return true;
1104 }
1105
1106 //------------------------------inline_string_indexOf------------------------
1107 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1108 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1109 return false;
1110 }
1111 Node* src = argument(0);
1112 Node* tgt = argument(1);
1113
1114 // Make the merge point
1115 RegionNode* result_rgn = new RegionNode(4);
1116 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1117
1118 src = must_be_not_null(src, true);
1119 tgt = must_be_not_null(tgt, true);
1120
1121 // Get start addr and length of source string
1122 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1123 Node* src_count = load_array_length(src);
1124
1125 // Get start addr and length of substring
1126 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1127 Node* tgt_count = load_array_length(tgt);
1128
1129 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1130 // Divide src size by 2 if String is UTF16 encoded
1131 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1132 }
1133 if (ae == StrIntrinsicNode::UU) {
1134 // Divide substring size by 2 if String is UTF16 encoded
1135 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1136 }
1137
1138 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1139 if (result != NULL) {
1140 result_phi->init_req(3, result);
1141 result_rgn->init_req(3, control());
1142 }
1143 set_control(_gvn.transform(result_rgn));
1144 record_for_igvn(result_rgn);
1145 set_result(_gvn.transform(result_phi));
1146
1147 return true;
1148 }
1149
1150 //-----------------------------inline_string_indexOf-----------------------
1151 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1152 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1153 return false;
1154 }
1155 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1156 return false;
1157 }
1158 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1159 Node* src = argument(0); // byte[]
1160 Node* src_count = argument(1); // char count
1161 Node* tgt = argument(2); // byte[]
1162 Node* tgt_count = argument(3); // char count
1163 Node* from_index = argument(4); // char index
1164
1165 src = must_be_not_null(src, true);
1166 tgt = must_be_not_null(tgt, true);
1167
1168 // Multiply byte array index by 2 if String is UTF16 encoded
1169 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1170 src_count = _gvn.transform(new SubINode(src_count, from_index));
1171 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1172 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1173
1174 // Range checks
1175 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1176 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1177 if (stopped()) {
1178 return true;
1179 }
1180
1181 RegionNode* region = new RegionNode(5);
1182 Node* phi = new PhiNode(region, TypeInt::INT);
1183
1184 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1185 if (result != NULL) {
1186 // The result is index relative to from_index if substring was found, -1 otherwise.
1187 // Generate code which will fold into cmove.
1188 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1189 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1190
1191 Node* if_lt = generate_slow_guard(bol, NULL);
1192 if (if_lt != NULL) {
1193 // result == -1
1194 phi->init_req(3, result);
1195 region->init_req(3, if_lt);
1196 }
1197 if (!stopped()) {
1198 result = _gvn.transform(new AddINode(result, from_index));
1199 phi->init_req(4, result);
1200 region->init_req(4, control());
1201 }
1202 }
1203
1204 set_control(_gvn.transform(region));
1205 record_for_igvn(region);
1206 set_result(_gvn.transform(phi));
1207 clear_upper_avx();
1208
1209 return true;
1210 }
1211
1212 // Create StrIndexOfNode with fast path checks
1213 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1214 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1215 // Check for substr count > string count
1216 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1217 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1218 Node* if_gt = generate_slow_guard(bol, NULL);
1219 if (if_gt != NULL) {
1220 phi->init_req(1, intcon(-1));
1221 region->init_req(1, if_gt);
1222 }
1223 if (!stopped()) {
1224 // Check for substr count == 0
1225 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1226 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1227 Node* if_zero = generate_slow_guard(bol, NULL);
1228 if (if_zero != NULL) {
1229 phi->init_req(2, intcon(0));
1230 region->init_req(2, if_zero);
1231 }
1232 }
1233 if (!stopped()) {
1234 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1235 }
1236 return NULL;
1237 }
1238
1239 //-----------------------------inline_string_indexOfChar-----------------------
1240 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1241 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1242 return false;
1243 }
1244 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1245 return false;
1246 }
1247 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1248 Node* src = argument(0); // byte[]
1249 Node* tgt = argument(1); // tgt is int ch
1250 Node* from_index = argument(2);
1251 Node* max = argument(3);
1252
1253 src = must_be_not_null(src, true);
1254
1255 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1256 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1257 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1258
1259 // Range checks
1260 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1261 if (stopped()) {
1262 return true;
1263 }
1264
1265 RegionNode* region = new RegionNode(3);
1266 Node* phi = new PhiNode(region, TypeInt::INT);
1267
1268 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1269 C->set_has_split_ifs(true); // Has chance for split-if optimization
1270 _gvn.transform(result);
1271
1272 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1273 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1274
1275 Node* if_lt = generate_slow_guard(bol, NULL);
1276 if (if_lt != NULL) {
1277 // result == -1
1278 phi->init_req(2, result);
1279 region->init_req(2, if_lt);
1280 }
1281 if (!stopped()) {
1282 result = _gvn.transform(new AddINode(result, from_index));
1283 phi->init_req(1, result);
1284 region->init_req(1, control());
1285 }
1286 set_control(_gvn.transform(region));
1287 record_for_igvn(region);
1288 set_result(_gvn.transform(phi));
1289
1290 return true;
1291 }
1292 //---------------------------inline_string_copy---------------------
1293 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1294 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1295 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1296 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1297 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1298 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1299 bool LibraryCallKit::inline_string_copy(bool compress) {
1300 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1301 return false;
1302 }
1303 int nargs = 5; // 2 oops, 3 ints
1304 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1305
1306 Node* src = argument(0);
1307 Node* src_offset = argument(1);
1308 Node* dst = argument(2);
1309 Node* dst_offset = argument(3);
1310 Node* length = argument(4);
1311
1312 // Check for allocation before we add nodes that would confuse
1313 // tightly_coupled_allocation()
1314 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1315
1316 // Figure out the size and type of the elements we will be copying.
1317 const Type* src_type = src->Value(&_gvn);
1318 const Type* dst_type = dst->Value(&_gvn);
1319 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1320 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1321 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1322 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1323 "Unsupported array types for inline_string_copy");
1324
1325 src = must_be_not_null(src, true);
1326 dst = must_be_not_null(dst, true);
1327
1328 // Convert char[] offsets to byte[] offsets
1329 bool convert_src = (compress && src_elem == T_BYTE);
1330 bool convert_dst = (!compress && dst_elem == T_BYTE);
1331 if (convert_src) {
1332 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1333 } else if (convert_dst) {
1334 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1335 }
1336
1337 // Range checks
1338 generate_string_range_check(src, src_offset, length, convert_src);
1339 generate_string_range_check(dst, dst_offset, length, convert_dst);
1340 if (stopped()) {
1341 return true;
1342 }
1343
1344 Node* src_start = array_element_address(src, src_offset, src_elem);
1345 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1346 // 'src_start' points to src array + scaled offset
1347 // 'dst_start' points to dst array + scaled offset
1348 Node* count = NULL;
1349 if (compress) {
1350 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1351 } else {
1352 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1353 }
1354
1355 if (alloc != NULL) {
1356 if (alloc->maybe_set_complete(&_gvn)) {
1357 // "You break it, you buy it."
1358 InitializeNode* init = alloc->initialization();
1359 assert(init->is_complete(), "we just did this");
1360 init->set_complete_with_arraycopy();
1361 assert(dst->is_CheckCastPP(), "sanity");
1362 assert(dst->in(0)->in(0) == init, "dest pinned");
1363 }
1364 // Do not let stores that initialize this object be reordered with
1365 // a subsequent store that would make this object accessible by
1366 // other threads.
1367 // Record what AllocateNode this StoreStore protects so that
1368 // escape analysis can go from the MemBarStoreStoreNode to the
1369 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1370 // based on the escape status of the AllocateNode.
1371 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1372 }
1373 if (compress) {
1374 set_result(_gvn.transform(count));
1375 }
1376 clear_upper_avx();
1377
1378 return true;
1379 }
1380
1381 #ifdef _LP64
1382 #define XTOP ,top() /*additional argument*/
1383 #else //_LP64
1384 #define XTOP /*no additional argument*/
1385 #endif //_LP64
1386
1387 //------------------------inline_string_toBytesU--------------------------
1388 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1389 bool LibraryCallKit::inline_string_toBytesU() {
1390 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1391 return false;
1392 }
1393 // Get the arguments.
1394 Node* value = argument(0);
1395 Node* offset = argument(1);
1396 Node* length = argument(2);
1397
1398 Node* newcopy = NULL;
1399
1400 // Set the original stack and the reexecute bit for the interpreter to reexecute
1401 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1402 { PreserveReexecuteState preexecs(this);
1403 jvms()->set_should_reexecute(true);
1404
1405 // Check if a null path was taken unconditionally.
1406 value = null_check(value);
1407
1408 RegionNode* bailout = new RegionNode(1);
1409 record_for_igvn(bailout);
1410
1411 // Range checks
1412 generate_negative_guard(offset, bailout);
1413 generate_negative_guard(length, bailout);
1414 generate_limit_guard(offset, length, load_array_length(value), bailout);
1415 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1416 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1417
1418 if (bailout->req() > 1) {
1419 PreserveJVMState pjvms(this);
1420 set_control(_gvn.transform(bailout));
1421 uncommon_trap(Deoptimization::Reason_intrinsic,
1422 Deoptimization::Action_maybe_recompile);
1423 }
1424 if (stopped()) {
1425 return true;
1426 }
1427
1428 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1429 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1430 newcopy = new_array(klass_node, size, 0); // no arguments to push
1431 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1432 guarantee(alloc != NULL, "created above");
1433
1434 // Calculate starting addresses.
1435 Node* src_start = array_element_address(value, offset, T_CHAR);
1436 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1437
1438 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1439 const TypeInt* toffset = gvn().type(offset)->is_int();
1440 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1441
1442 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1443 const char* copyfunc_name = "arraycopy";
1444 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1445 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1446 OptoRuntime::fast_arraycopy_Type(),
1447 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1448 src_start, dst_start, ConvI2X(length) XTOP);
1449 // Do not let reads from the cloned object float above the arraycopy.
1450 if (alloc->maybe_set_complete(&_gvn)) {
1451 // "You break it, you buy it."
1452 InitializeNode* init = alloc->initialization();
1453 assert(init->is_complete(), "we just did this");
1454 init->set_complete_with_arraycopy();
1455 assert(newcopy->is_CheckCastPP(), "sanity");
1456 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1457 }
1458 // Do not let stores that initialize this object be reordered with
1459 // a subsequent store that would make this object accessible by
1460 // other threads.
1461 // Record what AllocateNode this StoreStore protects so that
1462 // escape analysis can go from the MemBarStoreStoreNode to the
1463 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1464 // based on the escape status of the AllocateNode.
1465 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1466 } // original reexecute is set back here
1467
1468 C->set_has_split_ifs(true); // Has chance for split-if optimization
1469 if (!stopped()) {
1470 set_result(newcopy);
1471 }
1472 clear_upper_avx();
1473
1474 return true;
1475 }
1476
1477 //------------------------inline_string_getCharsU--------------------------
1478 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1479 bool LibraryCallKit::inline_string_getCharsU() {
1480 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1481 return false;
1482 }
1483
1484 // Get the arguments.
1485 Node* src = argument(0);
1486 Node* src_begin = argument(1);
1487 Node* src_end = argument(2); // exclusive offset (i < src_end)
1488 Node* dst = argument(3);
1489 Node* dst_begin = argument(4);
1490
1491 // Check for allocation before we add nodes that would confuse
1492 // tightly_coupled_allocation()
1493 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1494
1495 // Check if a null path was taken unconditionally.
1496 src = null_check(src);
1497 dst = null_check(dst);
1498 if (stopped()) {
1499 return true;
1500 }
1501
1502 // Get length and convert char[] offset to byte[] offset
1503 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1504 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1505
1506 // Range checks
1507 generate_string_range_check(src, src_begin, length, true);
1508 generate_string_range_check(dst, dst_begin, length, false);
1509 if (stopped()) {
1510 return true;
1511 }
1512
1513 if (!stopped()) {
1514 // Calculate starting addresses.
1515 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1516 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1517
1518 // Check if array addresses are aligned to HeapWordSize
1519 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1520 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1521 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1522 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1523
1524 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1525 const char* copyfunc_name = "arraycopy";
1526 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1527 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1528 OptoRuntime::fast_arraycopy_Type(),
1529 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1530 src_start, dst_start, ConvI2X(length) XTOP);
1531 // Do not let reads from the cloned object float above the arraycopy.
1532 if (alloc != NULL) {
1533 if (alloc->maybe_set_complete(&_gvn)) {
1534 // "You break it, you buy it."
1535 InitializeNode* init = alloc->initialization();
1536 assert(init->is_complete(), "we just did this");
1537 init->set_complete_with_arraycopy();
1538 assert(dst->is_CheckCastPP(), "sanity");
1539 assert(dst->in(0)->in(0) == init, "dest pinned");
1540 }
1541 // Do not let stores that initialize this object be reordered with
1542 // a subsequent store that would make this object accessible by
1543 // other threads.
1544 // Record what AllocateNode this StoreStore protects so that
1545 // escape analysis can go from the MemBarStoreStoreNode to the
1546 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1547 // based on the escape status of the AllocateNode.
1548 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1549 } else {
1550 insert_mem_bar(Op_MemBarCPUOrder);
1551 }
1552 }
1553
1554 C->set_has_split_ifs(true); // Has chance for split-if optimization
1555 return true;
1556 }
1557
1558 //----------------------inline_string_char_access----------------------------
1559 // Store/Load char to/from byte[] array.
1560 // static void StringUTF16.putChar(byte[] val, int index, int c)
1561 // static char StringUTF16.getChar(byte[] val, int index)
1562 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1563 Node* value = argument(0);
1564 Node* index = argument(1);
1565 Node* ch = is_store ? argument(2) : NULL;
1566
1567 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1568 // correctly requires matched array shapes.
1569 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1570 "sanity: byte[] and char[] bases agree");
1571 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1572 "sanity: byte[] and char[] scales agree");
1573
1574 // Bail when getChar over constants is requested: constant folding would
1575 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1576 // Java method would constant fold nicely instead.
1577 if (!is_store && value->is_Con() && index->is_Con()) {
1578 return false;
1579 }
1580
1581 value = must_be_not_null(value, true);
1582
1583 Node* adr = array_element_address(value, index, T_CHAR);
1584 if (adr->is_top()) {
1585 return false;
1586 }
1587 if (is_store) {
1588 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1589 } else {
1590 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1591 set_result(ch);
1592 }
1593 return true;
1594 }
1595
1596 //--------------------------round_double_node--------------------------------
1597 // Round a double node if necessary.
1598 Node* LibraryCallKit::round_double_node(Node* n) {
1599 if (Matcher::strict_fp_requires_explicit_rounding) {
1600 #ifdef IA32
1601 if (UseSSE < 2) {
1602 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1603 }
1604 #else
1605 Unimplemented();
1606 #endif // IA32
1607 }
1608 return n;
1609 }
1610
1611 //------------------------------inline_math-----------------------------------
1612 // public static double Math.abs(double)
1613 // public static double Math.sqrt(double)
1614 // public static double Math.log(double)
1615 // public static double Math.log10(double)
1616 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1617 Node* arg = round_double_node(argument(0));
1618 Node* n = NULL;
1619 switch (id) {
1620 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1621 case vmIntrinsics::_dsqrt:
1622 case vmIntrinsics::_dsqrt_strict:
1623 n = new SqrtDNode(C, control(), arg); break;
1624 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1625 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1626 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1627 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1628 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1629 default: fatal_unexpected_iid(id); break;
1630 }
1631 set_result(_gvn.transform(n));
1632 return true;
1633 }
1634
1635 //------------------------------inline_math-----------------------------------
1636 // public static float Math.abs(float)
1637 // public static int Math.abs(int)
1638 // public static long Math.abs(long)
1639 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1640 Node* arg = argument(0);
1641 Node* n = NULL;
1642 switch (id) {
1643 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1644 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1645 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1646 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1647 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1648 default: fatal_unexpected_iid(id); break;
1649 }
1650 set_result(_gvn.transform(n));
1651 return true;
1652 }
1653
1654 //------------------------------runtime_math-----------------------------
1655 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1656 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1657 "must be (DD)D or (D)D type");
1658
1659 // Inputs
1660 Node* a = round_double_node(argument(0));
1661 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1662
1663 const TypePtr* no_memory_effects = NULL;
1664 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1665 no_memory_effects,
1666 a, top(), b, b ? top() : NULL);
1667 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1668 #ifdef ASSERT
1669 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1670 assert(value_top == top(), "second value must be top");
1671 #endif
1672
1673 set_result(value);
1674 return true;
1675 }
1676
1677 //------------------------------inline_math_pow-----------------------------
1678 bool LibraryCallKit::inline_math_pow() {
1679 Node* exp = round_double_node(argument(2));
1680 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1681 if (d != NULL) {
1682 if (d->getd() == 2.0) {
1683 // Special case: pow(x, 2.0) => x * x
1684 Node* base = round_double_node(argument(0));
1685 set_result(_gvn.transform(new MulDNode(base, base)));
1686 return true;
1687 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1688 // Special case: pow(x, 0.5) => sqrt(x)
1689 Node* base = round_double_node(argument(0));
1690 Node* zero = _gvn.zerocon(T_DOUBLE);
1691
1692 RegionNode* region = new RegionNode(3);
1693 Node* phi = new PhiNode(region, Type::DOUBLE);
1694
1695 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1696 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1697 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1698 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1699 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1700
1701 Node* if_pow = generate_slow_guard(test, NULL);
1702 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1703 phi->init_req(1, value_sqrt);
1704 region->init_req(1, control());
1705
1706 if (if_pow != NULL) {
1707 set_control(if_pow);
1708 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1709 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1710 const TypePtr* no_memory_effects = NULL;
1711 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1712 no_memory_effects, base, top(), exp, top());
1713 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1714 #ifdef ASSERT
1715 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1716 assert(value_top == top(), "second value must be top");
1717 #endif
1718 phi->init_req(2, value_pow);
1719 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1720 }
1721
1722 C->set_has_split_ifs(true); // Has chance for split-if optimization
1723 set_control(_gvn.transform(region));
1724 record_for_igvn(region);
1725 set_result(_gvn.transform(phi));
1726
1727 return true;
1728 }
1729 }
1730
1731 return StubRoutines::dpow() != NULL ?
1732 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1733 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1734 }
1735
1736 //------------------------------inline_math_native-----------------------------
1737 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1738 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1739 switch (id) {
1740 // These intrinsics are not properly supported on all hardware
1741 case vmIntrinsics::_dsin:
1742 return StubRoutines::dsin() != NULL ?
1743 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1744 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1745 case vmIntrinsics::_dcos:
1746 return StubRoutines::dcos() != NULL ?
1747 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1748 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1749 case vmIntrinsics::_dtan:
1750 return StubRoutines::dtan() != NULL ?
1751 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1752 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1753 case vmIntrinsics::_dlog:
1754 return StubRoutines::dlog() != NULL ?
1755 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1756 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1757 case vmIntrinsics::_dlog10:
1758 return StubRoutines::dlog10() != NULL ?
1759 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1760 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1761
1762 // These intrinsics are supported on all hardware
1763 case vmIntrinsics::_ceil:
1764 case vmIntrinsics::_floor:
1765 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1766 case vmIntrinsics::_dsqrt:
1767 case vmIntrinsics::_dsqrt_strict:
1768 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1769 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1770 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1771 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1772 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1773
1774 case vmIntrinsics::_dexp:
1775 return StubRoutines::dexp() != NULL ?
1776 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1777 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1778 #undef FN_PTR
1779
1780 case vmIntrinsics::_dpow: return inline_math_pow();
1781 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1782 case vmIntrinsics::_fcopySign: return inline_math(id);
1783 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1784 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1785
1786 // These intrinsics are not yet correctly implemented
1787 case vmIntrinsics::_datan2:
1788 return false;
1789
1790 default:
1791 fatal_unexpected_iid(id);
1792 return false;
1793 }
1794 }
1795
1796 static bool is_simple_name(Node* n) {
1797 return (n->req() == 1 // constant
1798 || (n->is_Type() && n->as_Type()->type()->singleton())
1799 || n->is_Proj() // parameter or return value
1800 || n->is_Phi() // local of some sort
1801 );
1802 }
1803
1804 //----------------------------inline_notify-----------------------------------*
1805 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1806 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1807 address func;
1808 if (id == vmIntrinsics::_notify) {
1809 func = OptoRuntime::monitor_notify_Java();
1810 } else {
1811 func = OptoRuntime::monitor_notifyAll_Java();
1812 }
1813 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1814 make_slow_call_ex(call, env()->Throwable_klass(), false);
1815 return true;
1816 }
1817
1818
1819 //----------------------------inline_min_max-----------------------------------
1820 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1821 set_result(generate_min_max(id, argument(0), argument(1)));
1822 return true;
1823 }
1824
1825 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1826 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1827 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1828 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1829 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1830
1831 {
1832 PreserveJVMState pjvms(this);
1833 PreserveReexecuteState preexecs(this);
1834 jvms()->set_should_reexecute(true);
1835
1836 set_control(slow_path);
1837 set_i_o(i_o());
1838
1839 uncommon_trap(Deoptimization::Reason_intrinsic,
1840 Deoptimization::Action_none);
1841 }
1842
1843 set_control(fast_path);
1844 set_result(math);
1845 }
1846
1847 template <typename OverflowOp>
1848 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1849 typedef typename OverflowOp::MathOp MathOp;
1850
1851 MathOp* mathOp = new MathOp(arg1, arg2);
1852 Node* operation = _gvn.transform( mathOp );
1853 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1854 inline_math_mathExact(operation, ofcheck);
1855 return true;
1856 }
1857
1858 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1859 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1860 }
1861
1862 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1863 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1864 }
1865
1866 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1867 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1868 }
1869
1870 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1871 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1872 }
1873
1874 bool LibraryCallKit::inline_math_negateExactI() {
1875 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1876 }
1877
1878 bool LibraryCallKit::inline_math_negateExactL() {
1879 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1880 }
1881
1882 bool LibraryCallKit::inline_math_multiplyExactI() {
1883 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1884 }
1885
1886 bool LibraryCallKit::inline_math_multiplyExactL() {
1887 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1888 }
1889
1890 bool LibraryCallKit::inline_math_multiplyHigh() {
1891 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1892 return true;
1893 }
1894
1895 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1896 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1897 return true;
1898 }
1899
1900 Node*
1901 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1902 // These are the candidate return value:
1903 Node* xvalue = x0;
1904 Node* yvalue = y0;
1905
1906 if (xvalue == yvalue) {
1907 return xvalue;
1908 }
1909
1910 bool want_max = (id == vmIntrinsics::_max || id == vmIntrinsics::_max_strict);
1911
1912 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1913 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1914 if (txvalue == NULL || tyvalue == NULL) return top();
1915 // This is not really necessary, but it is consistent with a
1916 // hypothetical MaxINode::Value method:
1917 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1918
1919 // %%% This folding logic should (ideally) be in a different place.
1920 // Some should be inside IfNode, and there to be a more reliable
1921 // transformation of ?: style patterns into cmoves. We also want
1922 // more powerful optimizations around cmove and min/max.
1923
1924 // Try to find a dominating comparison of these guys.
1925 // It can simplify the index computation for Arrays.copyOf
1926 // and similar uses of System.arraycopy.
1927 // First, compute the normalized version of CmpI(x, y).
1928 int cmp_op = Op_CmpI;
1929 Node* xkey = xvalue;
1930 Node* ykey = yvalue;
1931 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1932 if (ideal_cmpxy->is_Cmp()) {
1933 // E.g., if we have CmpI(length - offset, count),
1934 // it might idealize to CmpI(length, count + offset)
1935 cmp_op = ideal_cmpxy->Opcode();
1936 xkey = ideal_cmpxy->in(1);
1937 ykey = ideal_cmpxy->in(2);
1938 }
1939
1940 // Start by locating any relevant comparisons.
1941 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1942 Node* cmpxy = NULL;
1943 Node* cmpyx = NULL;
1944 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1945 Node* cmp = start_from->fast_out(k);
1946 if (cmp->outcnt() > 0 && // must have prior uses
1947 cmp->in(0) == NULL && // must be context-independent
1948 cmp->Opcode() == cmp_op) { // right kind of compare
1949 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1950 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1951 }
1952 }
1953
1954 const int NCMPS = 2;
1955 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1956 int cmpn;
1957 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1958 if (cmps[cmpn] != NULL) break; // find a result
1959 }
1960 if (cmpn < NCMPS) {
1961 // Look for a dominating test that tells us the min and max.
1962 int depth = 0; // Limit search depth for speed
1963 Node* dom = control();
1964 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1965 if (++depth >= 100) break;
1966 Node* ifproj = dom;
1967 if (!ifproj->is_Proj()) continue;
1968 Node* iff = ifproj->in(0);
1969 if (!iff->is_If()) continue;
1970 Node* bol = iff->in(1);
1971 if (!bol->is_Bool()) continue;
1972 Node* cmp = bol->in(1);
1973 if (cmp == NULL) continue;
1974 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1975 if (cmps[cmpn] == cmp) break;
1976 if (cmpn == NCMPS) continue;
1977 BoolTest::mask btest = bol->as_Bool()->_test._test;
1978 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1979 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1980 // At this point, we know that 'x btest y' is true.
1981 switch (btest) {
1982 case BoolTest::eq:
1983 // They are proven equal, so we can collapse the min/max.
1984 // Either value is the answer. Choose the simpler.
1985 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1986 return yvalue;
1987 return xvalue;
1988 case BoolTest::lt: // x < y
1989 case BoolTest::le: // x <= y
1990 return (want_max ? yvalue : xvalue);
1991 case BoolTest::gt: // x > y
1992 case BoolTest::ge: // x >= y
1993 return (want_max ? xvalue : yvalue);
1994 default:
1995 break;
1996 }
1997 }
1998 }
1999
2000 // We failed to find a dominating test.
2001 // Let's pick a test that might GVN with prior tests.
2002 Node* best_bol = NULL;
2003 BoolTest::mask best_btest = BoolTest::illegal;
2004 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2005 Node* cmp = cmps[cmpn];
2006 if (cmp == NULL) continue;
2007 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2008 Node* bol = cmp->fast_out(j);
2009 if (!bol->is_Bool()) continue;
2010 BoolTest::mask btest = bol->as_Bool()->_test._test;
2011 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2012 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2013 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2014 best_bol = bol->as_Bool();
2015 best_btest = btest;
2016 }
2017 }
2018 }
2019
2020 Node* answer_if_true = NULL;
2021 Node* answer_if_false = NULL;
2022 switch (best_btest) {
2023 default:
2024 if (cmpxy == NULL)
2025 cmpxy = ideal_cmpxy;
2026 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2027 // and fall through:
2028 case BoolTest::lt: // x < y
2029 case BoolTest::le: // x <= y
2030 answer_if_true = (want_max ? yvalue : xvalue);
2031 answer_if_false = (want_max ? xvalue : yvalue);
2032 break;
2033 case BoolTest::gt: // x > y
2034 case BoolTest::ge: // x >= y
2035 answer_if_true = (want_max ? xvalue : yvalue);
2036 answer_if_false = (want_max ? yvalue : xvalue);
2037 break;
2038 }
2039
2040 jint hi, lo;
2041 if (want_max) {
2042 // We can sharpen the minimum.
2043 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2044 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2045 } else {
2046 // We can sharpen the maximum.
2047 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2048 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2049 }
2050
2051 // Use a flow-free graph structure, to avoid creating excess control edges
2052 // which could hinder other optimizations.
2053 // Since Math.min/max is often used with arraycopy, we want
2054 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2055 Node* cmov = CMoveNode::make(NULL, best_bol,
2056 answer_if_false, answer_if_true,
2057 TypeInt::make(lo, hi, widen));
2058
2059 return _gvn.transform(cmov);
2060
2061 /*
2062 // This is not as desirable as it may seem, since Min and Max
2063 // nodes do not have a full set of optimizations.
2064 // And they would interfere, anyway, with 'if' optimizations
2065 // and with CMoveI canonical forms.
2066 switch (id) {
2067 case vmIntrinsics::_min:
2068 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2069 case vmIntrinsics::_max:
2070 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2071 default:
2072 ShouldNotReachHere();
2073 }
2074 */
2075 }
2076
2077 inline int
2078 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2079 const TypePtr* base_type = TypePtr::NULL_PTR;
2080 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2081 if (base_type == NULL) {
2082 // Unknown type.
2083 return Type::AnyPtr;
2084 } else if (base_type == TypePtr::NULL_PTR) {
2085 // Since this is a NULL+long form, we have to switch to a rawptr.
2086 base = _gvn.transform(new CastX2PNode(offset));
2087 offset = MakeConX(0);
2088 return Type::RawPtr;
2089 } else if (base_type->base() == Type::RawPtr) {
2090 return Type::RawPtr;
2091 } else if (base_type->isa_oopptr()) {
2092 // Base is never null => always a heap address.
2093 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2094 return Type::OopPtr;
2095 }
2096 // Offset is small => always a heap address.
2097 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2098 if (offset_type != NULL &&
2099 base_type->offset() == 0 && // (should always be?)
2100 offset_type->_lo >= 0 &&
2101 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2102 return Type::OopPtr;
2103 } else if (type == T_OBJECT) {
2104 // off heap access to an oop doesn't make any sense. Has to be on
2105 // heap.
2106 return Type::OopPtr;
2107 }
2108 // Otherwise, it might either be oop+off or NULL+addr.
2109 return Type::AnyPtr;
2110 } else {
2111 // No information:
2112 return Type::AnyPtr;
2113 }
2114 }
2115
2116 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2117 Node* uncasted_base = base;
2118 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2119 if (kind == Type::RawPtr) {
2120 return basic_plus_adr(top(), uncasted_base, offset);
2121 } else if (kind == Type::AnyPtr) {
2122 assert(base == uncasted_base, "unexpected base change");
2123 if (can_cast) {
2124 if (!_gvn.type(base)->speculative_maybe_null() &&
2125 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2126 // According to profiling, this access is always on
2127 // heap. Casting the base to not null and thus avoiding membars
2128 // around the access should allow better optimizations
2129 Node* null_ctl = top();
2130 base = null_check_oop(base, &null_ctl, true, true, true);
2131 assert(null_ctl->is_top(), "no null control here");
2132 return basic_plus_adr(base, offset);
2133 } else if (_gvn.type(base)->speculative_always_null() &&
2134 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2135 // According to profiling, this access is always off
2136 // heap.
2137 base = null_assert(base);
2138 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2139 offset = MakeConX(0);
2140 return basic_plus_adr(top(), raw_base, offset);
2141 }
2142 }
2143 // We don't know if it's an on heap or off heap access. Fall back
2144 // to raw memory access.
2145 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2146 return basic_plus_adr(top(), raw, offset);
2147 } else {
2148 assert(base == uncasted_base, "unexpected base change");
2149 // We know it's an on heap access so base can't be null
2150 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2151 base = must_be_not_null(base, true);
2152 }
2153 return basic_plus_adr(base, offset);
2154 }
2155 }
2156
2157 //--------------------------inline_number_methods-----------------------------
2158 // inline int Integer.numberOfLeadingZeros(int)
2159 // inline int Long.numberOfLeadingZeros(long)
2160 //
2161 // inline int Integer.numberOfTrailingZeros(int)
2162 // inline int Long.numberOfTrailingZeros(long)
2163 //
2164 // inline int Integer.bitCount(int)
2165 // inline int Long.bitCount(long)
2166 //
2167 // inline char Character.reverseBytes(char)
2168 // inline short Short.reverseBytes(short)
2169 // inline int Integer.reverseBytes(int)
2170 // inline long Long.reverseBytes(long)
2171 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2172 Node* arg = argument(0);
2173 Node* n = NULL;
2174 switch (id) {
2175 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2176 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2177 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2178 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2179 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2180 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2181 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2182 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2183 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2184 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2185 default: fatal_unexpected_iid(id); break;
2186 }
2187 set_result(_gvn.transform(n));
2188 return true;
2189 }
2190
2191 //----------------------------inline_unsafe_access----------------------------
2192
2193 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2194 // Attempt to infer a sharper value type from the offset and base type.
2195 ciKlass* sharpened_klass = NULL;
2196 bool null_free = false;
2197
2198 // See if it is an instance field, with an object type.
2199 if (alias_type->field() != NULL) {
2200 if (alias_type->field()->type()->is_klass()) {
2201 sharpened_klass = alias_type->field()->type()->as_klass();
2202 null_free = alias_type->field()->is_null_free();
2203 }
2204 }
2205
2206 // See if it is a narrow oop array.
2207 if (adr_type->isa_aryptr()) {
2208 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2209 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2210 null_free = adr_type->is_aryptr()->is_null_free();
2211 if (elem_type != NULL) {
2212 sharpened_klass = elem_type->klass();
2213 }
2214 }
2215 }
2216
2217 // The sharpened class might be unloaded if there is no class loader
2218 // contraint in place.
2219 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2220 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2221 if (null_free) {
2222 tjp = tjp->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2223 }
2224
2225 #ifndef PRODUCT
2226 if (C->print_intrinsics() || C->print_inlining()) {
2227 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2228 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2229 }
2230 #endif
2231 // Sharpen the value type.
2232 return tjp;
2233 }
2234 return NULL;
2235 }
2236
2237 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2238 switch (kind) {
2239 case Relaxed:
2240 return MO_UNORDERED;
2241 case Opaque:
2242 return MO_RELAXED;
2243 case Acquire:
2244 return MO_ACQUIRE;
2245 case Release:
2246 return MO_RELEASE;
2247 case Volatile:
2248 return MO_SEQ_CST;
2249 default:
2250 ShouldNotReachHere();
2251 return 0;
2252 }
2253 }
2254
2255 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2256 if (callee()->is_static()) return false; // caller must have the capability!
2257 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2258 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2259 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2260 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2261
2262 if (is_reference_type(type)) {
2263 decorators |= ON_UNKNOWN_OOP_REF;
2264 }
2265
2266 if (unaligned) {
2267 decorators |= C2_UNALIGNED;
2268 }
2269
2270 #ifndef PRODUCT
2271 {
2272 ResourceMark rm;
2273 // Check the signatures.
2274 ciSignature* sig = callee()->signature();
2275 #ifdef ASSERT
2276 if (!is_store) {
2277 // Object getReference(Object base, int/long offset), etc.
2278 BasicType rtype = sig->return_type()->basic_type();
2279 assert(rtype == type || (rtype == T_OBJECT && type == T_PRIMITIVE_OBJECT), "getter must return the expected value");
2280 assert(sig->count() == 2 || (type == T_PRIMITIVE_OBJECT && sig->count() == 3), "oop getter has 2 or 3 arguments");
2281 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2282 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2283 } else {
2284 // void putReference(Object base, int/long offset, Object x), etc.
2285 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2286 assert(sig->count() == 3 || (type == T_PRIMITIVE_OBJECT && sig->count() == 4), "oop putter has 3 arguments");
2287 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2288 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2289 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2290 assert(vtype == type || (type == T_PRIMITIVE_OBJECT && vtype == T_OBJECT), "putter must accept the expected value");
2291 }
2292 #endif // ASSERT
2293 }
2294 #endif //PRODUCT
2295
2296 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2297
2298 Node* receiver = argument(0); // type: oop
2299
2300 // Build address expression.
2301 Node* heap_base_oop = top();
2302
2303 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2304 Node* base = argument(1); // type: oop
2305 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2306 Node* offset = argument(2); // type: long
2307 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2308 // to be plain byte offsets, which are also the same as those accepted
2309 // by oopDesc::field_addr.
2310 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2311 "fieldOffset must be byte-scaled");
2312
2313 ciInlineKlass* inline_klass = NULL;
2314 if (type == T_PRIMITIVE_OBJECT) {
2315 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2316 if (cls == NULL || cls->const_oop() == NULL) {
2317 return false;
2318 }
2319 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2320 if (!mirror_type->is_inlinetype()) {
2321 return false;
2322 }
2323 inline_klass = mirror_type->as_inline_klass();
2324 }
2325
2326 if (base->is_InlineTypeBase()) {
2327 InlineTypeBaseNode* vt = base->as_InlineTypeBase();
2328 if (is_store) {
2329 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->isa_inlinetype() || !_gvn.type(vt)->is_inlinetype()->larval()) {
2330 return false;
2331 }
2332 base = vt->get_oop();
2333 } else {
2334 if (offset->is_Con()) {
2335 long off = find_long_con(offset, 0);
2336 ciInlineKlass* vk = vt->type()->inline_klass();
2337 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2338 return false;
2339 }
2340
2341 ciField* field = vk->get_non_flattened_field_by_offset(off);
2342 if (field != NULL) {
2343 BasicType bt = field->layout_type();
2344 if (bt == T_ARRAY || bt == T_NARROWOOP || (bt == T_PRIMITIVE_OBJECT && !field->is_flattened())) {
2345 bt = T_OBJECT;
2346 }
2347 if (bt == type && (bt != T_PRIMITIVE_OBJECT || field->type() == inline_klass)) {
2348 set_result(vt->field_value_by_offset(off, false));
2349 return true;
2350 }
2351 }
2352 }
2353 if (vt->is_InlineType()) {
2354 // Re-execute the unsafe access if allocation triggers deoptimization.
2355 PreserveReexecuteState preexecs(this);
2356 jvms()->set_should_reexecute(true);
2357 vt = vt->buffer(this);
2358 }
2359 base = vt->get_oop();
2360 }
2361 }
2362
2363 // 32-bit machines ignore the high half!
2364 offset = ConvL2X(offset);
2365
2366 // Save state and restore on bailout
2367 uint old_sp = sp();
2368 SafePointNode* old_map = clone_map();
2369
2370 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2371
2372 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2373 if (type != T_OBJECT && (inline_klass == NULL || !inline_klass->has_object_fields())) {
2374 decorators |= IN_NATIVE; // off-heap primitive access
2375 } else {
2376 set_map(old_map);
2377 set_sp(old_sp);
2378 return false; // off-heap oop accesses are not supported
2379 }
2380 } else {
2381 heap_base_oop = base; // on-heap or mixed access
2382 }
2383
2384 // Can base be NULL? Otherwise, always on-heap access.
2385 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2386
2387 if (!can_access_non_heap) {
2388 decorators |= IN_HEAP;
2389 }
2390
2391 Node* val = is_store ? argument(4 + (type == T_PRIMITIVE_OBJECT ? 1 : 0)) : NULL;
2392
2393 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2394 if (adr_type == TypePtr::NULL_PTR) {
2395 set_map(old_map);
2396 set_sp(old_sp);
2397 return false; // off-heap access with zero address
2398 }
2399
2400 // Try to categorize the address.
2401 Compile::AliasType* alias_type = C->alias_type(adr_type);
2402 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2403
2404 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2405 alias_type->adr_type() == TypeAryPtr::RANGE) {
2406 set_map(old_map);
2407 set_sp(old_sp);
2408 return false; // not supported
2409 }
2410
2411 bool mismatched = false;
2412 BasicType bt = T_ILLEGAL;
2413 ciField* field = NULL;
2414 if (adr_type->isa_instptr()) {
2415 const TypeInstPtr* instptr = adr_type->is_instptr();
2416 ciInstanceKlass* k = instptr->klass()->as_instance_klass();
2417 int off = instptr->offset();
2418 if (instptr->const_oop() != NULL &&
2419 instptr->klass() == ciEnv::current()->Class_klass() &&
2420 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) {
2421 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2422 field = k->get_field_by_offset(off, true);
2423 } else {
2424 field = k->get_non_flattened_field_by_offset(off);
2425 }
2426 if (field != NULL) {
2427 bt = field->layout_type();
2428 }
2429 assert(bt == alias_type->basic_type() || bt == T_PRIMITIVE_OBJECT, "should match");
2430 if (field != NULL && bt == T_PRIMITIVE_OBJECT && !field->is_flattened()) {
2431 bt = T_OBJECT;
2432 }
2433 } else {
2434 bt = alias_type->basic_type();
2435 }
2436
2437 if (bt != T_ILLEGAL) {
2438 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2439 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2440 // Alias type doesn't differentiate between byte[] and boolean[]).
2441 // Use address type to get the element type.
2442 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2443 }
2444 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2445 // accessing an array field with getReference is not a mismatch
2446 bt = T_OBJECT;
2447 }
2448 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2449 // Don't intrinsify mismatched object accesses
2450 set_map(old_map);
2451 set_sp(old_sp);
2452 return false;
2453 }
2454 mismatched = (bt != type);
2455 } else if (alias_type->adr_type()->isa_oopptr()) {
2456 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2457 }
2458
2459 if (type == T_PRIMITIVE_OBJECT) {
2460 if (adr_type->isa_instptr()) {
2461 if (field == NULL || field->type() != inline_klass) {
2462 mismatched = true;
2463 }
2464 } else if (adr_type->isa_aryptr()) {
2465 const Type* elem = adr_type->is_aryptr()->elem();
2466 if (!elem->isa_inlinetype()) {
2467 mismatched = true;
2468 } else if (elem->inline_klass() != inline_klass) {
2469 mismatched = true;
2470 }
2471 } else {
2472 mismatched = true;
2473 }
2474 if (is_store) {
2475 const Type* val_t = _gvn.type(val);
2476 if (!(val_t->isa_inlinetype() || val_t->is_inlinetypeptr()) || val_t->inline_klass() != inline_klass) {
2477 set_map(old_map);
2478 set_sp(old_sp);
2479 return false;
2480 }
2481 }
2482 }
2483
2484 old_map->destruct(&_gvn);
2485 assert(!mismatched || type == T_PRIMITIVE_OBJECT || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2486
2487 if (mismatched) {
2488 decorators |= C2_MISMATCHED;
2489 }
2490
2491 // First guess at the value type.
2492 const Type *value_type = Type::get_const_basic_type(type);
2493
2494 // Figure out the memory ordering.
2495 decorators |= mo_decorator_for_access_kind(kind);
2496
2497 if (!is_store) {
2498 if (type == T_OBJECT) {
2499 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2500 if (tjp != NULL) {
2501 value_type = tjp;
2502 }
2503 } else if (type == T_PRIMITIVE_OBJECT) {
2504 value_type = NULL;
2505 }
2506 }
2507
2508 receiver = null_check(receiver);
2509 if (stopped()) {
2510 return true;
2511 }
2512 // Heap pointers get a null-check from the interpreter,
2513 // as a courtesy. However, this is not guaranteed by Unsafe,
2514 // and it is not possible to fully distinguish unintended nulls
2515 // from intended ones in this API.
2516
2517 if (!is_store) {
2518 Node* p = NULL;
2519 // Try to constant fold a load from a constant field
2520
2521 if (heap_base_oop != top() && field != NULL && field->is_constant() && !field->is_flattened() && !mismatched) {
2522 // final or stable field
2523 p = make_constant_from_field(field, heap_base_oop);
2524 }
2525
2526 if (p == NULL) { // Could not constant fold the load
2527 if (type == T_PRIMITIVE_OBJECT) {
2528 if (adr_type->isa_instptr() && !mismatched) {
2529 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2530 int offset = adr_type->is_instptr()->offset();
2531 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, base, holder, offset, decorators);
2532 } else {
2533 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, adr, NULL, 0, decorators);
2534 }
2535 } else {
2536 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2537 const TypeOopPtr* ptr = value_type->make_oopptr();
2538 if (ptr != NULL && ptr->is_inlinetypeptr()) {
2539 // Load a non-flattened inline type from memory
2540 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2541 }
2542 }
2543 // Normalize the value returned by getBoolean in the following cases
2544 if (type == T_BOOLEAN &&
2545 (mismatched ||
2546 heap_base_oop == top() || // - heap_base_oop is NULL or
2547 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2548 // and the unsafe access is made to large offset
2549 // (i.e., larger than the maximum offset necessary for any
2550 // field access)
2551 ) {
2552 IdealKit ideal = IdealKit(this);
2553 #define __ ideal.
2554 IdealVariable normalized_result(ideal);
2555 __ declarations_done();
2556 __ set(normalized_result, p);
2557 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2558 __ set(normalized_result, ideal.ConI(1));
2559 ideal.end_if();
2560 final_sync(ideal);
2561 p = __ value(normalized_result);
2562 #undef __
2563 }
2564 }
2565 if (type == T_ADDRESS) {
2566 p = gvn().transform(new CastP2XNode(NULL, p));
2567 p = ConvX2UL(p);
2568 }
2569 // The load node has the control of the preceding MemBarCPUOrder. All
2570 // following nodes will have the control of the MemBarCPUOrder inserted at
2571 // the end of this method. So, pushing the load onto the stack at a later
2572 // point is fine.
2573 set_result(p);
2574 } else {
2575 if (bt == T_ADDRESS) {
2576 // Repackage the long as a pointer.
2577 val = ConvL2X(val);
2578 val = gvn().transform(new CastX2PNode(val));
2579 }
2580 if (type == T_PRIMITIVE_OBJECT) {
2581 if (adr_type->isa_instptr() && !mismatched) {
2582 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2583 int offset = adr_type->is_instptr()->offset();
2584 val->as_InlineTypeBase()->store_flattened(this, base, base, holder, offset, decorators);
2585 } else {
2586 val->as_InlineTypeBase()->store_flattened(this, base, adr, NULL, 0, decorators);
2587 }
2588 } else {
2589 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2590 }
2591 }
2592
2593 if (argument(1)->is_InlineType() && is_store) {
2594 InlineTypeBaseNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(base)->inline_klass());
2595 value = value->make_larval(this, false);
2596 replace_in_map(argument(1), value);
2597 }
2598
2599 return true;
2600 }
2601
2602 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2603 Node* receiver = argument(0);
2604 Node* value = argument(1);
2605 if (!value->is_InlineTypeBase()) {
2606 return false;
2607 }
2608
2609 receiver = null_check(receiver);
2610 if (stopped()) {
2611 return true;
2612 }
2613
2614 set_result(value->as_InlineTypeBase()->make_larval(this, true));
2615 return true;
2616 }
2617
2618 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2619 Node* receiver = argument(0);
2620 Node* buffer = argument(1);
2621 if (!buffer->is_InlineType()) {
2622 return false;
2623 }
2624 InlineTypeNode* vt = buffer->as_InlineType();
2625 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_inlinetype()->larval()) {
2626 return false;
2627 }
2628
2629 receiver = null_check(receiver);
2630 if (stopped()) {
2631 return true;
2632 }
2633
2634 set_result(vt->finish_larval(this));
2635 return true;
2636 }
2637
2638 //----------------------------inline_unsafe_load_store----------------------------
2639 // This method serves a couple of different customers (depending on LoadStoreKind):
2640 //
2641 // LS_cmp_swap:
2642 //
2643 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2644 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2645 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2646 //
2647 // LS_cmp_swap_weak:
2648 //
2649 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2650 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2651 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2652 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2653 //
2654 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2655 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2656 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2657 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2658 //
2659 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2660 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2661 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2662 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2663 //
2664 // LS_cmp_exchange:
2665 //
2666 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2667 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2668 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2669 //
2670 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2671 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2672 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2673 //
2674 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2675 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2676 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2677 //
2678 // LS_get_add:
2679 //
2680 // int getAndAddInt( Object o, long offset, int delta)
2681 // long getAndAddLong(Object o, long offset, long delta)
2682 //
2683 // LS_get_set:
2684 //
2685 // int getAndSet(Object o, long offset, int newValue)
2686 // long getAndSet(Object o, long offset, long newValue)
2687 // Object getAndSet(Object o, long offset, Object newValue)
2688 //
2689 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2690 // This basic scheme here is the same as inline_unsafe_access, but
2691 // differs in enough details that combining them would make the code
2692 // overly confusing. (This is a true fact! I originally combined
2693 // them, but even I was confused by it!) As much code/comments as
2694 // possible are retained from inline_unsafe_access though to make
2695 // the correspondences clearer. - dl
2696
2697 if (callee()->is_static()) return false; // caller must have the capability!
2698
2699 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2700 decorators |= mo_decorator_for_access_kind(access_kind);
2701
2702 #ifndef PRODUCT
2703 BasicType rtype;
2704 {
2705 ResourceMark rm;
2706 // Check the signatures.
2707 ciSignature* sig = callee()->signature();
2708 rtype = sig->return_type()->basic_type();
2709 switch(kind) {
2710 case LS_get_add:
2711 case LS_get_set: {
2712 // Check the signatures.
2713 #ifdef ASSERT
2714 assert(rtype == type, "get and set must return the expected type");
2715 assert(sig->count() == 3, "get and set has 3 arguments");
2716 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2717 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2718 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2719 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2720 #endif // ASSERT
2721 break;
2722 }
2723 case LS_cmp_swap:
2724 case LS_cmp_swap_weak: {
2725 // Check the signatures.
2726 #ifdef ASSERT
2727 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2728 assert(sig->count() == 4, "CAS has 4 arguments");
2729 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2730 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2731 #endif // ASSERT
2732 break;
2733 }
2734 case LS_cmp_exchange: {
2735 // Check the signatures.
2736 #ifdef ASSERT
2737 assert(rtype == type, "CAS must return the expected type");
2738 assert(sig->count() == 4, "CAS has 4 arguments");
2739 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2740 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2741 #endif // ASSERT
2742 break;
2743 }
2744 default:
2745 ShouldNotReachHere();
2746 }
2747 }
2748 #endif //PRODUCT
2749
2750 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2751
2752 // Get arguments:
2753 Node* receiver = NULL;
2754 Node* base = NULL;
2755 Node* offset = NULL;
2756 Node* oldval = NULL;
2757 Node* newval = NULL;
2758 switch(kind) {
2759 case LS_cmp_swap:
2760 case LS_cmp_swap_weak:
2761 case LS_cmp_exchange: {
2762 const bool two_slot_type = type2size[type] == 2;
2763 receiver = argument(0); // type: oop
2764 base = argument(1); // type: oop
2765 offset = argument(2); // type: long
2766 oldval = argument(4); // type: oop, int, or long
2767 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2768 break;
2769 }
2770 case LS_get_add:
2771 case LS_get_set: {
2772 receiver = argument(0); // type: oop
2773 base = argument(1); // type: oop
2774 offset = argument(2); // type: long
2775 oldval = NULL;
2776 newval = argument(4); // type: oop, int, or long
2777 break;
2778 }
2779 default:
2780 ShouldNotReachHere();
2781 }
2782
2783 // Build field offset expression.
2784 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2785 // to be plain byte offsets, which are also the same as those accepted
2786 // by oopDesc::field_addr.
2787 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2788 // 32-bit machines ignore the high half of long offsets
2789 offset = ConvL2X(offset);
2790 // Save state and restore on bailout
2791 uint old_sp = sp();
2792 SafePointNode* old_map = clone_map();
2793 Node* adr = make_unsafe_address(base, offset,type, false);
2794 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2795
2796 Compile::AliasType* alias_type = C->alias_type(adr_type);
2797 BasicType bt = alias_type->basic_type();
2798 if (bt != T_ILLEGAL &&
2799 (is_reference_type(bt) != (type == T_OBJECT))) {
2800 // Don't intrinsify mismatched object accesses.
2801 set_map(old_map);
2802 set_sp(old_sp);
2803 return false;
2804 }
2805
2806 old_map->destruct(&_gvn);
2807
2808 // For CAS, unlike inline_unsafe_access, there seems no point in
2809 // trying to refine types. Just use the coarse types here.
2810 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2811 const Type *value_type = Type::get_const_basic_type(type);
2812
2813 switch (kind) {
2814 case LS_get_set:
2815 case LS_cmp_exchange: {
2816 if (type == T_OBJECT) {
2817 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2818 if (tjp != NULL) {
2819 value_type = tjp;
2820 }
2821 }
2822 break;
2823 }
2824 case LS_cmp_swap:
2825 case LS_cmp_swap_weak:
2826 case LS_get_add:
2827 break;
2828 default:
2829 ShouldNotReachHere();
2830 }
2831
2832 // Null check receiver.
2833 receiver = null_check(receiver);
2834 if (stopped()) {
2835 return true;
2836 }
2837
2838 int alias_idx = C->get_alias_index(adr_type);
2839
2840 if (is_reference_type(type)) {
2841 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2842
2843 if (oldval != NULL && oldval->is_InlineType()) {
2844 // Re-execute the unsafe access if allocation triggers deoptimization.
2845 PreserveReexecuteState preexecs(this);
2846 jvms()->set_should_reexecute(true);
2847 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2848 }
2849 if (newval != NULL && newval->is_InlineType()) {
2850 // Re-execute the unsafe access if allocation triggers deoptimization.
2851 PreserveReexecuteState preexecs(this);
2852 jvms()->set_should_reexecute(true);
2853 newval = newval->as_InlineType()->buffer(this)->get_oop();
2854 }
2855
2856 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2857 // could be delayed during Parse (for example, in adjust_map_after_if()).
2858 // Execute transformation here to avoid barrier generation in such case.
2859 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2860 newval = _gvn.makecon(TypePtr::NULL_PTR);
2861
2862 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2863 // Refine the value to a null constant, when it is known to be null
2864 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2865 }
2866 }
2867
2868 Node* result = NULL;
2869 switch (kind) {
2870 case LS_cmp_exchange: {
2871 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2872 oldval, newval, value_type, type, decorators);
2873 break;
2874 }
2875 case LS_cmp_swap_weak:
2876 decorators |= C2_WEAK_CMPXCHG;
2877 case LS_cmp_swap: {
2878 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2879 oldval, newval, value_type, type, decorators);
2880 break;
2881 }
2882 case LS_get_set: {
2883 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2884 newval, value_type, type, decorators);
2885 break;
2886 }
2887 case LS_get_add: {
2888 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2889 newval, value_type, type, decorators);
2890 break;
2891 }
2892 default:
2893 ShouldNotReachHere();
2894 }
2895
2896 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2897 set_result(result);
2898 return true;
2899 }
2900
2901 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2902 // Regardless of form, don't allow previous ld/st to move down,
2903 // then issue acquire, release, or volatile mem_bar.
2904 insert_mem_bar(Op_MemBarCPUOrder);
2905 switch(id) {
2906 case vmIntrinsics::_loadFence:
2907 insert_mem_bar(Op_LoadFence);
2908 return true;
2909 case vmIntrinsics::_storeFence:
2910 insert_mem_bar(Op_StoreFence);
2911 return true;
2912 case vmIntrinsics::_storeStoreFence:
2913 insert_mem_bar(Op_StoreStoreFence);
2914 return true;
2915 case vmIntrinsics::_fullFence:
2916 insert_mem_bar(Op_MemBarVolatile);
2917 return true;
2918 default:
2919 fatal_unexpected_iid(id);
2920 return false;
2921 }
2922 }
2923
2924 bool LibraryCallKit::inline_onspinwait() {
2925 insert_mem_bar(Op_OnSpinWait);
2926 return true;
2927 }
2928
2929 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2930 if (!kls->is_Con()) {
2931 return true;
2932 }
2933 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2934 if (klsptr == NULL) {
2935 return true;
2936 }
2937 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2938 // don't need a guard for a klass that is already initialized
2939 return !ik->is_initialized();
2940 }
2941
2942 //----------------------------inline_unsafe_writeback0-------------------------
2943 // public native void Unsafe.writeback0(long address)
2944 bool LibraryCallKit::inline_unsafe_writeback0() {
2945 if (!Matcher::has_match_rule(Op_CacheWB)) {
2946 return false;
2947 }
2948 #ifndef PRODUCT
2949 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2950 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2951 ciSignature* sig = callee()->signature();
2952 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2953 #endif
2954 null_check_receiver(); // null-check, then ignore
2955 Node *addr = argument(1);
2956 addr = new CastX2PNode(addr);
2957 addr = _gvn.transform(addr);
2958 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2959 flush = _gvn.transform(flush);
2960 set_memory(flush, TypeRawPtr::BOTTOM);
2961 return true;
2962 }
2963
2964 //----------------------------inline_unsafe_writeback0-------------------------
2965 // public native void Unsafe.writeback0(long address)
2966 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2967 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2968 return false;
2969 }
2970 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2971 return false;
2972 }
2973 #ifndef PRODUCT
2974 assert(Matcher::has_match_rule(Op_CacheWB),
2975 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2976 : "found match rule for CacheWBPostSync but not CacheWB"));
2977
2978 #endif
2979 null_check_receiver(); // null-check, then ignore
2980 Node *sync;
2981 if (is_pre) {
2982 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2983 } else {
2984 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2985 }
2986 sync = _gvn.transform(sync);
2987 set_memory(sync, TypeRawPtr::BOTTOM);
2988 return true;
2989 }
2990
2991 //----------------------------inline_unsafe_allocate---------------------------
2992 // public native Object Unsafe.allocateInstance(Class<?> cls);
2993 bool LibraryCallKit::inline_unsafe_allocate() {
2994 if (callee()->is_static()) return false; // caller must have the capability!
2995
2996 null_check_receiver(); // null-check, then ignore
2997 Node* cls = null_check(argument(1));
2998 if (stopped()) return true;
2999
3000 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3001 kls = null_check(kls);
3002 if (stopped()) return true; // argument was like int.class
3003
3004 Node* test = NULL;
3005 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3006 // Note: The argument might still be an illegal value like
3007 // Serializable.class or Object[].class. The runtime will handle it.
3008 // But we must make an explicit check for initialization.
3009 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3010 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3011 // can generate code to load it as unsigned byte.
3012 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3013 Node* bits = intcon(InstanceKlass::fully_initialized);
3014 test = _gvn.transform(new SubINode(inst, bits));
3015 // The 'test' is non-zero if we need to take a slow path.
3016 }
3017 Node* obj = NULL;
3018 ciKlass* klass = _gvn.type(kls)->is_klassptr()->klass();
3019 if (klass->is_inlinetype()) {
3020 obj = InlineTypeNode::make_default(_gvn, klass->as_inline_klass())->buffer(this);
3021 } else {
3022 obj = new_instance(kls, test);
3023 }
3024 set_result(obj);
3025 return true;
3026 }
3027
3028 //------------------------inline_native_time_funcs--------------
3029 // inline code for System.currentTimeMillis() and System.nanoTime()
3030 // these have the same type and signature
3031 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3032 const TypeFunc* tf = OptoRuntime::void_long_Type();
3033 const TypePtr* no_memory_effects = NULL;
3034 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3035 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3036 #ifdef ASSERT
3037 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3038 assert(value_top == top(), "second value must be top");
3039 #endif
3040 set_result(value);
3041 return true;
3042 }
3043
3044 #ifdef JFR_HAVE_INTRINSICS
3045
3046 /**
3047 * if oop->klass != null
3048 * // normal class
3049 * epoch = _epoch_state ? 2 : 1
3050 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3051 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3052 * }
3053 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3054 * else
3055 * // primitive class
3056 * if oop->array_klass != null
3057 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3058 * else
3059 * id = LAST_TYPE_ID + 1 // void class path
3060 * if (!signaled)
3061 * signaled = true
3062 */
3063 bool LibraryCallKit::inline_native_classID() {
3064 Node* cls = argument(0);
3065
3066 IdealKit ideal(this);
3067 #define __ ideal.
3068 IdealVariable result(ideal); __ declarations_done();
3069 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3070 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3071 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3072
3073
3074 __ if_then(kls, BoolTest::ne, null()); {
3075 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3076 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3077
3078 Node* epoch_address = makecon(TypeRawPtr::make(Jfr::epoch_address()));
3079 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3080 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3081 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3082 mask = _gvn.transform(new OrLNode(mask, epoch));
3083 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3084
3085 float unlikely = PROB_UNLIKELY(0.999);
3086 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3087 sync_kit(ideal);
3088 make_runtime_call(RC_LEAF,
3089 OptoRuntime::get_class_id_intrinsic_Type(),
3090 CAST_FROM_FN_PTR(address, Jfr::get_class_id_intrinsic),
3091 "get_class_id_intrinsic",
3092 TypePtr::BOTTOM,
3093 kls);
3094 ideal.sync_kit(this);
3095 } __ end_if();
3096
3097 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3098 } __ else_(); {
3099 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3100 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3101 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3102 __ if_then(array_kls, BoolTest::ne, null()); {
3103 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3104 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3105 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3106 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3107 } __ else_(); {
3108 // void class case
3109 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3110 } __ end_if();
3111
3112 Node* signaled_flag_address = makecon(TypeRawPtr::make(Jfr::signal_address()));
3113 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3114 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3115 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3116 } __ end_if();
3117 } __ end_if();
3118
3119 final_sync(ideal);
3120 set_result(ideal.value(result));
3121 #undef __
3122 return true;
3123 }
3124
3125 bool LibraryCallKit::inline_native_getEventWriter() {
3126 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3127
3128 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3129 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3130
3131 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3132
3133 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3134 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3135
3136 IfNode* iff_jobj_null =
3137 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3138
3139 enum { _normal_path = 1,
3140 _null_path = 2,
3141 PATH_LIMIT };
3142
3143 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3144 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3145
3146 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3147 result_rgn->init_req(_null_path, jobj_is_null);
3148 result_val->init_req(_null_path, null());
3149
3150 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3151 set_control(jobj_is_not_null);
3152 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3153 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3154 result_rgn->init_req(_normal_path, control());
3155 result_val->init_req(_normal_path, res);
3156
3157 set_result(result_rgn, result_val);
3158
3159 return true;
3160 }
3161
3162 #endif // JFR_HAVE_INTRINSICS
3163
3164 //------------------------inline_native_currentThread------------------
3165 bool LibraryCallKit::inline_native_currentThread() {
3166 Node* junk = NULL;
3167 set_result(generate_current_thread(junk));
3168 return true;
3169 }
3170
3171 //-----------------------load_klass_from_mirror_common-------------------------
3172 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3173 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3174 // and branch to the given path on the region.
3175 // If never_see_null, take an uncommon trap on null, so we can optimistically
3176 // compile for the non-null case.
3177 // If the region is NULL, force never_see_null = true.
3178 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3179 bool never_see_null,
3180 RegionNode* region,
3181 int null_path,
3182 int offset) {
3183 if (region == NULL) never_see_null = true;
3184 Node* p = basic_plus_adr(mirror, offset);
3185 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3186 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3187 Node* null_ctl = top();
3188 kls = null_check_oop(kls, &null_ctl, never_see_null);
3189 if (region != NULL) {
3190 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3191 region->init_req(null_path, null_ctl);
3192 } else {
3193 assert(null_ctl == top(), "no loose ends");
3194 }
3195 return kls;
3196 }
3197
3198 //--------------------(inline_native_Class_query helpers)---------------------
3199 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3200 // Fall through if (mods & mask) == bits, take the guard otherwise.
3201 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3202 // Branch around if the given klass has the given modifier bit set.
3203 // Like generate_guard, adds a new path onto the region.
3204 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3205 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3206 Node* mask = intcon(modifier_mask);
3207 Node* bits = intcon(modifier_bits);
3208 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3209 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3210 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3211 return generate_fair_guard(bol, region);
3212 }
3213
3214 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3215 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3216 }
3217 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3218 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3219 }
3220
3221 //-------------------------inline_native_Class_query-------------------
3222 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3223 const Type* return_type = TypeInt::BOOL;
3224 Node* prim_return_value = top(); // what happens if it's a primitive class?
3225 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3226 bool expect_prim = false; // most of these guys expect to work on refs
3227
3228 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3229
3230 Node* mirror = argument(0);
3231 Node* obj = top();
3232
3233 switch (id) {
3234 case vmIntrinsics::_isInstance:
3235 // nothing is an instance of a primitive type
3236 prim_return_value = intcon(0);
3237 obj = argument(1);
3238 break;
3239 case vmIntrinsics::_getModifiers:
3240 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3241 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3242 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3243 break;
3244 case vmIntrinsics::_isInterface:
3245 prim_return_value = intcon(0);
3246 break;
3247 case vmIntrinsics::_isArray:
3248 prim_return_value = intcon(0);
3249 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3250 break;
3251 case vmIntrinsics::_isPrimitive:
3252 prim_return_value = intcon(1);
3253 expect_prim = true; // obviously
3254 break;
3255 case vmIntrinsics::_isHidden:
3256 prim_return_value = intcon(0);
3257 break;
3258 case vmIntrinsics::_getSuperclass:
3259 prim_return_value = null();
3260 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3261 break;
3262 case vmIntrinsics::_getClassAccessFlags:
3263 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3264 return_type = TypeInt::INT; // not bool! 6297094
3265 break;
3266 default:
3267 fatal_unexpected_iid(id);
3268 break;
3269 }
3270
3271 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3272 if (mirror_con == NULL) return false; // cannot happen?
3273
3274 #ifndef PRODUCT
3275 if (C->print_intrinsics() || C->print_inlining()) {
3276 ciType* k = mirror_con->java_mirror_type();
3277 if (k) {
3278 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3279 k->print_name();
3280 tty->cr();
3281 }
3282 }
3283 #endif
3284
3285 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3286 RegionNode* region = new RegionNode(PATH_LIMIT);
3287 record_for_igvn(region);
3288 PhiNode* phi = new PhiNode(region, return_type);
3289
3290 // The mirror will never be null of Reflection.getClassAccessFlags, however
3291 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3292 // if it is. See bug 4774291.
3293
3294 // For Reflection.getClassAccessFlags(), the null check occurs in
3295 // the wrong place; see inline_unsafe_access(), above, for a similar
3296 // situation.
3297 mirror = null_check(mirror);
3298 // If mirror or obj is dead, only null-path is taken.
3299 if (stopped()) return true;
3300
3301 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3302
3303 // Now load the mirror's klass metaobject, and null-check it.
3304 // Side-effects region with the control path if the klass is null.
3305 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3306 // If kls is null, we have a primitive mirror.
3307 phi->init_req(_prim_path, prim_return_value);
3308 if (stopped()) { set_result(region, phi); return true; }
3309 bool safe_for_replace = (region->in(_prim_path) == top());
3310
3311 Node* p; // handy temp
3312 Node* null_ctl;
3313
3314 // Now that we have the non-null klass, we can perform the real query.
3315 // For constant classes, the query will constant-fold in LoadNode::Value.
3316 Node* query_value = top();
3317 switch (id) {
3318 case vmIntrinsics::_isInstance:
3319 // nothing is an instance of a primitive type
3320 query_value = gen_instanceof(obj, kls, safe_for_replace);
3321 break;
3322
3323 case vmIntrinsics::_getModifiers:
3324 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3325 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3326 break;
3327
3328 case vmIntrinsics::_isInterface:
3329 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3330 if (generate_interface_guard(kls, region) != NULL)
3331 // A guard was added. If the guard is taken, it was an interface.
3332 phi->add_req(intcon(1));
3333 // If we fall through, it's a plain class.
3334 query_value = intcon(0);
3335 break;
3336
3337 case vmIntrinsics::_isArray:
3338 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3339 if (generate_array_guard(kls, region) != NULL)
3340 // A guard was added. If the guard is taken, it was an array.
3341 phi->add_req(intcon(1));
3342 // If we fall through, it's a plain class.
3343 query_value = intcon(0);
3344 break;
3345
3346 case vmIntrinsics::_isPrimitive:
3347 query_value = intcon(0); // "normal" path produces false
3348 break;
3349
3350 case vmIntrinsics::_isHidden:
3351 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3352 if (generate_hidden_class_guard(kls, region) != NULL)
3353 // A guard was added. If the guard is taken, it was an hidden class.
3354 phi->add_req(intcon(1));
3355 // If we fall through, it's a plain class.
3356 query_value = intcon(0);
3357 break;
3358
3359
3360 case vmIntrinsics::_getSuperclass:
3361 // The rules here are somewhat unfortunate, but we can still do better
3362 // with random logic than with a JNI call.
3363 // Interfaces store null or Object as _super, but must report null.
3364 // Arrays store an intermediate super as _super, but must report Object.
3365 // Other types can report the actual _super.
3366 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3367 if (generate_interface_guard(kls, region) != NULL)
3368 // A guard was added. If the guard is taken, it was an interface.
3369 phi->add_req(null());
3370 if (generate_array_guard(kls, region) != NULL)
3371 // A guard was added. If the guard is taken, it was an array.
3372 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3373 // If we fall through, it's a plain class. Get its _super.
3374 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3375 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3376 null_ctl = top();
3377 kls = null_check_oop(kls, &null_ctl);
3378 if (null_ctl != top()) {
3379 // If the guard is taken, Object.superClass is null (both klass and mirror).
3380 region->add_req(null_ctl);
3381 phi ->add_req(null());
3382 }
3383 if (!stopped()) {
3384 query_value = load_mirror_from_klass(kls);
3385 }
3386 break;
3387
3388 case vmIntrinsics::_getClassAccessFlags:
3389 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3390 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3391 break;
3392
3393 default:
3394 fatal_unexpected_iid(id);
3395 break;
3396 }
3397
3398 // Fall-through is the normal case of a query to a real class.
3399 phi->init_req(1, query_value);
3400 region->init_req(1, control());
3401
3402 C->set_has_split_ifs(true); // Has chance for split-if optimization
3403 set_result(region, phi);
3404 return true;
3405 }
3406
3407 //-------------------------inline_primitive_Class_conversion-------------------
3408 // public Class<T> java.lang.Class.asPrimaryType();
3409 // public Class<T> java.lang.Class.asValueType()
3410 bool LibraryCallKit::inline_primitive_Class_conversion(vmIntrinsics::ID id) {
3411 Node* mirror = argument(0); // Receiver Class
3412 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3413 if (mirror_con == NULL) {
3414 return false;
3415 }
3416
3417 bool is_val_mirror = true;
3418 ciType* tm = mirror_con->java_mirror_type(&is_val_mirror);
3419 if (tm != NULL) {
3420 Node* result = mirror;
3421 if (id == vmIntrinsics::_asPrimaryType && is_val_mirror) {
3422 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->ref_mirror()));
3423 } else if (id == vmIntrinsics::_asValueType) {
3424 if (!tm->is_inlinetype()) {
3425 return false; // Throw UnsupportedOperationException
3426 } else if (!is_val_mirror) {
3427 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->val_mirror()));
3428 }
3429 }
3430 set_result(result);
3431 return true;
3432 }
3433 return false;
3434 }
3435
3436 //-------------------------inline_Class_cast-------------------
3437 bool LibraryCallKit::inline_Class_cast() {
3438 Node* mirror = argument(0); // Class
3439 Node* obj = argument(1);
3440 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3441 if (mirror_con == NULL) {
3442 return false; // dead path (mirror->is_top()).
3443 }
3444 if (obj == NULL || obj->is_top()) {
3445 return false; // dead path
3446 }
3447 ciKlass* obj_klass = NULL;
3448 const Type* obj_t = _gvn.type(obj);
3449 if (obj->is_InlineType()) {
3450 obj_klass = obj_t->inline_klass();
3451 } else if (obj_t->isa_oopptr()) {
3452 obj_klass = obj_t->is_oopptr()->klass();
3453 }
3454
3455 // First, see if Class.cast() can be folded statically.
3456 // java_mirror_type() returns non-null for compile-time Class constants.
3457 bool requires_null_check = false;
3458 ciType* tm = mirror_con->java_mirror_type(&requires_null_check);
3459 // Check for null if casting to QMyValue
3460 if (tm != NULL && tm->is_klass() && obj_klass != NULL) {
3461 if (!obj_klass->is_loaded()) {
3462 // Don't use intrinsic when class is not loaded.
3463 return false;
3464 } else {
3465 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass);
3466 if (static_res == Compile::SSC_always_true) {
3467 // isInstance() is true - fold the code.
3468 if (requires_null_check) {
3469 obj = null_check(obj);
3470 }
3471 set_result(obj);
3472 return true;
3473 } else if (static_res == Compile::SSC_always_false) {
3474 // Don't use intrinsic, have to throw ClassCastException.
3475 // If the reference is null, the non-intrinsic bytecode will
3476 // be optimized appropriately.
3477 return false;
3478 }
3479 }
3480 }
3481
3482 // Bailout intrinsic and do normal inlining if exception path is frequent.
3483 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3484 return false;
3485 }
3486
3487 // Generate dynamic checks.
3488 // Class.cast() is java implementation of _checkcast bytecode.
3489 // Do checkcast (Parse::do_checkcast()) optimizations here.
3490
3491 if (requires_null_check) {
3492 obj = null_check(obj);
3493 }
3494 mirror = null_check(mirror);
3495 // If mirror is dead, only null-path is taken.
3496 if (stopped()) {
3497 return true;
3498 }
3499
3500 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3501 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
3502 RegionNode* region = new RegionNode(PATH_LIMIT);
3503 record_for_igvn(region);
3504
3505 // Now load the mirror's klass metaobject, and null-check it.
3506 // If kls is null, we have a primitive mirror and
3507 // nothing is an instance of a primitive type.
3508 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3509
3510 Node* res = top();
3511 if (!stopped()) {
3512 if (EnableValhalla && !requires_null_check) {
3513 // Check if we are casting to QMyValue
3514 Node* ctrl_val_mirror = generate_fair_guard(is_val_mirror(mirror), NULL);
3515 if (ctrl_val_mirror != NULL) {
3516 RegionNode* r = new RegionNode(3);
3517 record_for_igvn(r);
3518 r->init_req(1, control());
3519
3520 // Casting to QMyValue, check for null
3521 set_control(ctrl_val_mirror);
3522 { // PreserveJVMState because null check replaces obj in map
3523 PreserveJVMState pjvms(this);
3524 Node* null_ctr = top();
3525 null_check_oop(obj, &null_ctr);
3526 region->init_req(_npe_path, null_ctr);
3527 r->init_req(2, control());
3528 }
3529 set_control(_gvn.transform(r));
3530 }
3531 }
3532
3533 Node* bad_type_ctrl = top();
3534 // Do checkcast optimizations.
3535 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3536 region->init_req(_bad_type_path, bad_type_ctrl);
3537 }
3538 if (region->in(_prim_path) != top() ||
3539 region->in(_bad_type_path) != top() ||
3540 region->in(_npe_path) != top()) {
3541 // Let Interpreter throw ClassCastException.
3542 PreserveJVMState pjvms(this);
3543 set_control(_gvn.transform(region));
3544 uncommon_trap(Deoptimization::Reason_intrinsic,
3545 Deoptimization::Action_maybe_recompile);
3546 }
3547 if (!stopped()) {
3548 set_result(res);
3549 }
3550 return true;
3551 }
3552
3553
3554 //--------------------------inline_native_subtype_check------------------------
3555 // This intrinsic takes the JNI calls out of the heart of
3556 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3557 bool LibraryCallKit::inline_native_subtype_check() {
3558 // Pull both arguments off the stack.
3559 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3560 args[0] = argument(0);
3561 args[1] = argument(1);
3562 Node* klasses[2]; // corresponding Klasses: superk, subk
3563 klasses[0] = klasses[1] = top();
3564
3565 enum {
3566 // A full decision tree on {superc is prim, subc is prim}:
3567 _prim_0_path = 1, // {P,N} => false
3568 // {P,P} & superc!=subc => false
3569 _prim_same_path, // {P,P} & superc==subc => true
3570 _prim_1_path, // {N,P} => false
3571 _ref_subtype_path, // {N,N} & subtype check wins => true
3572 _both_ref_path, // {N,N} & subtype check loses => false
3573 PATH_LIMIT
3574 };
3575
3576 RegionNode* region = new RegionNode(PATH_LIMIT);
3577 RegionNode* prim_region = new RegionNode(2);
3578 Node* phi = new PhiNode(region, TypeInt::BOOL);
3579 record_for_igvn(region);
3580 record_for_igvn(prim_region);
3581
3582 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3583 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3584 int class_klass_offset = java_lang_Class::klass_offset();
3585
3586 // First null-check both mirrors and load each mirror's klass metaobject.
3587 int which_arg;
3588 for (which_arg = 0; which_arg <= 1; which_arg++) {
3589 Node* arg = args[which_arg];
3590 arg = null_check(arg);
3591 if (stopped()) break;
3592 args[which_arg] = arg;
3593
3594 Node* p = basic_plus_adr(arg, class_klass_offset);
3595 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3596 klasses[which_arg] = _gvn.transform(kls);
3597 }
3598
3599 // Having loaded both klasses, test each for null.
3600 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3601 for (which_arg = 0; which_arg <= 1; which_arg++) {
3602 Node* kls = klasses[which_arg];
3603 Node* null_ctl = top();
3604 kls = null_check_oop(kls, &null_ctl, never_see_null);
3605 if (which_arg == 0) {
3606 prim_region->init_req(1, null_ctl);
3607 } else {
3608 region->init_req(_prim_1_path, null_ctl);
3609 }
3610 if (stopped()) break;
3611 klasses[which_arg] = kls;
3612 }
3613
3614 if (!stopped()) {
3615 // now we have two reference types, in klasses[0..1]
3616 Node* subk = klasses[1]; // the argument to isAssignableFrom
3617 Node* superk = klasses[0]; // the receiver
3618 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3619 // If superc is an inline mirror, we also need to check if superc == subc because LMyValue
3620 // is not a subtype of QMyValue but due to subk == superk the subtype check will pass.
3621 generate_fair_guard(is_val_mirror(args[0]), prim_region);
3622 // now we have a successful reference subtype check
3623 region->set_req(_ref_subtype_path, control());
3624 }
3625
3626 // If both operands are primitive (both klasses null), then
3627 // we must return true when they are identical primitives.
3628 // It is convenient to test this after the first null klass check.
3629 // This path is also used if superc is a value mirror.
3630 set_control(_gvn.transform(prim_region));
3631 if (!stopped()) {
3632 // Since superc is primitive, make a guard for the superc==subc case.
3633 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3634 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3635 generate_fair_guard(bol_eq, region);
3636 if (region->req() == PATH_LIMIT+1) {
3637 // A guard was added. If the added guard is taken, superc==subc.
3638 region->swap_edges(PATH_LIMIT, _prim_same_path);
3639 region->del_req(PATH_LIMIT);
3640 }
3641 region->set_req(_prim_0_path, control()); // Not equal after all.
3642 }
3643
3644 // these are the only paths that produce 'true':
3645 phi->set_req(_prim_same_path, intcon(1));
3646 phi->set_req(_ref_subtype_path, intcon(1));
3647
3648 // pull together the cases:
3649 assert(region->req() == PATH_LIMIT, "sane region");
3650 for (uint i = 1; i < region->req(); i++) {
3651 Node* ctl = region->in(i);
3652 if (ctl == NULL || ctl == top()) {
3653 region->set_req(i, top());
3654 phi ->set_req(i, top());
3655 } else if (phi->in(i) == NULL) {
3656 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3657 }
3658 }
3659
3660 set_control(_gvn.transform(region));
3661 set_result(_gvn.transform(phi));
3662 return true;
3663 }
3664
3665 //---------------------generate_array_guard_common------------------------
3666 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
3667
3668 if (stopped()) {
3669 return NULL;
3670 }
3671
3672 // Like generate_guard, adds a new path onto the region.
3673 jint layout_con = 0;
3674 Node* layout_val = get_layout_helper(kls, layout_con);
3675 if (layout_val == NULL) {
3676 bool query = 0;
3677 switch(kind) {
3678 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
3679 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
3680 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
3681 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
3682 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
3683 default:
3684 ShouldNotReachHere();
3685 }
3686 if (!query) {
3687 return NULL; // never a branch
3688 } else { // always a branch
3689 Node* always_branch = control();
3690 if (region != NULL)
3691 region->add_req(always_branch);
3692 set_control(top());
3693 return always_branch;
3694 }
3695 }
3696 unsigned int value = 0;
3697 BoolTest::mask btest = BoolTest::illegal;
3698 switch(kind) {
3699 case ObjectArray:
3700 case NonObjectArray: {
3701 value = Klass::_lh_array_tag_obj_value;
3702 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3703 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
3704 break;
3705 }
3706 case TypeArray: {
3707 value = Klass::_lh_array_tag_type_value;
3708 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3709 btest = BoolTest::eq;
3710 break;
3711 }
3712 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
3713 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
3714 default:
3715 ShouldNotReachHere();
3716 }
3717 // Now test the correct condition.
3718 jint nval = (jint)value;
3719 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3720 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3721 return generate_fair_guard(bol, region);
3722 }
3723
3724
3725 //-----------------------inline_native_newArray--------------------------
3726 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
3727 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3728 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3729 Node* mirror;
3730 Node* count_val;
3731 if (uninitialized) {
3732 mirror = argument(1);
3733 count_val = argument(2);
3734 } else {
3735 mirror = argument(0);
3736 count_val = argument(1);
3737 }
3738
3739 mirror = null_check(mirror);
3740 // If mirror or obj is dead, only null-path is taken.
3741 if (stopped()) return true;
3742
3743 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3744 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3745 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3746 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3747 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3748
3749 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3750 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3751 result_reg, _slow_path);
3752 Node* normal_ctl = control();
3753 Node* no_array_ctl = result_reg->in(_slow_path);
3754
3755 // Generate code for the slow case. We make a call to newArray().
3756 set_control(no_array_ctl);
3757 if (!stopped()) {
3758 // Either the input type is void.class, or else the
3759 // array klass has not yet been cached. Either the
3760 // ensuing call will throw an exception, or else it
3761 // will cache the array klass for next time.
3762 PreserveJVMState pjvms(this);
3763 CallJavaNode* slow_call = NULL;
3764 if (uninitialized) {
3765 // Generate optimized virtual call (holder class 'Unsafe' is final)
3766 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3767 } else {
3768 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3769 }
3770 Node* slow_result = set_results_for_java_call(slow_call);
3771 // this->control() comes from set_results_for_java_call
3772 result_reg->set_req(_slow_path, control());
3773 result_val->set_req(_slow_path, slow_result);
3774 result_io ->set_req(_slow_path, i_o());
3775 result_mem->set_req(_slow_path, reset_memory());
3776 }
3777
3778 set_control(normal_ctl);
3779 if (!stopped()) {
3780 // Normal case: The array type has been cached in the java.lang.Class.
3781 // The following call works fine even if the array type is polymorphic.
3782 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3783 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3784 result_reg->init_req(_normal_path, control());
3785 result_val->init_req(_normal_path, obj);
3786 result_io ->init_req(_normal_path, i_o());
3787 result_mem->init_req(_normal_path, reset_memory());
3788
3789 if (uninitialized) {
3790 // Mark the allocation so that zeroing is skipped
3791 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3792 alloc->maybe_set_complete(&_gvn);
3793 }
3794 }
3795
3796 // Return the combined state.
3797 set_i_o( _gvn.transform(result_io) );
3798 set_all_memory( _gvn.transform(result_mem));
3799
3800 C->set_has_split_ifs(true); // Has chance for split-if optimization
3801 set_result(result_reg, result_val);
3802 return true;
3803 }
3804
3805 //----------------------inline_native_getLength--------------------------
3806 // public static native int java.lang.reflect.Array.getLength(Object array);
3807 bool LibraryCallKit::inline_native_getLength() {
3808 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3809
3810 Node* array = null_check(argument(0));
3811 // If array is dead, only null-path is taken.
3812 if (stopped()) return true;
3813
3814 // Deoptimize if it is a non-array.
3815 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3816
3817 if (non_array != NULL) {
3818 PreserveJVMState pjvms(this);
3819 set_control(non_array);
3820 uncommon_trap(Deoptimization::Reason_intrinsic,
3821 Deoptimization::Action_maybe_recompile);
3822 }
3823
3824 // If control is dead, only non-array-path is taken.
3825 if (stopped()) return true;
3826
3827 // The works fine even if the array type is polymorphic.
3828 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3829 Node* result = load_array_length(array);
3830
3831 C->set_has_split_ifs(true); // Has chance for split-if optimization
3832 set_result(result);
3833 return true;
3834 }
3835
3836 //------------------------inline_array_copyOf----------------------------
3837 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3838 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3839 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3840 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3841
3842 // Get the arguments.
3843 Node* original = argument(0);
3844 Node* start = is_copyOfRange? argument(1): intcon(0);
3845 Node* end = is_copyOfRange? argument(2): argument(1);
3846 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3847
3848 Node* newcopy = NULL;
3849
3850 // Set the original stack and the reexecute bit for the interpreter to reexecute
3851 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3852 { PreserveReexecuteState preexecs(this);
3853 jvms()->set_should_reexecute(true);
3854
3855 array_type_mirror = null_check(array_type_mirror);
3856 original = null_check(original);
3857
3858 // Check if a null path was taken unconditionally.
3859 if (stopped()) return true;
3860
3861 Node* orig_length = load_array_length(original);
3862
3863 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3864 klass_node = null_check(klass_node);
3865
3866 RegionNode* bailout = new RegionNode(1);
3867 record_for_igvn(bailout);
3868
3869 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3870 // Bail out if that is so.
3871 // Inline type array may have object field that would require a
3872 // write barrier. Conservatively, go to slow path.
3873 // TODO 8251971: Optimize for the case when flat src/dst are later found
3874 // to not contain oops (i.e., move this check to the macro expansion phase).
3875 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
3876 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
3877 ciKlass* klass = _gvn.type(klass_node)->is_klassptr()->klass();
3878 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
3879 // Can src array be flat and contain oops?
3880 (orig_t == NULL || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
3881 // Can dest array be flat and contain oops?
3882 klass->can_be_inline_array_klass() && (!klass->is_flat_array_klass() || klass->as_flat_array_klass()->element_klass()->as_inline_klass()->contains_oops());
3883 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
3884 if (not_objArray != NULL) {
3885 // Improve the klass node's type from the new optimistic assumption:
3886 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3887 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
3888 Node* cast = new CastPPNode(klass_node, akls);
3889 cast->init_req(0, control());
3890 klass_node = _gvn.transform(cast);
3891 }
3892
3893 // Bail out if either start or end is negative.
3894 generate_negative_guard(start, bailout, &start);
3895 generate_negative_guard(end, bailout, &end);
3896
3897 Node* length = end;
3898 if (_gvn.type(start) != TypeInt::ZERO) {
3899 length = _gvn.transform(new SubINode(end, start));
3900 }
3901
3902 // Bail out if length is negative.
3903 // Without this the new_array would throw
3904 // NegativeArraySizeException but IllegalArgumentException is what
3905 // should be thrown
3906 generate_negative_guard(length, bailout, &length);
3907
3908 // Handle inline type arrays
3909 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
3910 if (!stopped()) {
3911 orig_t = _gvn.type(original)->isa_aryptr();
3912 if (orig_t != NULL && orig_t->is_flat()) {
3913 // Src is flat, check that dest is flat as well
3914 if (exclude_flat) {
3915 // Dest can't be flat, bail out
3916 bailout->add_req(control());
3917 set_control(top());
3918 } else {
3919 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
3920 }
3921 } else if (UseFlatArray && (orig_t == NULL || !orig_t->is_not_flat()) &&
3922 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
3923 ((!klass->is_flat_array_klass() && klass->can_be_inline_array_klass()) || !can_validate)) {
3924 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
3925 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
3926 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
3927 if (orig_t != NULL) {
3928 orig_t = orig_t->cast_to_not_flat();
3929 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
3930 }
3931 }
3932 if (!can_validate) {
3933 // No validation. The subtype check emitted at macro expansion time will not go to the slow
3934 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
3935 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
3936 generate_fair_guard(null_free_array_test(klass_node), bailout);
3937 }
3938 }
3939
3940 if (bailout->req() > 1) {
3941 PreserveJVMState pjvms(this);
3942 set_control(_gvn.transform(bailout));
3943 uncommon_trap(Deoptimization::Reason_intrinsic,
3944 Deoptimization::Action_maybe_recompile);
3945 }
3946
3947 if (!stopped()) {
3948 // How many elements will we copy from the original?
3949 // The answer is MinI(orig_length - start, length).
3950 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3951 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3952
3953 // Generate a direct call to the right arraycopy function(s).
3954 // We know the copy is disjoint but we might not know if the
3955 // oop stores need checking.
3956 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3957 // This will fail a store-check if x contains any non-nulls.
3958
3959 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3960 // loads/stores but it is legal only if we're sure the
3961 // Arrays.copyOf would succeed. So we need all input arguments
3962 // to the copyOf to be validated, including that the copy to the
3963 // new array won't trigger an ArrayStoreException. That subtype
3964 // check can be optimized if we know something on the type of
3965 // the input array from type speculation.
3966 if (_gvn.type(klass_node)->singleton()) {
3967 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3968 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3969
3970 int test = C->static_subtype_check(superk, subk);
3971 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3972 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3973 if (t_original->speculative_type() != NULL) {
3974 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3975 }
3976 }
3977 }
3978
3979 bool validated = false;
3980 // Reason_class_check rather than Reason_intrinsic because we
3981 // want to intrinsify even if this traps.
3982 if (can_validate) {
3983 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3984
3985 if (not_subtype_ctrl != top()) {
3986 PreserveJVMState pjvms(this);
3987 set_control(not_subtype_ctrl);
3988 uncommon_trap(Deoptimization::Reason_class_check,
3989 Deoptimization::Action_make_not_entrant);
3990 assert(stopped(), "Should be stopped");
3991 }
3992 validated = true;
3993 }
3994
3995 if (!stopped()) {
3996 newcopy = new_array(klass_node, length, 0); // no arguments to push
3997
3998 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3999 load_object_klass(original), klass_node);
4000 if (!is_copyOfRange) {
4001 ac->set_copyof(validated);
4002 } else {
4003 ac->set_copyofrange(validated);
4004 }
4005 Node* n = _gvn.transform(ac);
4006 if (n == ac) {
4007 ac->connect_outputs(this);
4008 } else {
4009 assert(validated, "shouldn't transform if all arguments not validated");
4010 set_all_memory(n);
4011 }
4012 }
4013 }
4014 } // original reexecute is set back here
4015
4016 C->set_has_split_ifs(true); // Has chance for split-if optimization
4017 if (!stopped()) {
4018 set_result(newcopy);
4019 }
4020 return true;
4021 }
4022
4023
4024 //----------------------generate_virtual_guard---------------------------
4025 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4026 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4027 RegionNode* slow_region) {
4028 ciMethod* method = callee();
4029 int vtable_index = method->vtable_index();
4030 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4031 "bad index %d", vtable_index);
4032 // Get the Method* out of the appropriate vtable entry.
4033 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4034 vtable_index*vtableEntry::size_in_bytes() +
4035 vtableEntry::method_offset_in_bytes();
4036 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4037 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4038
4039 // Compare the target method with the expected method (e.g., Object.hashCode).
4040 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4041
4042 Node* native_call = makecon(native_call_addr);
4043 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4044 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4045
4046 return generate_slow_guard(test_native, slow_region);
4047 }
4048
4049 //-----------------------generate_method_call----------------------------
4050 // Use generate_method_call to make a slow-call to the real
4051 // method if the fast path fails. An alternative would be to
4052 // use a stub like OptoRuntime::slow_arraycopy_Java.
4053 // This only works for expanding the current library call,
4054 // not another intrinsic. (E.g., don't use this for making an
4055 // arraycopy call inside of the copyOf intrinsic.)
4056 CallJavaNode*
4057 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4058 // When compiling the intrinsic method itself, do not use this technique.
4059 guarantee(callee() != C->method(), "cannot make slow-call to self");
4060
4061 ciMethod* method = callee();
4062 // ensure the JVMS we have will be correct for this call
4063 guarantee(method_id == method->intrinsic_id(), "must match");
4064
4065 const TypeFunc* tf = TypeFunc::make(method);
4066 CallJavaNode* slow_call;
4067 if (is_static) {
4068 assert(!is_virtual, "");
4069 slow_call = new CallStaticJavaNode(C, tf,
4070 SharedRuntime::get_resolve_static_call_stub(), method);
4071 } else if (is_virtual) {
4072 null_check_receiver();
4073 int vtable_index = Method::invalid_vtable_index;
4074 if (UseInlineCaches) {
4075 // Suppress the vtable call
4076 } else {
4077 // hashCode and clone are not a miranda methods,
4078 // so the vtable index is fixed.
4079 // No need to use the linkResolver to get it.
4080 vtable_index = method->vtable_index();
4081 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4082 "bad index %d", vtable_index);
4083 }
4084 slow_call = new CallDynamicJavaNode(tf,
4085 SharedRuntime::get_resolve_virtual_call_stub(),
4086 method, vtable_index);
4087 } else { // neither virtual nor static: opt_virtual
4088 null_check_receiver();
4089 slow_call = new CallStaticJavaNode(C, tf,
4090 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4091 slow_call->set_optimized_virtual(true);
4092 }
4093 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4094 // To be able to issue a direct call (optimized virtual or virtual)
4095 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4096 // about the method being invoked should be attached to the call site to
4097 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4098 slow_call->set_override_symbolic_info(true);
4099 }
4100 set_arguments_for_java_call(slow_call);
4101 set_edges_for_java_call(slow_call);
4102 return slow_call;
4103 }
4104
4105
4106 /**
4107 * Build special case code for calls to hashCode on an object. This call may
4108 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4109 * slightly different code.
4110 */
4111 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4112 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4113 assert(!(is_virtual && is_static), "either virtual, special, or static");
4114
4115 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4116
4117 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4118 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4119 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4120 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4121 Node* obj = argument(0);
4122
4123 if (obj->is_InlineType() || gvn().type(obj)->is_inlinetypeptr()) {
4124 return false;
4125 }
4126
4127 if (!is_static) {
4128 // Check for hashing null object
4129 obj = null_check_receiver();
4130 if (stopped()) return true; // unconditionally null
4131 result_reg->init_req(_null_path, top());
4132 result_val->init_req(_null_path, top());
4133 } else {
4134 // Do a null check, and return zero if null.
4135 // System.identityHashCode(null) == 0
4136 Node* null_ctl = top();
4137 obj = null_check_oop(obj, &null_ctl);
4138 result_reg->init_req(_null_path, null_ctl);
4139 result_val->init_req(_null_path, _gvn.intcon(0));
4140 }
4141
4142 // Unconditionally null? Then return right away.
4143 if (stopped()) {
4144 set_control( result_reg->in(_null_path));
4145 if (!stopped())
4146 set_result(result_val->in(_null_path));
4147 return true;
4148 }
4149
4150 // We only go to the fast case code if we pass a number of guards. The
4151 // paths which do not pass are accumulated in the slow_region.
4152 RegionNode* slow_region = new RegionNode(1);
4153 record_for_igvn(slow_region);
4154
4155 // If this is a virtual call, we generate a funny guard. We pull out
4156 // the vtable entry corresponding to hashCode() from the target object.
4157 // If the target method which we are calling happens to be the native
4158 // Object hashCode() method, we pass the guard. We do not need this
4159 // guard for non-virtual calls -- the caller is known to be the native
4160 // Object hashCode().
4161 if (is_virtual) {
4162 // After null check, get the object's klass.
4163 Node* obj_klass = load_object_klass(obj);
4164 generate_virtual_guard(obj_klass, slow_region);
4165 }
4166
4167 // Get the header out of the object, use LoadMarkNode when available
4168 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4169 // The control of the load must be NULL. Otherwise, the load can move before
4170 // the null check after castPP removal.
4171 Node* no_ctrl = NULL;
4172 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4173
4174 // Test the header to see if it is unlocked.
4175 // This also serves as guard against inline types
4176 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4177 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4178 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4179 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4180 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4181
4182 generate_slow_guard(test_unlocked, slow_region);
4183
4184 // Get the hash value and check to see that it has been properly assigned.
4185 // We depend on hash_mask being at most 32 bits and avoid the use of
4186 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4187 // vm: see markWord.hpp.
4188 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4189 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4190 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4191 // This hack lets the hash bits live anywhere in the mark object now, as long
4192 // as the shift drops the relevant bits into the low 32 bits. Note that
4193 // Java spec says that HashCode is an int so there's no point in capturing
4194 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4195 hshifted_header = ConvX2I(hshifted_header);
4196 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4197
4198 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4199 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4200 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4201
4202 generate_slow_guard(test_assigned, slow_region);
4203
4204 Node* init_mem = reset_memory();
4205 // fill in the rest of the null path:
4206 result_io ->init_req(_null_path, i_o());
4207 result_mem->init_req(_null_path, init_mem);
4208
4209 result_val->init_req(_fast_path, hash_val);
4210 result_reg->init_req(_fast_path, control());
4211 result_io ->init_req(_fast_path, i_o());
4212 result_mem->init_req(_fast_path, init_mem);
4213
4214 // Generate code for the slow case. We make a call to hashCode().
4215 set_control(_gvn.transform(slow_region));
4216 if (!stopped()) {
4217 // No need for PreserveJVMState, because we're using up the present state.
4218 set_all_memory(init_mem);
4219 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4220 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4221 Node* slow_result = set_results_for_java_call(slow_call);
4222 // this->control() comes from set_results_for_java_call
4223 result_reg->init_req(_slow_path, control());
4224 result_val->init_req(_slow_path, slow_result);
4225 result_io ->set_req(_slow_path, i_o());
4226 result_mem ->set_req(_slow_path, reset_memory());
4227 }
4228
4229 // Return the combined state.
4230 set_i_o( _gvn.transform(result_io) );
4231 set_all_memory( _gvn.transform(result_mem));
4232
4233 set_result(result_reg, result_val);
4234 return true;
4235 }
4236
4237 //---------------------------inline_native_getClass----------------------------
4238 // public final native Class<?> java.lang.Object.getClass();
4239 //
4240 // Build special case code for calls to getClass on an object.
4241 bool LibraryCallKit::inline_native_getClass() {
4242 Node* obj = argument(0);
4243 if (obj->is_InlineTypeBase()) {
4244 const Type* t = _gvn.type(obj);
4245 if (t->maybe_null()) {
4246 null_check(obj);
4247 }
4248 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4249 return true;
4250 }
4251 obj = null_check_receiver();
4252 if (stopped()) return true;
4253 set_result(load_mirror_from_klass(load_object_klass(obj)));
4254 return true;
4255 }
4256
4257 //-----------------inline_native_Reflection_getCallerClass---------------------
4258 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4259 //
4260 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4261 //
4262 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4263 // in that it must skip particular security frames and checks for
4264 // caller sensitive methods.
4265 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4266 #ifndef PRODUCT
4267 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4268 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4269 }
4270 #endif
4271
4272 if (!jvms()->has_method()) {
4273 #ifndef PRODUCT
4274 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4275 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4276 }
4277 #endif
4278 return false;
4279 }
4280
4281 // Walk back up the JVM state to find the caller at the required
4282 // depth.
4283 JVMState* caller_jvms = jvms();
4284
4285 // Cf. JVM_GetCallerClass
4286 // NOTE: Start the loop at depth 1 because the current JVM state does
4287 // not include the Reflection.getCallerClass() frame.
4288 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4289 ciMethod* m = caller_jvms->method();
4290 switch (n) {
4291 case 0:
4292 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4293 break;
4294 case 1:
4295 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4296 if (!m->caller_sensitive()) {
4297 #ifndef PRODUCT
4298 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4299 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4300 }
4301 #endif
4302 return false; // bail-out; let JVM_GetCallerClass do the work
4303 }
4304 break;
4305 default:
4306 if (!m->is_ignored_by_security_stack_walk()) {
4307 // We have reached the desired frame; return the holder class.
4308 // Acquire method holder as java.lang.Class and push as constant.
4309 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4310 ciInstance* caller_mirror = caller_klass->java_mirror();
4311 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4312
4313 #ifndef PRODUCT
4314 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4315 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());
4316 tty->print_cr(" JVM state at this point:");
4317 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4318 ciMethod* m = jvms()->of_depth(i)->method();
4319 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4320 }
4321 }
4322 #endif
4323 return true;
4324 }
4325 break;
4326 }
4327 }
4328
4329 #ifndef PRODUCT
4330 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4331 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4332 tty->print_cr(" JVM state at this point:");
4333 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4334 ciMethod* m = jvms()->of_depth(i)->method();
4335 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4336 }
4337 }
4338 #endif
4339
4340 return false; // bail-out; let JVM_GetCallerClass do the work
4341 }
4342
4343 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4344 Node* arg = argument(0);
4345 Node* result = NULL;
4346
4347 switch (id) {
4348 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4349 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4350 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4351 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4352
4353 case vmIntrinsics::_doubleToLongBits: {
4354 // two paths (plus control) merge in a wood
4355 RegionNode *r = new RegionNode(3);
4356 Node *phi = new PhiNode(r, TypeLong::LONG);
4357
4358 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4359 // Build the boolean node
4360 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4361
4362 // Branch either way.
4363 // NaN case is less traveled, which makes all the difference.
4364 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4365 Node *opt_isnan = _gvn.transform(ifisnan);
4366 assert( opt_isnan->is_If(), "Expect an IfNode");
4367 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4368 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4369
4370 set_control(iftrue);
4371
4372 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4373 Node *slow_result = longcon(nan_bits); // return NaN
4374 phi->init_req(1, _gvn.transform( slow_result ));
4375 r->init_req(1, iftrue);
4376
4377 // Else fall through
4378 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4379 set_control(iffalse);
4380
4381 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4382 r->init_req(2, iffalse);
4383
4384 // Post merge
4385 set_control(_gvn.transform(r));
4386 record_for_igvn(r);
4387
4388 C->set_has_split_ifs(true); // Has chance for split-if optimization
4389 result = phi;
4390 assert(result->bottom_type()->isa_long(), "must be");
4391 break;
4392 }
4393
4394 case vmIntrinsics::_floatToIntBits: {
4395 // two paths (plus control) merge in a wood
4396 RegionNode *r = new RegionNode(3);
4397 Node *phi = new PhiNode(r, TypeInt::INT);
4398
4399 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4400 // Build the boolean node
4401 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4402
4403 // Branch either way.
4404 // NaN case is less traveled, which makes all the difference.
4405 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4406 Node *opt_isnan = _gvn.transform(ifisnan);
4407 assert( opt_isnan->is_If(), "Expect an IfNode");
4408 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4409 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4410
4411 set_control(iftrue);
4412
4413 static const jint nan_bits = 0x7fc00000;
4414 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4415 phi->init_req(1, _gvn.transform( slow_result ));
4416 r->init_req(1, iftrue);
4417
4418 // Else fall through
4419 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4420 set_control(iffalse);
4421
4422 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4423 r->init_req(2, iffalse);
4424
4425 // Post merge
4426 set_control(_gvn.transform(r));
4427 record_for_igvn(r);
4428
4429 C->set_has_split_ifs(true); // Has chance for split-if optimization
4430 result = phi;
4431 assert(result->bottom_type()->isa_int(), "must be");
4432 break;
4433 }
4434
4435 default:
4436 fatal_unexpected_iid(id);
4437 break;
4438 }
4439 set_result(_gvn.transform(result));
4440 return true;
4441 }
4442
4443 //----------------------inline_unsafe_copyMemory-------------------------
4444 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4445
4446 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4447 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4448 const Type* base_t = gvn.type(base);
4449
4450 bool in_native = (base_t == TypePtr::NULL_PTR);
4451 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4452 bool is_mixed = !in_heap && !in_native;
4453
4454 if (is_mixed) {
4455 return true; // mixed accesses can touch both on-heap and off-heap memory
4456 }
4457 if (in_heap) {
4458 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4459 if (!is_prim_array) {
4460 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4461 // there's not enough type information available to determine proper memory slice for it.
4462 return true;
4463 }
4464 }
4465 return false;
4466 }
4467
4468 bool LibraryCallKit::inline_unsafe_copyMemory() {
4469 if (callee()->is_static()) return false; // caller must have the capability!
4470 null_check_receiver(); // null-check receiver
4471 if (stopped()) return true;
4472
4473 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4474
4475 Node* src_base = argument(1); // type: oop
4476 Node* src_off = ConvL2X(argument(2)); // type: long
4477 Node* dst_base = argument(4); // type: oop
4478 Node* dst_off = ConvL2X(argument(5)); // type: long
4479 Node* size = ConvL2X(argument(7)); // type: long
4480
4481 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4482 "fieldOffset must be byte-scaled");
4483
4484 Node* src_addr = make_unsafe_address(src_base, src_off);
4485 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4486
4487 Node* thread = _gvn.transform(new ThreadLocalNode());
4488 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4489 BasicType doing_unsafe_access_bt = T_BYTE;
4490 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4491
4492 // update volatile field
4493 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4494
4495 int flags = RC_LEAF | RC_NO_FP;
4496
4497 const TypePtr* dst_type = TypePtr::BOTTOM;
4498
4499 // Adjust memory effects of the runtime call based on input values.
4500 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4501 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4502 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4503
4504 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4505 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4506 flags |= RC_NARROW_MEM; // narrow in memory
4507 }
4508 }
4509
4510 // Call it. Note that the length argument is not scaled.
4511 make_runtime_call(flags,
4512 OptoRuntime::fast_arraycopy_Type(),
4513 StubRoutines::unsafe_arraycopy(),
4514 "unsafe_arraycopy",
4515 dst_type,
4516 src_addr, dst_addr, size XTOP);
4517
4518 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4519
4520 return true;
4521 }
4522
4523 #undef XTOP
4524
4525 //------------------------clone_coping-----------------------------------
4526 // Helper function for inline_native_clone.
4527 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4528 assert(obj_size != NULL, "");
4529 Node* raw_obj = alloc_obj->in(1);
4530 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4531
4532 AllocateNode* alloc = NULL;
4533 if (ReduceBulkZeroing) {
4534 // We will be completely responsible for initializing this object -
4535 // mark Initialize node as complete.
4536 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4537 // The object was just allocated - there should be no any stores!
4538 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4539 // Mark as complete_with_arraycopy so that on AllocateNode
4540 // expansion, we know this AllocateNode is initialized by an array
4541 // copy and a StoreStore barrier exists after the array copy.
4542 alloc->initialization()->set_complete_with_arraycopy();
4543 }
4544
4545 Node* size = _gvn.transform(obj_size);
4546 access_clone(obj, alloc_obj, size, is_array);
4547
4548 // Do not let reads from the cloned object float above the arraycopy.
4549 if (alloc != NULL) {
4550 // Do not let stores that initialize this object be reordered with
4551 // a subsequent store that would make this object accessible by
4552 // other threads.
4553 // Record what AllocateNode this StoreStore protects so that
4554 // escape analysis can go from the MemBarStoreStoreNode to the
4555 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4556 // based on the escape status of the AllocateNode.
4557 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4558 } else {
4559 insert_mem_bar(Op_MemBarCPUOrder);
4560 }
4561 }
4562
4563 //------------------------inline_native_clone----------------------------
4564 // protected native Object java.lang.Object.clone();
4565 //
4566 // Here are the simple edge cases:
4567 // null receiver => normal trap
4568 // virtual and clone was overridden => slow path to out-of-line clone
4569 // not cloneable or finalizer => slow path to out-of-line Object.clone
4570 //
4571 // The general case has two steps, allocation and copying.
4572 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4573 //
4574 // Copying also has two cases, oop arrays and everything else.
4575 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4576 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4577 //
4578 // These steps fold up nicely if and when the cloned object's klass
4579 // can be sharply typed as an object array, a type array, or an instance.
4580 //
4581 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4582 PhiNode* result_val;
4583
4584 // Set the reexecute bit for the interpreter to reexecute
4585 // the bytecode that invokes Object.clone if deoptimization happens.
4586 { PreserveReexecuteState preexecs(this);
4587 jvms()->set_should_reexecute(true);
4588
4589 Node* obj = argument(0);
4590 if (obj->is_InlineType()) {
4591 return false;
4592 }
4593
4594 obj = null_check_receiver();
4595 if (stopped()) return true;
4596
4597 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4598
4599 // If we are going to clone an instance, we need its exact type to
4600 // know the number and types of fields to convert the clone to
4601 // loads/stores. Maybe a speculative type can help us.
4602 if (!obj_type->klass_is_exact() &&
4603 obj_type->speculative_type() != NULL &&
4604 obj_type->speculative_type()->is_instance_klass() &&
4605 !obj_type->speculative_type()->is_inlinetype()) {
4606 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4607 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4608 !spec_ik->has_injected_fields()) {
4609 ciKlass* k = obj_type->klass();
4610 if (!k->is_instance_klass() ||
4611 k->as_instance_klass()->is_interface() ||
4612 k->as_instance_klass()->has_subklass()) {
4613 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4614 }
4615 }
4616 }
4617
4618 // Conservatively insert a memory barrier on all memory slices.
4619 // Do not let writes into the original float below the clone.
4620 insert_mem_bar(Op_MemBarCPUOrder);
4621
4622 // paths into result_reg:
4623 enum {
4624 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4625 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4626 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4627 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4628 PATH_LIMIT
4629 };
4630 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4631 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4632 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4633 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4634 record_for_igvn(result_reg);
4635
4636 Node* obj_klass = load_object_klass(obj);
4637 // We only go to the fast case code if we pass a number of guards.
4638 // The paths which do not pass are accumulated in the slow_region.
4639 RegionNode* slow_region = new RegionNode(1);
4640 record_for_igvn(slow_region);
4641
4642 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4643 if (array_ctl != NULL) {
4644 // It's an array.
4645 PreserveJVMState pjvms(this);
4646 set_control(array_ctl);
4647
4648 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4649 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
4650 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
4651 obj_type->klass()->can_be_inline_array_klass() &&
4652 (ary_ptr == NULL || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
4653 // Flattened inline type array may have object field that would require a
4654 // write barrier. Conservatively, go to slow path.
4655 generate_fair_guard(flat_array_test(obj_klass), slow_region);
4656 }
4657
4658 if (!stopped()) {
4659 Node* obj_length = load_array_length(obj);
4660 Node* obj_size = NULL;
4661 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4662
4663 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4664 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4665 // If it is an oop array, it requires very special treatment,
4666 // because gc barriers are required when accessing the array.
4667 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4668 if (is_obja != NULL) {
4669 PreserveJVMState pjvms2(this);
4670 set_control(is_obja);
4671 // Generate a direct call to the right arraycopy function(s).
4672 // Clones are always tightly coupled.
4673 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4674 ac->set_clone_oop_array();
4675 Node* n = _gvn.transform(ac);
4676 assert(n == ac, "cannot disappear");
4677 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4678
4679 result_reg->init_req(_objArray_path, control());
4680 result_val->init_req(_objArray_path, alloc_obj);
4681 result_i_o ->set_req(_objArray_path, i_o());
4682 result_mem ->set_req(_objArray_path, reset_memory());
4683 }
4684 }
4685 // Otherwise, there are no barriers to worry about.
4686 // (We can dispense with card marks if we know the allocation
4687 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4688 // causes the non-eden paths to take compensating steps to
4689 // simulate a fresh allocation, so that no further
4690 // card marks are required in compiled code to initialize
4691 // the object.)
4692
4693 if (!stopped()) {
4694 copy_to_clone(obj, alloc_obj, obj_size, true);
4695
4696 // Present the results of the copy.
4697 result_reg->init_req(_array_path, control());
4698 result_val->init_req(_array_path, alloc_obj);
4699 result_i_o ->set_req(_array_path, i_o());
4700 result_mem ->set_req(_array_path, reset_memory());
4701 }
4702 }
4703 }
4704
4705 if (!stopped()) {
4706 // It's an instance (we did array above). Make the slow-path tests.
4707 // If this is a virtual call, we generate a funny guard. We grab
4708 // the vtable entry corresponding to clone() from the target object.
4709 // If the target method which we are calling happens to be the
4710 // Object clone() method, we pass the guard. We do not need this
4711 // guard for non-virtual calls; the caller is known to be the native
4712 // Object clone().
4713 if (is_virtual) {
4714 generate_virtual_guard(obj_klass, slow_region);
4715 }
4716
4717 // The object must be easily cloneable and must not have a finalizer.
4718 // Both of these conditions may be checked in a single test.
4719 // We could optimize the test further, but we don't care.
4720 generate_access_flags_guard(obj_klass,
4721 // Test both conditions:
4722 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4723 // Must be cloneable but not finalizer:
4724 JVM_ACC_IS_CLONEABLE_FAST,
4725 slow_region);
4726 }
4727
4728 if (!stopped()) {
4729 // It's an instance, and it passed the slow-path tests.
4730 PreserveJVMState pjvms(this);
4731 Node* obj_size = NULL;
4732 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4733 // is reexecuted if deoptimization occurs and there could be problems when merging
4734 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4735 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4736
4737 copy_to_clone(obj, alloc_obj, obj_size, false);
4738
4739 // Present the results of the slow call.
4740 result_reg->init_req(_instance_path, control());
4741 result_val->init_req(_instance_path, alloc_obj);
4742 result_i_o ->set_req(_instance_path, i_o());
4743 result_mem ->set_req(_instance_path, reset_memory());
4744 }
4745
4746 // Generate code for the slow case. We make a call to clone().
4747 set_control(_gvn.transform(slow_region));
4748 if (!stopped()) {
4749 PreserveJVMState pjvms(this);
4750 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4751 // We need to deoptimize on exception (see comment above)
4752 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4753 // this->control() comes from set_results_for_java_call
4754 result_reg->init_req(_slow_path, control());
4755 result_val->init_req(_slow_path, slow_result);
4756 result_i_o ->set_req(_slow_path, i_o());
4757 result_mem ->set_req(_slow_path, reset_memory());
4758 }
4759
4760 // Return the combined state.
4761 set_control( _gvn.transform(result_reg));
4762 set_i_o( _gvn.transform(result_i_o));
4763 set_all_memory( _gvn.transform(result_mem));
4764 } // original reexecute is set back here
4765
4766 set_result(_gvn.transform(result_val));
4767 return true;
4768 }
4769
4770 // If we have a tightly coupled allocation, the arraycopy may take care
4771 // of the array initialization. If one of the guards we insert between
4772 // the allocation and the arraycopy causes a deoptimization, an
4773 // unitialized array will escape the compiled method. To prevent that
4774 // we set the JVM state for uncommon traps between the allocation and
4775 // the arraycopy to the state before the allocation so, in case of
4776 // deoptimization, we'll reexecute the allocation and the
4777 // initialization.
4778 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4779 if (alloc != NULL) {
4780 ciMethod* trap_method = alloc->jvms()->method();
4781 int trap_bci = alloc->jvms()->bci();
4782
4783 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4784 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4785 // Make sure there's no store between the allocation and the
4786 // arraycopy otherwise visible side effects could be rexecuted
4787 // in case of deoptimization and cause incorrect execution.
4788 bool no_interfering_store = true;
4789 Node* mem = alloc->in(TypeFunc::Memory);
4790 if (mem->is_MergeMem()) {
4791 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4792 Node* n = mms.memory();
4793 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4794 assert(n->is_Store(), "what else?");
4795 no_interfering_store = false;
4796 break;
4797 }
4798 }
4799 } else {
4800 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4801 Node* n = mms.memory();
4802 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4803 assert(n->is_Store(), "what else?");
4804 no_interfering_store = false;
4805 break;
4806 }
4807 }
4808 }
4809
4810 if (no_interfering_store) {
4811 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4812 uint size = alloc->req();
4813 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4814 old_jvms->set_map(sfpt);
4815 for (uint i = 0; i < size; i++) {
4816 sfpt->init_req(i, alloc->in(i));
4817 }
4818 // re-push array length for deoptimization
4819 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4820 old_jvms->set_sp(old_jvms->sp()+1);
4821 old_jvms->set_monoff(old_jvms->monoff()+1);
4822 old_jvms->set_scloff(old_jvms->scloff()+1);
4823 old_jvms->set_endoff(old_jvms->endoff()+1);
4824 old_jvms->set_should_reexecute(true);
4825
4826 sfpt->set_i_o(map()->i_o());
4827 sfpt->set_memory(map()->memory());
4828 sfpt->set_control(map()->control());
4829
4830 JVMState* saved_jvms = jvms();
4831 saved_reexecute_sp = _reexecute_sp;
4832
4833 set_jvms(sfpt->jvms());
4834 _reexecute_sp = jvms()->sp();
4835
4836 return saved_jvms;
4837 }
4838 }
4839 }
4840 return NULL;
4841 }
4842
4843 // In case of a deoptimization, we restart execution at the
4844 // allocation, allocating a new array. We would leave an uninitialized
4845 // array in the heap that GCs wouldn't expect. Move the allocation
4846 // after the traps so we don't allocate the array if we
4847 // deoptimize. This is possible because tightly_coupled_allocation()
4848 // guarantees there's no observer of the allocated array at this point
4849 // and the control flow is simple enough.
4850 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4851 int saved_reexecute_sp, uint new_idx) {
4852 if (saved_jvms != NULL && !stopped()) {
4853 assert(alloc != NULL, "only with a tightly coupled allocation");
4854 // restore JVM state to the state at the arraycopy
4855 saved_jvms->map()->set_control(map()->control());
4856 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4857 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4858 // If we've improved the types of some nodes (null check) while
4859 // emitting the guards, propagate them to the current state
4860 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4861 set_jvms(saved_jvms);
4862 _reexecute_sp = saved_reexecute_sp;
4863
4864 // Remove the allocation from above the guards
4865 CallProjections* callprojs = alloc->extract_projections(true);
4866 InitializeNode* init = alloc->initialization();
4867 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4868 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4869 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4870
4871 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4872 // the allocation (i.e. is only valid if the allocation succeeds):
4873 // 1) replace CastIINode with AllocateArrayNode's length here
4874 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4875 //
4876 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4877 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4878 Node* init_control = init->proj_out(TypeFunc::Control);
4879 Node* alloc_length = alloc->Ideal_length();
4880 #ifdef ASSERT
4881 Node* prev_cast = NULL;
4882 #endif
4883 for (uint i = 0; i < init_control->outcnt(); i++) {
4884 Node* init_out = init_control->raw_out(i);
4885 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4886 #ifdef ASSERT
4887 if (prev_cast == NULL) {
4888 prev_cast = init_out;
4889 } else {
4890 if (prev_cast->cmp(*init_out) == false) {
4891 prev_cast->dump();
4892 init_out->dump();
4893 assert(false, "not equal CastIINode");
4894 }
4895 }
4896 #endif
4897 C->gvn_replace_by(init_out, alloc_length);
4898 }
4899 }
4900 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4901
4902 // move the allocation here (after the guards)
4903 _gvn.hash_delete(alloc);
4904 alloc->set_req(TypeFunc::Control, control());
4905 alloc->set_req(TypeFunc::I_O, i_o());
4906 Node *mem = reset_memory();
4907 set_all_memory(mem);
4908 alloc->set_req(TypeFunc::Memory, mem);
4909 set_control(init->proj_out_or_null(TypeFunc::Control));
4910 set_i_o(callprojs->fallthrough_ioproj);
4911
4912 // Update memory as done in GraphKit::set_output_for_allocation()
4913 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4914 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4915 if (ary_type->isa_aryptr() && length_type != NULL) {
4916 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4917 }
4918 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4919 int elemidx = C->get_alias_index(telemref);
4920 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4921 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4922
4923 Node* allocx = _gvn.transform(alloc);
4924 assert(allocx == alloc, "where has the allocation gone?");
4925 assert(dest->is_CheckCastPP(), "not an allocation result?");
4926
4927 _gvn.hash_delete(dest);
4928 dest->set_req(0, control());
4929 Node* destx = _gvn.transform(dest);
4930 assert(destx == dest, "where has the allocation result gone?");
4931
4932 array_ideal_length(alloc, ary_type, true);
4933 }
4934 }
4935
4936
4937 //------------------------------inline_arraycopy-----------------------
4938 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4939 // Object dest, int destPos,
4940 // int length);
4941 bool LibraryCallKit::inline_arraycopy() {
4942 // Get the arguments.
4943 Node* src = argument(0); // type: oop
4944 Node* src_offset = argument(1); // type: int
4945 Node* dest = argument(2); // type: oop
4946 Node* dest_offset = argument(3); // type: int
4947 Node* length = argument(4); // type: int
4948
4949 uint new_idx = C->unique();
4950
4951 // Check for allocation before we add nodes that would confuse
4952 // tightly_coupled_allocation()
4953 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
4954
4955 int saved_reexecute_sp = -1;
4956 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4957 // See arraycopy_restore_alloc_state() comment
4958 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4959 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4960 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4961 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4962
4963 // The following tests must be performed
4964 // (1) src and dest are arrays.
4965 // (2) src and dest arrays must have elements of the same BasicType
4966 // (3) src and dest must not be null.
4967 // (4) src_offset must not be negative.
4968 // (5) dest_offset must not be negative.
4969 // (6) length must not be negative.
4970 // (7) src_offset + length must not exceed length of src.
4971 // (8) dest_offset + length must not exceed length of dest.
4972 // (9) each element of an oop array must be assignable
4973
4974 // (3) src and dest must not be null.
4975 // always do this here because we need the JVM state for uncommon traps
4976 Node* null_ctl = top();
4977 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4978 assert(null_ctl->is_top(), "no null control here");
4979 dest = null_check(dest, T_ARRAY);
4980
4981 if (!can_emit_guards) {
4982 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4983 // guards but the arraycopy node could still take advantage of a
4984 // tightly allocated allocation. tightly_coupled_allocation() is
4985 // called again to make sure it takes the null check above into
4986 // account: the null check is mandatory and if it caused an
4987 // uncommon trap to be emitted then the allocation can't be
4988 // considered tightly coupled in this context.
4989 alloc = tightly_coupled_allocation(dest);
4990 }
4991
4992 bool validated = false;
4993
4994 const Type* src_type = _gvn.type(src);
4995 const Type* dest_type = _gvn.type(dest);
4996 const TypeAryPtr* top_src = src_type->isa_aryptr();
4997 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4998
4999 // Do we have the type of src?
5000 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5001 // Do we have the type of dest?
5002 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5003 // Is the type for src from speculation?
5004 bool src_spec = false;
5005 // Is the type for dest from speculation?
5006 bool dest_spec = false;
5007
5008 if ((!has_src || !has_dest) && can_emit_guards) {
5009 // We don't have sufficient type information, let's see if
5010 // speculative types can help. We need to have types for both src
5011 // and dest so that it pays off.
5012
5013 // Do we already have or could we have type information for src
5014 bool could_have_src = has_src;
5015 // Do we already have or could we have type information for dest
5016 bool could_have_dest = has_dest;
5017
5018 ciKlass* src_k = NULL;
5019 if (!has_src) {
5020 src_k = src_type->speculative_type_not_null();
5021 if (src_k != NULL && src_k->is_array_klass()) {
5022 could_have_src = true;
5023 }
5024 }
5025
5026 ciKlass* dest_k = NULL;
5027 if (!has_dest) {
5028 dest_k = dest_type->speculative_type_not_null();
5029 if (dest_k != NULL && dest_k->is_array_klass()) {
5030 could_have_dest = true;
5031 }
5032 }
5033
5034 if (could_have_src && could_have_dest) {
5035 // This is going to pay off so emit the required guards
5036 if (!has_src) {
5037 src = maybe_cast_profiled_obj(src, src_k, true);
5038 src_type = _gvn.type(src);
5039 top_src = src_type->isa_aryptr();
5040 has_src = (top_src != NULL && top_src->klass() != NULL);
5041 src_spec = true;
5042 }
5043 if (!has_dest) {
5044 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5045 dest_type = _gvn.type(dest);
5046 top_dest = dest_type->isa_aryptr();
5047 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5048 dest_spec = true;
5049 }
5050 }
5051 }
5052
5053 if (has_src && has_dest && can_emit_guards) {
5054 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5055 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5056 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
5057 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
5058
5059 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5060 // If both arrays are object arrays then having the exact types
5061 // for both will remove the need for a subtype check at runtime
5062 // before the call and may make it possible to pick a faster copy
5063 // routine (without a subtype check on every element)
5064 // Do we have the exact type of src?
5065 bool could_have_src = src_spec;
5066 // Do we have the exact type of dest?
5067 bool could_have_dest = dest_spec;
5068 ciKlass* src_k = top_src->klass();
5069 ciKlass* dest_k = top_dest->klass();
5070 if (!src_spec) {
5071 src_k = src_type->speculative_type_not_null();
5072 if (src_k != NULL && src_k->is_array_klass()) {
5073 could_have_src = true;
5074 }
5075 }
5076 if (!dest_spec) {
5077 dest_k = dest_type->speculative_type_not_null();
5078 if (dest_k != NULL && dest_k->is_array_klass()) {
5079 could_have_dest = true;
5080 }
5081 }
5082 if (could_have_src && could_have_dest) {
5083 // If we can have both exact types, emit the missing guards
5084 if (could_have_src && !src_spec) {
5085 src = maybe_cast_profiled_obj(src, src_k, true);
5086 src_type = _gvn.type(src);
5087 top_src = src_type->isa_aryptr();
5088 }
5089 if (could_have_dest && !dest_spec) {
5090 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5091 dest_type = _gvn.type(dest);
5092 top_dest = dest_type->isa_aryptr();
5093 }
5094 }
5095 }
5096 }
5097
5098 ciMethod* trap_method = method();
5099 int trap_bci = bci();
5100 if (saved_jvms != NULL) {
5101 trap_method = alloc->jvms()->method();
5102 trap_bci = alloc->jvms()->bci();
5103 }
5104
5105 bool negative_length_guard_generated = false;
5106
5107 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5108 can_emit_guards && !src->is_top() && !dest->is_top()) {
5109 // validate arguments: enables transformation the ArrayCopyNode
5110 validated = true;
5111
5112 RegionNode* slow_region = new RegionNode(1);
5113 record_for_igvn(slow_region);
5114
5115 // (1) src and dest are arrays.
5116 generate_non_array_guard(load_object_klass(src), slow_region);
5117 generate_non_array_guard(load_object_klass(dest), slow_region);
5118
5119 // (2) src and dest arrays must have elements of the same BasicType
5120 // done at macro expansion or at Ideal transformation time
5121
5122 // (4) src_offset must not be negative.
5123 generate_negative_guard(src_offset, slow_region);
5124
5125 // (5) dest_offset must not be negative.
5126 generate_negative_guard(dest_offset, slow_region);
5127
5128 // (7) src_offset + length must not exceed length of src.
5129 generate_limit_guard(src_offset, length,
5130 load_array_length(src),
5131 slow_region);
5132
5133 // (8) dest_offset + length must not exceed length of dest.
5134 generate_limit_guard(dest_offset, length,
5135 load_array_length(dest),
5136 slow_region);
5137
5138 // (6) length must not be negative.
5139 // This is also checked in generate_arraycopy() during macro expansion, but
5140 // we also have to check it here for the case where the ArrayCopyNode will
5141 // be eliminated by Escape Analysis.
5142 if (EliminateAllocations) {
5143 generate_negative_guard(length, slow_region);
5144 negative_length_guard_generated = true;
5145 }
5146
5147 // (9) each element of an oop array must be assignable
5148 Node* dest_klass = load_object_klass(dest);
5149 if (src != dest) {
5150 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5151 slow_region->add_req(not_subtype_ctrl);
5152 }
5153
5154 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5155 const Type* toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
5156 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5157 src_type = _gvn.type(src);
5158 top_src = src_type->isa_aryptr();
5159
5160 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
5161 if (!stopped() && UseFlatArray) {
5162 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
5163 assert(top_dest == NULL || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
5164 if (top_src != NULL && top_src->is_flat()) {
5165 // Src is flat, check that dest is flat as well
5166 if (top_dest != NULL && !top_dest->is_flat()) {
5167 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
5168 // Since dest is flat and src <: dest, dest must have the same type as src.
5169 top_dest = TypeOopPtr::make_from_klass(top_src->klass())->isa_aryptr();
5170 assert(top_dest->is_flat(), "dest must be flat");
5171 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
5172 }
5173 } else if (top_src == NULL || !top_src->is_not_flat()) {
5174 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5175 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5176 assert(top_dest == NULL || !top_dest->is_flat(), "dest array must not be flat");
5177 generate_fair_guard(flat_array_test(src), slow_region);
5178 if (top_src != NULL) {
5179 top_src = top_src->cast_to_not_flat();
5180 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
5181 }
5182 }
5183 }
5184
5185 {
5186 PreserveJVMState pjvms(this);
5187 set_control(_gvn.transform(slow_region));
5188 uncommon_trap(Deoptimization::Reason_intrinsic,
5189 Deoptimization::Action_make_not_entrant);
5190 assert(stopped(), "Should be stopped");
5191 }
5192 }
5193
5194 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5195
5196 if (stopped()) {
5197 return true;
5198 }
5199
5200 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5201 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5202 // so the compiler has a chance to eliminate them: during macro expansion,
5203 // we have to set their control (CastPP nodes are eliminated).
5204 load_object_klass(src), load_object_klass(dest),
5205 load_array_length(src), load_array_length(dest));
5206
5207 ac->set_arraycopy(validated);
5208
5209 Node* n = _gvn.transform(ac);
5210 if (n == ac) {
5211 ac->connect_outputs(this);
5212 } else {
5213 assert(validated, "shouldn't transform if all arguments not validated");
5214 set_all_memory(n);
5215 }
5216 clear_upper_avx();
5217
5218
5219 return true;
5220 }
5221
5222
5223 // Helper function which determines if an arraycopy immediately follows
5224 // an allocation, with no intervening tests or other escapes for the object.
5225 AllocateArrayNode*
5226 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5227 if (stopped()) return NULL; // no fast path
5228 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5229
5230 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5231 if (alloc == NULL) return NULL;
5232
5233 Node* rawmem = memory(Compile::AliasIdxRaw);
5234 // Is the allocation's memory state untouched?
5235 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5236 // Bail out if there have been raw-memory effects since the allocation.
5237 // (Example: There might have been a call or safepoint.)
5238 return NULL;
5239 }
5240 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5241 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5242 return NULL;
5243 }
5244
5245 // There must be no unexpected observers of this allocation.
5246 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5247 Node* obs = ptr->fast_out(i);
5248 if (obs != this->map()) {
5249 return NULL;
5250 }
5251 }
5252
5253 // This arraycopy must unconditionally follow the allocation of the ptr.
5254 Node* alloc_ctl = ptr->in(0);
5255 Node* ctl = control();
5256 while (ctl != alloc_ctl) {
5257 // There may be guards which feed into the slow_region.
5258 // Any other control flow means that we might not get a chance
5259 // to finish initializing the allocated object.
5260 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5261 IfNode* iff = ctl->in(0)->as_If();
5262 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5263 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5264 // One more try: Various low-level checks bottom out in
5265 // uncommon traps. If the debug-info of the trap omits
5266 // any reference to the allocation, as we've already
5267 // observed, then there can be no objection to the trap.
5268 bool found_trap = false;
5269 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5270 Node* obs = not_ctl->fast_out(j);
5271 if (obs->in(0) == not_ctl && obs->is_Call() &&
5272 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5273 found_trap = true; break;
5274 }
5275 }
5276 if (found_trap) {
5277 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5278 continue;
5279 }
5280 }
5281 return NULL;
5282 }
5283
5284 // If we get this far, we have an allocation which immediately
5285 // precedes the arraycopy, and we can take over zeroing the new object.
5286 // The arraycopy will finish the initialization, and provide
5287 // a new control state to which we will anchor the destination pointer.
5288
5289 return alloc;
5290 }
5291
5292 //-------------inline_encodeISOArray-----------------------------------
5293 // encode char[] to byte[] in ISO_8859_1 or ASCII
5294 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5295 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5296 // no receiver since it is static method
5297 Node *src = argument(0);
5298 Node *src_offset = argument(1);
5299 Node *dst = argument(2);
5300 Node *dst_offset = argument(3);
5301 Node *length = argument(4);
5302
5303 src = must_be_not_null(src, true);
5304 dst = must_be_not_null(dst, true);
5305
5306 const Type* src_type = src->Value(&_gvn);
5307 const Type* dst_type = dst->Value(&_gvn);
5308 const TypeAryPtr* top_src = src_type->isa_aryptr();
5309 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5310 if (top_src == NULL || top_src->klass() == NULL ||
5311 top_dest == NULL || top_dest->klass() == NULL) {
5312 // failed array check
5313 return false;
5314 }
5315
5316 // Figure out the size and type of the elements we will be copying.
5317 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5318 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5319 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5320 return false;
5321 }
5322
5323 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5324 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5325 // 'src_start' points to src array + scaled offset
5326 // 'dst_start' points to dst array + scaled offset
5327
5328 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5329 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5330 enc = _gvn.transform(enc);
5331 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5332 set_memory(res_mem, mtype);
5333 set_result(enc);
5334 clear_upper_avx();
5335
5336 return true;
5337 }
5338
5339 //-------------inline_multiplyToLen-----------------------------------
5340 bool LibraryCallKit::inline_multiplyToLen() {
5341 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5342
5343 address stubAddr = StubRoutines::multiplyToLen();
5344 if (stubAddr == NULL) {
5345 return false; // Intrinsic's stub is not implemented on this platform
5346 }
5347 const char* stubName = "multiplyToLen";
5348
5349 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5350
5351 // no receiver because it is a static method
5352 Node* x = argument(0);
5353 Node* xlen = argument(1);
5354 Node* y = argument(2);
5355 Node* ylen = argument(3);
5356 Node* z = argument(4);
5357
5358 x = must_be_not_null(x, true);
5359 y = must_be_not_null(y, true);
5360
5361 const Type* x_type = x->Value(&_gvn);
5362 const Type* y_type = y->Value(&_gvn);
5363 const TypeAryPtr* top_x = x_type->isa_aryptr();
5364 const TypeAryPtr* top_y = y_type->isa_aryptr();
5365 if (top_x == NULL || top_x->klass() == NULL ||
5366 top_y == NULL || top_y->klass() == NULL) {
5367 // failed array check
5368 return false;
5369 }
5370
5371 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5372 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5373 if (x_elem != T_INT || y_elem != T_INT) {
5374 return false;
5375 }
5376
5377 // Set the original stack and the reexecute bit for the interpreter to reexecute
5378 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5379 // on the return from z array allocation in runtime.
5380 { PreserveReexecuteState preexecs(this);
5381 jvms()->set_should_reexecute(true);
5382
5383 Node* x_start = array_element_address(x, intcon(0), x_elem);
5384 Node* y_start = array_element_address(y, intcon(0), y_elem);
5385 // 'x_start' points to x array + scaled xlen
5386 // 'y_start' points to y array + scaled ylen
5387
5388 // Allocate the result array
5389 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5390 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5391 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5392
5393 IdealKit ideal(this);
5394
5395 #define __ ideal.
5396 Node* one = __ ConI(1);
5397 Node* zero = __ ConI(0);
5398 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5399 __ set(need_alloc, zero);
5400 __ set(z_alloc, z);
5401 __ if_then(z, BoolTest::eq, null()); {
5402 __ increment (need_alloc, one);
5403 } __ else_(); {
5404 // Update graphKit memory and control from IdealKit.
5405 sync_kit(ideal);
5406 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5407 cast->init_req(0, control());
5408 _gvn.set_type(cast, cast->bottom_type());
5409 C->record_for_igvn(cast);
5410
5411 Node* zlen_arg = load_array_length(cast);
5412 // Update IdealKit memory and control from graphKit.
5413 __ sync_kit(this);
5414 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5415 __ increment (need_alloc, one);
5416 } __ end_if();
5417 } __ end_if();
5418
5419 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5420 // Update graphKit memory and control from IdealKit.
5421 sync_kit(ideal);
5422 Node * narr = new_array(klass_node, zlen, 1);
5423 // Update IdealKit memory and control from graphKit.
5424 __ sync_kit(this);
5425 __ set(z_alloc, narr);
5426 } __ end_if();
5427
5428 sync_kit(ideal);
5429 z = __ value(z_alloc);
5430 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5431 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5432 // Final sync IdealKit and GraphKit.
5433 final_sync(ideal);
5434 #undef __
5435
5436 Node* z_start = array_element_address(z, intcon(0), T_INT);
5437
5438 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5439 OptoRuntime::multiplyToLen_Type(),
5440 stubAddr, stubName, TypePtr::BOTTOM,
5441 x_start, xlen, y_start, ylen, z_start, zlen);
5442 } // original reexecute is set back here
5443
5444 C->set_has_split_ifs(true); // Has chance for split-if optimization
5445 set_result(z);
5446 return true;
5447 }
5448
5449 //-------------inline_squareToLen------------------------------------
5450 bool LibraryCallKit::inline_squareToLen() {
5451 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5452
5453 address stubAddr = StubRoutines::squareToLen();
5454 if (stubAddr == NULL) {
5455 return false; // Intrinsic's stub is not implemented on this platform
5456 }
5457 const char* stubName = "squareToLen";
5458
5459 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5460
5461 Node* x = argument(0);
5462 Node* len = argument(1);
5463 Node* z = argument(2);
5464 Node* zlen = argument(3);
5465
5466 x = must_be_not_null(x, true);
5467 z = must_be_not_null(z, true);
5468
5469 const Type* x_type = x->Value(&_gvn);
5470 const Type* z_type = z->Value(&_gvn);
5471 const TypeAryPtr* top_x = x_type->isa_aryptr();
5472 const TypeAryPtr* top_z = z_type->isa_aryptr();
5473 if (top_x == NULL || top_x->klass() == NULL ||
5474 top_z == NULL || top_z->klass() == NULL) {
5475 // failed array check
5476 return false;
5477 }
5478
5479 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5480 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5481 if (x_elem != T_INT || z_elem != T_INT) {
5482 return false;
5483 }
5484
5485
5486 Node* x_start = array_element_address(x, intcon(0), x_elem);
5487 Node* z_start = array_element_address(z, intcon(0), z_elem);
5488
5489 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5490 OptoRuntime::squareToLen_Type(),
5491 stubAddr, stubName, TypePtr::BOTTOM,
5492 x_start, len, z_start, zlen);
5493
5494 set_result(z);
5495 return true;
5496 }
5497
5498 //-------------inline_mulAdd------------------------------------------
5499 bool LibraryCallKit::inline_mulAdd() {
5500 assert(UseMulAddIntrinsic, "not implemented on this platform");
5501
5502 address stubAddr = StubRoutines::mulAdd();
5503 if (stubAddr == NULL) {
5504 return false; // Intrinsic's stub is not implemented on this platform
5505 }
5506 const char* stubName = "mulAdd";
5507
5508 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5509
5510 Node* out = argument(0);
5511 Node* in = argument(1);
5512 Node* offset = argument(2);
5513 Node* len = argument(3);
5514 Node* k = argument(4);
5515
5516 out = must_be_not_null(out, true);
5517
5518 const Type* out_type = out->Value(&_gvn);
5519 const Type* in_type = in->Value(&_gvn);
5520 const TypeAryPtr* top_out = out_type->isa_aryptr();
5521 const TypeAryPtr* top_in = in_type->isa_aryptr();
5522 if (top_out == NULL || top_out->klass() == NULL ||
5523 top_in == NULL || top_in->klass() == NULL) {
5524 // failed array check
5525 return false;
5526 }
5527
5528 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5529 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5530 if (out_elem != T_INT || in_elem != T_INT) {
5531 return false;
5532 }
5533
5534 Node* outlen = load_array_length(out);
5535 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5536 Node* out_start = array_element_address(out, intcon(0), out_elem);
5537 Node* in_start = array_element_address(in, intcon(0), in_elem);
5538
5539 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5540 OptoRuntime::mulAdd_Type(),
5541 stubAddr, stubName, TypePtr::BOTTOM,
5542 out_start,in_start, new_offset, len, k);
5543 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5544 set_result(result);
5545 return true;
5546 }
5547
5548 //-------------inline_montgomeryMultiply-----------------------------------
5549 bool LibraryCallKit::inline_montgomeryMultiply() {
5550 address stubAddr = StubRoutines::montgomeryMultiply();
5551 if (stubAddr == NULL) {
5552 return false; // Intrinsic's stub is not implemented on this platform
5553 }
5554
5555 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5556 const char* stubName = "montgomery_multiply";
5557
5558 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5559
5560 Node* a = argument(0);
5561 Node* b = argument(1);
5562 Node* n = argument(2);
5563 Node* len = argument(3);
5564 Node* inv = argument(4);
5565 Node* m = argument(6);
5566
5567 const Type* a_type = a->Value(&_gvn);
5568 const TypeAryPtr* top_a = a_type->isa_aryptr();
5569 const Type* b_type = b->Value(&_gvn);
5570 const TypeAryPtr* top_b = b_type->isa_aryptr();
5571 const Type* n_type = a->Value(&_gvn);
5572 const TypeAryPtr* top_n = n_type->isa_aryptr();
5573 const Type* m_type = a->Value(&_gvn);
5574 const TypeAryPtr* top_m = m_type->isa_aryptr();
5575 if (top_a == NULL || top_a->klass() == NULL ||
5576 top_b == NULL || top_b->klass() == NULL ||
5577 top_n == NULL || top_n->klass() == NULL ||
5578 top_m == NULL || top_m->klass() == NULL) {
5579 // failed array check
5580 return false;
5581 }
5582
5583 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5584 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5585 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5586 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5587 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5588 return false;
5589 }
5590
5591 // Make the call
5592 {
5593 Node* a_start = array_element_address(a, intcon(0), a_elem);
5594 Node* b_start = array_element_address(b, intcon(0), b_elem);
5595 Node* n_start = array_element_address(n, intcon(0), n_elem);
5596 Node* m_start = array_element_address(m, intcon(0), m_elem);
5597
5598 Node* call = make_runtime_call(RC_LEAF,
5599 OptoRuntime::montgomeryMultiply_Type(),
5600 stubAddr, stubName, TypePtr::BOTTOM,
5601 a_start, b_start, n_start, len, inv, top(),
5602 m_start);
5603 set_result(m);
5604 }
5605
5606 return true;
5607 }
5608
5609 bool LibraryCallKit::inline_montgomerySquare() {
5610 address stubAddr = StubRoutines::montgomerySquare();
5611 if (stubAddr == NULL) {
5612 return false; // Intrinsic's stub is not implemented on this platform
5613 }
5614
5615 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5616 const char* stubName = "montgomery_square";
5617
5618 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5619
5620 Node* a = argument(0);
5621 Node* n = argument(1);
5622 Node* len = argument(2);
5623 Node* inv = argument(3);
5624 Node* m = argument(5);
5625
5626 const Type* a_type = a->Value(&_gvn);
5627 const TypeAryPtr* top_a = a_type->isa_aryptr();
5628 const Type* n_type = a->Value(&_gvn);
5629 const TypeAryPtr* top_n = n_type->isa_aryptr();
5630 const Type* m_type = a->Value(&_gvn);
5631 const TypeAryPtr* top_m = m_type->isa_aryptr();
5632 if (top_a == NULL || top_a->klass() == NULL ||
5633 top_n == NULL || top_n->klass() == NULL ||
5634 top_m == NULL || top_m->klass() == NULL) {
5635 // failed array check
5636 return false;
5637 }
5638
5639 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5640 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5641 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5642 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5643 return false;
5644 }
5645
5646 // Make the call
5647 {
5648 Node* a_start = array_element_address(a, intcon(0), a_elem);
5649 Node* n_start = array_element_address(n, intcon(0), n_elem);
5650 Node* m_start = array_element_address(m, intcon(0), m_elem);
5651
5652 Node* call = make_runtime_call(RC_LEAF,
5653 OptoRuntime::montgomerySquare_Type(),
5654 stubAddr, stubName, TypePtr::BOTTOM,
5655 a_start, n_start, len, inv, top(),
5656 m_start);
5657 set_result(m);
5658 }
5659
5660 return true;
5661 }
5662
5663 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5664 address stubAddr = NULL;
5665 const char* stubName = NULL;
5666
5667 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5668 if (stubAddr == NULL) {
5669 return false; // Intrinsic's stub is not implemented on this platform
5670 }
5671
5672 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5673
5674 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5675
5676 Node* newArr = argument(0);
5677 Node* oldArr = argument(1);
5678 Node* newIdx = argument(2);
5679 Node* shiftCount = argument(3);
5680 Node* numIter = argument(4);
5681
5682 const Type* newArr_type = newArr->Value(&_gvn);
5683 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5684 const Type* oldArr_type = oldArr->Value(&_gvn);
5685 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5686 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5687 || top_oldArr->klass() == NULL) {
5688 return false;
5689 }
5690
5691 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5692 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5693 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5694 return false;
5695 }
5696
5697 // Make the call
5698 {
5699 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5700 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5701
5702 Node* call = make_runtime_call(RC_LEAF,
5703 OptoRuntime::bigIntegerShift_Type(),
5704 stubAddr,
5705 stubName,
5706 TypePtr::BOTTOM,
5707 newArr_start,
5708 oldArr_start,
5709 newIdx,
5710 shiftCount,
5711 numIter);
5712 }
5713
5714 return true;
5715 }
5716
5717 //-------------inline_vectorizedMismatch------------------------------
5718 bool LibraryCallKit::inline_vectorizedMismatch() {
5719 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5720
5721 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5722 Node* obja = argument(0); // Object
5723 Node* aoffset = argument(1); // long
5724 Node* objb = argument(3); // Object
5725 Node* boffset = argument(4); // long
5726 Node* length = argument(6); // int
5727 Node* scale = argument(7); // int
5728
5729 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5730 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5731 if (obja_t == NULL || obja_t->klass() == NULL ||
5732 objb_t == NULL || objb_t->klass() == NULL ||
5733 scale == top()) {
5734 return false; // failed input validation
5735 }
5736
5737 Node* obja_adr = make_unsafe_address(obja, aoffset);
5738 Node* objb_adr = make_unsafe_address(objb, boffset);
5739
5740 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5741 //
5742 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5743 // if (length <= inline_limit) {
5744 // inline_path:
5745 // vmask = VectorMaskGen length
5746 // vload1 = LoadVectorMasked obja, vmask
5747 // vload2 = LoadVectorMasked objb, vmask
5748 // result1 = VectorCmpMasked vload1, vload2, vmask
5749 // } else {
5750 // call_stub_path:
5751 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5752 // }
5753 // exit_block:
5754 // return Phi(result1, result2);
5755 //
5756 enum { inline_path = 1, // input is small enough to process it all at once
5757 stub_path = 2, // input is too large; call into the VM
5758 PATH_LIMIT = 3
5759 };
5760
5761 Node* exit_block = new RegionNode(PATH_LIMIT);
5762 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5763 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5764
5765 Node* call_stub_path = control();
5766
5767 BasicType elem_bt = T_ILLEGAL;
5768
5769 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5770 if (scale_t->is_con()) {
5771 switch (scale_t->get_con()) {
5772 case 0: elem_bt = T_BYTE; break;
5773 case 1: elem_bt = T_SHORT; break;
5774 case 2: elem_bt = T_INT; break;
5775 case 3: elem_bt = T_LONG; break;
5776
5777 default: elem_bt = T_ILLEGAL; break; // not supported
5778 }
5779 }
5780
5781 int inline_limit = 0;
5782 bool do_partial_inline = false;
5783
5784 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5785 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5786 do_partial_inline = inline_limit >= 16;
5787 }
5788
5789 if (do_partial_inline) {
5790 assert(elem_bt != T_ILLEGAL, "sanity");
5791
5792 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5793 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5794 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5795
5796 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5797 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5798 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5799
5800 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5801
5802 if (!stopped()) {
5803 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5804
5805 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5806 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5807 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5808 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5809
5810 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
5811 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5812 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5813 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5814
5815 exit_block->init_req(inline_path, control());
5816 memory_phi->init_req(inline_path, map()->memory());
5817 result_phi->init_req(inline_path, result);
5818
5819 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5820 clear_upper_avx();
5821 }
5822 }
5823 }
5824
5825 if (call_stub_path != NULL) {
5826 set_control(call_stub_path);
5827
5828 Node* call = make_runtime_call(RC_LEAF,
5829 OptoRuntime::vectorizedMismatch_Type(),
5830 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5831 obja_adr, objb_adr, length, scale);
5832
5833 exit_block->init_req(stub_path, control());
5834 memory_phi->init_req(stub_path, map()->memory());
5835 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5836 }
5837
5838 exit_block = _gvn.transform(exit_block);
5839 memory_phi = _gvn.transform(memory_phi);
5840 result_phi = _gvn.transform(result_phi);
5841
5842 set_control(exit_block);
5843 set_all_memory(memory_phi);
5844 set_result(result_phi);
5845
5846 return true;
5847 }
5848
5849 /**
5850 * Calculate CRC32 for byte.
5851 * int java.util.zip.CRC32.update(int crc, int b)
5852 */
5853 bool LibraryCallKit::inline_updateCRC32() {
5854 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5855 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5856 // no receiver since it is static method
5857 Node* crc = argument(0); // type: int
5858 Node* b = argument(1); // type: int
5859
5860 /*
5861 * int c = ~ crc;
5862 * b = timesXtoThe32[(b ^ c) & 0xFF];
5863 * b = b ^ (c >>> 8);
5864 * crc = ~b;
5865 */
5866
5867 Node* M1 = intcon(-1);
5868 crc = _gvn.transform(new XorINode(crc, M1));
5869 Node* result = _gvn.transform(new XorINode(crc, b));
5870 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5871
5872 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5873 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5874 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5875 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5876
5877 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5878 result = _gvn.transform(new XorINode(crc, result));
5879 result = _gvn.transform(new XorINode(result, M1));
5880 set_result(result);
5881 return true;
5882 }
5883
5884 /**
5885 * Calculate CRC32 for byte[] array.
5886 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5887 */
5888 bool LibraryCallKit::inline_updateBytesCRC32() {
5889 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5890 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5891 // no receiver since it is static method
5892 Node* crc = argument(0); // type: int
5893 Node* src = argument(1); // type: oop
5894 Node* offset = argument(2); // type: int
5895 Node* length = argument(3); // type: int
5896
5897 const Type* src_type = src->Value(&_gvn);
5898 const TypeAryPtr* top_src = src_type->isa_aryptr();
5899 if (top_src == NULL || top_src->klass() == NULL) {
5900 // failed array check
5901 return false;
5902 }
5903
5904 // Figure out the size and type of the elements we will be copying.
5905 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5906 if (src_elem != T_BYTE) {
5907 return false;
5908 }
5909
5910 // 'src_start' points to src array + scaled offset
5911 src = must_be_not_null(src, true);
5912 Node* src_start = array_element_address(src, offset, src_elem);
5913
5914 // We assume that range check is done by caller.
5915 // TODO: generate range check (offset+length < src.length) in debug VM.
5916
5917 // Call the stub.
5918 address stubAddr = StubRoutines::updateBytesCRC32();
5919 const char *stubName = "updateBytesCRC32";
5920
5921 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5922 stubAddr, stubName, TypePtr::BOTTOM,
5923 crc, src_start, length);
5924 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5925 set_result(result);
5926 return true;
5927 }
5928
5929 /**
5930 * Calculate CRC32 for ByteBuffer.
5931 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5932 */
5933 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5934 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5935 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5936 // no receiver since it is static method
5937 Node* crc = argument(0); // type: int
5938 Node* src = argument(1); // type: long
5939 Node* offset = argument(3); // type: int
5940 Node* length = argument(4); // type: int
5941
5942 src = ConvL2X(src); // adjust Java long to machine word
5943 Node* base = _gvn.transform(new CastX2PNode(src));
5944 offset = ConvI2X(offset);
5945
5946 // 'src_start' points to src array + scaled offset
5947 Node* src_start = basic_plus_adr(top(), base, offset);
5948
5949 // Call the stub.
5950 address stubAddr = StubRoutines::updateBytesCRC32();
5951 const char *stubName = "updateBytesCRC32";
5952
5953 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5954 stubAddr, stubName, TypePtr::BOTTOM,
5955 crc, src_start, length);
5956 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5957 set_result(result);
5958 return true;
5959 }
5960
5961 //------------------------------get_table_from_crc32c_class-----------------------
5962 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5963 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
5964 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5965
5966 return table;
5967 }
5968
5969 //------------------------------inline_updateBytesCRC32C-----------------------
5970 //
5971 // Calculate CRC32C for byte[] array.
5972 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5973 //
5974 bool LibraryCallKit::inline_updateBytesCRC32C() {
5975 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5976 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5977 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5978 // no receiver since it is a static method
5979 Node* crc = argument(0); // type: int
5980 Node* src = argument(1); // type: oop
5981 Node* offset = argument(2); // type: int
5982 Node* end = argument(3); // type: int
5983
5984 Node* length = _gvn.transform(new SubINode(end, offset));
5985
5986 const Type* src_type = src->Value(&_gvn);
5987 const TypeAryPtr* top_src = src_type->isa_aryptr();
5988 if (top_src == NULL || top_src->klass() == NULL) {
5989 // failed array check
5990 return false;
5991 }
5992
5993 // Figure out the size and type of the elements we will be copying.
5994 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5995 if (src_elem != T_BYTE) {
5996 return false;
5997 }
5998
5999 // 'src_start' points to src array + scaled offset
6000 src = must_be_not_null(src, true);
6001 Node* src_start = array_element_address(src, offset, src_elem);
6002
6003 // static final int[] byteTable in class CRC32C
6004 Node* table = get_table_from_crc32c_class(callee()->holder());
6005 table = must_be_not_null(table, true);
6006 Node* table_start = array_element_address(table, intcon(0), T_INT);
6007
6008 // We assume that range check is done by caller.
6009 // TODO: generate range check (offset+length < src.length) in debug VM.
6010
6011 // Call the stub.
6012 address stubAddr = StubRoutines::updateBytesCRC32C();
6013 const char *stubName = "updateBytesCRC32C";
6014
6015 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6016 stubAddr, stubName, TypePtr::BOTTOM,
6017 crc, src_start, length, table_start);
6018 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6019 set_result(result);
6020 return true;
6021 }
6022
6023 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6024 //
6025 // Calculate CRC32C for DirectByteBuffer.
6026 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6027 //
6028 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6029 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6030 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6031 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6032 // no receiver since it is a static method
6033 Node* crc = argument(0); // type: int
6034 Node* src = argument(1); // type: long
6035 Node* offset = argument(3); // type: int
6036 Node* end = argument(4); // type: int
6037
6038 Node* length = _gvn.transform(new SubINode(end, offset));
6039
6040 src = ConvL2X(src); // adjust Java long to machine word
6041 Node* base = _gvn.transform(new CastX2PNode(src));
6042 offset = ConvI2X(offset);
6043
6044 // 'src_start' points to src array + scaled offset
6045 Node* src_start = basic_plus_adr(top(), base, offset);
6046
6047 // static final int[] byteTable in class CRC32C
6048 Node* table = get_table_from_crc32c_class(callee()->holder());
6049 table = must_be_not_null(table, true);
6050 Node* table_start = array_element_address(table, intcon(0), T_INT);
6051
6052 // Call the stub.
6053 address stubAddr = StubRoutines::updateBytesCRC32C();
6054 const char *stubName = "updateBytesCRC32C";
6055
6056 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6057 stubAddr, stubName, TypePtr::BOTTOM,
6058 crc, src_start, length, table_start);
6059 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6060 set_result(result);
6061 return true;
6062 }
6063
6064 //------------------------------inline_updateBytesAdler32----------------------
6065 //
6066 // Calculate Adler32 checksum for byte[] array.
6067 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6068 //
6069 bool LibraryCallKit::inline_updateBytesAdler32() {
6070 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6071 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6072 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6073 // no receiver since it is static method
6074 Node* crc = argument(0); // type: int
6075 Node* src = argument(1); // type: oop
6076 Node* offset = argument(2); // type: int
6077 Node* length = argument(3); // type: int
6078
6079 const Type* src_type = src->Value(&_gvn);
6080 const TypeAryPtr* top_src = src_type->isa_aryptr();
6081 if (top_src == NULL || top_src->klass() == NULL) {
6082 // failed array check
6083 return false;
6084 }
6085
6086 // Figure out the size and type of the elements we will be copying.
6087 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6088 if (src_elem != T_BYTE) {
6089 return false;
6090 }
6091
6092 // 'src_start' points to src array + scaled offset
6093 Node* src_start = array_element_address(src, offset, src_elem);
6094
6095 // We assume that range check is done by caller.
6096 // TODO: generate range check (offset+length < src.length) in debug VM.
6097
6098 // Call the stub.
6099 address stubAddr = StubRoutines::updateBytesAdler32();
6100 const char *stubName = "updateBytesAdler32";
6101
6102 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6103 stubAddr, stubName, TypePtr::BOTTOM,
6104 crc, src_start, length);
6105 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6106 set_result(result);
6107 return true;
6108 }
6109
6110 //------------------------------inline_updateByteBufferAdler32---------------
6111 //
6112 // Calculate Adler32 checksum for DirectByteBuffer.
6113 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6114 //
6115 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6116 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6117 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6118 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6119 // no receiver since it is static method
6120 Node* crc = argument(0); // type: int
6121 Node* src = argument(1); // type: long
6122 Node* offset = argument(3); // type: int
6123 Node* length = argument(4); // type: int
6124
6125 src = ConvL2X(src); // adjust Java long to machine word
6126 Node* base = _gvn.transform(new CastX2PNode(src));
6127 offset = ConvI2X(offset);
6128
6129 // 'src_start' points to src array + scaled offset
6130 Node* src_start = basic_plus_adr(top(), base, offset);
6131
6132 // Call the stub.
6133 address stubAddr = StubRoutines::updateBytesAdler32();
6134 const char *stubName = "updateBytesAdler32";
6135
6136 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6137 stubAddr, stubName, TypePtr::BOTTOM,
6138 crc, src_start, length);
6139
6140 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6141 set_result(result);
6142 return true;
6143 }
6144
6145 //----------------------------inline_reference_get----------------------------
6146 // public T java.lang.ref.Reference.get();
6147 bool LibraryCallKit::inline_reference_get() {
6148 const int referent_offset = java_lang_ref_Reference::referent_offset();
6149
6150 // Get the argument:
6151 Node* reference_obj = null_check_receiver();
6152 if (stopped()) return true;
6153
6154 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6155 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6156 decorators, /*is_static*/ false, NULL);
6157 if (result == NULL) return false;
6158
6159 // Add memory barrier to prevent commoning reads from this field
6160 // across safepoint since GC can change its value.
6161 insert_mem_bar(Op_MemBarCPUOrder);
6162
6163 set_result(result);
6164 return true;
6165 }
6166
6167 //----------------------------inline_reference_refersTo0----------------------------
6168 // bool java.lang.ref.Reference.refersTo0();
6169 // bool java.lang.ref.PhantomReference.refersTo0();
6170 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6171 // Get arguments:
6172 Node* reference_obj = null_check_receiver();
6173 Node* other_obj = argument(1);
6174 if (stopped()) return true;
6175
6176 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6177 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6178 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6179 decorators, /*is_static*/ false, NULL);
6180 if (referent == NULL) return false;
6181
6182 // Add memory barrier to prevent commoning reads from this field
6183 // across safepoint since GC can change its value.
6184 insert_mem_bar(Op_MemBarCPUOrder);
6185
6186 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6187 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6188 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6189
6190 RegionNode* region = new RegionNode(3);
6191 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6192
6193 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6194 region->init_req(1, if_true);
6195 phi->init_req(1, intcon(1));
6196
6197 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6198 region->init_req(2, if_false);
6199 phi->init_req(2, intcon(0));
6200
6201 set_control(_gvn.transform(region));
6202 record_for_igvn(region);
6203 set_result(_gvn.transform(phi));
6204 return true;
6205 }
6206
6207
6208 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6209 DecoratorSet decorators = IN_HEAP, bool is_static = false,
6210 ciInstanceKlass* fromKls = NULL) {
6211 if (fromKls == NULL) {
6212 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6213 assert(tinst != NULL, "obj is null");
6214 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6215 fromKls = tinst->klass()->as_instance_klass();
6216 } else {
6217 assert(is_static, "only for static field access");
6218 }
6219 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6220 ciSymbol::make(fieldTypeString),
6221 is_static);
6222
6223 assert (field != NULL, "undefined field");
6224 if (field == NULL) return (Node *) NULL;
6225
6226 if (is_static) {
6227 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6228 fromObj = makecon(tip);
6229 }
6230
6231 // Next code copied from Parse::do_get_xxx():
6232
6233 // Compute address and memory type.
6234 int offset = field->offset_in_bytes();
6235 bool is_vol = field->is_volatile();
6236 ciType* field_klass = field->type();
6237 assert(field_klass->is_loaded(), "should be loaded");
6238 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6239 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6240 BasicType bt = field->layout_type();
6241
6242 // Build the resultant type of the load
6243 const Type *type;
6244 if (bt == T_OBJECT) {
6245 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6246 } else {
6247 type = Type::get_const_basic_type(bt);
6248 }
6249
6250 if (is_vol) {
6251 decorators |= MO_SEQ_CST;
6252 }
6253
6254 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6255 }
6256
6257 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6258 bool is_exact = true, bool is_static = false,
6259 ciInstanceKlass * fromKls = NULL) {
6260 if (fromKls == NULL) {
6261 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6262 assert(tinst != NULL, "obj is null");
6263 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6264 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6265 fromKls = tinst->klass()->as_instance_klass();
6266 }
6267 else {
6268 assert(is_static, "only for static field access");
6269 }
6270 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6271 ciSymbol::make(fieldTypeString),
6272 is_static);
6273
6274 assert(field != NULL, "undefined field");
6275 assert(!field->is_volatile(), "not defined for volatile fields");
6276
6277 if (is_static) {
6278 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6279 fromObj = makecon(tip);
6280 }
6281
6282 // Next code copied from Parse::do_get_xxx():
6283
6284 // Compute address and memory type.
6285 int offset = field->offset_in_bytes();
6286 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6287
6288 return adr;
6289 }
6290
6291 //------------------------------inline_aescrypt_Block-----------------------
6292 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6293 address stubAddr = NULL;
6294 const char *stubName;
6295 assert(UseAES, "need AES instruction support");
6296
6297 switch(id) {
6298 case vmIntrinsics::_aescrypt_encryptBlock:
6299 stubAddr = StubRoutines::aescrypt_encryptBlock();
6300 stubName = "aescrypt_encryptBlock";
6301 break;
6302 case vmIntrinsics::_aescrypt_decryptBlock:
6303 stubAddr = StubRoutines::aescrypt_decryptBlock();
6304 stubName = "aescrypt_decryptBlock";
6305 break;
6306 default:
6307 break;
6308 }
6309 if (stubAddr == NULL) return false;
6310
6311 Node* aescrypt_object = argument(0);
6312 Node* src = argument(1);
6313 Node* src_offset = argument(2);
6314 Node* dest = argument(3);
6315 Node* dest_offset = argument(4);
6316
6317 src = must_be_not_null(src, true);
6318 dest = must_be_not_null(dest, true);
6319
6320 // (1) src and dest are arrays.
6321 const Type* src_type = src->Value(&_gvn);
6322 const Type* dest_type = dest->Value(&_gvn);
6323 const TypeAryPtr* top_src = src_type->isa_aryptr();
6324 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6325 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6326
6327 // for the quick and dirty code we will skip all the checks.
6328 // we are just trying to get the call to be generated.
6329 Node* src_start = src;
6330 Node* dest_start = dest;
6331 if (src_offset != NULL || dest_offset != NULL) {
6332 assert(src_offset != NULL && dest_offset != NULL, "");
6333 src_start = array_element_address(src, src_offset, T_BYTE);
6334 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6335 }
6336
6337 // now need to get the start of its expanded key array
6338 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6339 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6340 if (k_start == NULL) return false;
6341
6342 // Call the stub.
6343 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6344 stubAddr, stubName, TypePtr::BOTTOM,
6345 src_start, dest_start, k_start);
6346
6347 return true;
6348 }
6349
6350 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6351 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6352 address stubAddr = NULL;
6353 const char *stubName = NULL;
6354
6355 assert(UseAES, "need AES instruction support");
6356
6357 switch(id) {
6358 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6359 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6360 stubName = "cipherBlockChaining_encryptAESCrypt";
6361 break;
6362 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6363 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6364 stubName = "cipherBlockChaining_decryptAESCrypt";
6365 break;
6366 default:
6367 break;
6368 }
6369 if (stubAddr == NULL) return false;
6370
6371 Node* cipherBlockChaining_object = argument(0);
6372 Node* src = argument(1);
6373 Node* src_offset = argument(2);
6374 Node* len = argument(3);
6375 Node* dest = argument(4);
6376 Node* dest_offset = argument(5);
6377
6378 src = must_be_not_null(src, false);
6379 dest = must_be_not_null(dest, false);
6380
6381 // (1) src and dest are arrays.
6382 const Type* src_type = src->Value(&_gvn);
6383 const Type* dest_type = dest->Value(&_gvn);
6384 const TypeAryPtr* top_src = src_type->isa_aryptr();
6385 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6386 assert (top_src != NULL && top_src->klass() != NULL
6387 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6388
6389 // checks are the responsibility of the caller
6390 Node* src_start = src;
6391 Node* dest_start = dest;
6392 if (src_offset != NULL || dest_offset != NULL) {
6393 assert(src_offset != NULL && dest_offset != NULL, "");
6394 src_start = array_element_address(src, src_offset, T_BYTE);
6395 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6396 }
6397
6398 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6399 // (because of the predicated logic executed earlier).
6400 // so we cast it here safely.
6401 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6402
6403 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6404 if (embeddedCipherObj == NULL) return false;
6405
6406 // cast it to what we know it will be at runtime
6407 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6408 assert(tinst != NULL, "CBC obj is null");
6409 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6410 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6411 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6412
6413 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6414 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6415 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6416 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6417 aescrypt_object = _gvn.transform(aescrypt_object);
6418
6419 // we need to get the start of the aescrypt_object's expanded key array
6420 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6421 if (k_start == NULL) return false;
6422
6423 // similarly, get the start address of the r vector
6424 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6425 if (objRvec == NULL) return false;
6426 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6427
6428 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6429 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6430 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6431 stubAddr, stubName, TypePtr::BOTTOM,
6432 src_start, dest_start, k_start, r_start, len);
6433
6434 // return cipher length (int)
6435 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6436 set_result(retvalue);
6437 return true;
6438 }
6439
6440 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6441 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6442 address stubAddr = NULL;
6443 const char *stubName = NULL;
6444
6445 assert(UseAES, "need AES instruction support");
6446
6447 switch (id) {
6448 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6449 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6450 stubName = "electronicCodeBook_encryptAESCrypt";
6451 break;
6452 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6453 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6454 stubName = "electronicCodeBook_decryptAESCrypt";
6455 break;
6456 default:
6457 break;
6458 }
6459
6460 if (stubAddr == NULL) return false;
6461
6462 Node* electronicCodeBook_object = argument(0);
6463 Node* src = argument(1);
6464 Node* src_offset = argument(2);
6465 Node* len = argument(3);
6466 Node* dest = argument(4);
6467 Node* dest_offset = argument(5);
6468
6469 // (1) src and dest are arrays.
6470 const Type* src_type = src->Value(&_gvn);
6471 const Type* dest_type = dest->Value(&_gvn);
6472 const TypeAryPtr* top_src = src_type->isa_aryptr();
6473 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6474 assert(top_src != NULL && top_src->klass() != NULL
6475 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6476
6477 // checks are the responsibility of the caller
6478 Node* src_start = src;
6479 Node* dest_start = dest;
6480 if (src_offset != NULL || dest_offset != NULL) {
6481 assert(src_offset != NULL && dest_offset != NULL, "");
6482 src_start = array_element_address(src, src_offset, T_BYTE);
6483 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6484 }
6485
6486 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6487 // (because of the predicated logic executed earlier).
6488 // so we cast it here safely.
6489 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6490
6491 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6492 if (embeddedCipherObj == NULL) return false;
6493
6494 // cast it to what we know it will be at runtime
6495 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6496 assert(tinst != NULL, "ECB obj is null");
6497 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6498 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6499 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6500
6501 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6502 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6503 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6504 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6505 aescrypt_object = _gvn.transform(aescrypt_object);
6506
6507 // we need to get the start of the aescrypt_object's expanded key array
6508 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6509 if (k_start == NULL) return false;
6510
6511 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6512 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6513 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6514 stubAddr, stubName, TypePtr::BOTTOM,
6515 src_start, dest_start, k_start, len);
6516
6517 // return cipher length (int)
6518 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6519 set_result(retvalue);
6520 return true;
6521 }
6522
6523 //------------------------------inline_counterMode_AESCrypt-----------------------
6524 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6525 assert(UseAES, "need AES instruction support");
6526 if (!UseAESCTRIntrinsics) return false;
6527
6528 address stubAddr = NULL;
6529 const char *stubName = NULL;
6530 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6531 stubAddr = StubRoutines::counterMode_AESCrypt();
6532 stubName = "counterMode_AESCrypt";
6533 }
6534 if (stubAddr == NULL) return false;
6535
6536 Node* counterMode_object = argument(0);
6537 Node* src = argument(1);
6538 Node* src_offset = argument(2);
6539 Node* len = argument(3);
6540 Node* dest = argument(4);
6541 Node* dest_offset = argument(5);
6542
6543 // (1) src and dest are arrays.
6544 const Type* src_type = src->Value(&_gvn);
6545 const Type* dest_type = dest->Value(&_gvn);
6546 const TypeAryPtr* top_src = src_type->isa_aryptr();
6547 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6548 assert(top_src != NULL && top_src->klass() != NULL &&
6549 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6550
6551 // checks are the responsibility of the caller
6552 Node* src_start = src;
6553 Node* dest_start = dest;
6554 if (src_offset != NULL || dest_offset != NULL) {
6555 assert(src_offset != NULL && dest_offset != NULL, "");
6556 src_start = array_element_address(src, src_offset, T_BYTE);
6557 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6558 }
6559
6560 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6561 // (because of the predicated logic executed earlier).
6562 // so we cast it here safely.
6563 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6564 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6565 if (embeddedCipherObj == NULL) return false;
6566 // cast it to what we know it will be at runtime
6567 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6568 assert(tinst != NULL, "CTR obj is null");
6569 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6570 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6571 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6572 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6573 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6574 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6575 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6576 aescrypt_object = _gvn.transform(aescrypt_object);
6577 // we need to get the start of the aescrypt_object's expanded key array
6578 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6579 if (k_start == NULL) return false;
6580 // similarly, get the start address of the r vector
6581 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6582 if (obj_counter == NULL) return false;
6583 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6584
6585 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6586 if (saved_encCounter == NULL) return false;
6587 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6588 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6589
6590 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6591 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6592 OptoRuntime::counterMode_aescrypt_Type(),
6593 stubAddr, stubName, TypePtr::BOTTOM,
6594 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6595
6596 // return cipher length (int)
6597 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6598 set_result(retvalue);
6599 return true;
6600 }
6601
6602 //------------------------------get_key_start_from_aescrypt_object-----------------------
6603 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6604 #if defined(PPC64) || defined(S390)
6605 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6606 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6607 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6608 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6609 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6610 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6611 if (objSessionK == NULL) {
6612 return (Node *) NULL;
6613 }
6614 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6615 #else
6616 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6617 #endif // PPC64
6618 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6619 if (objAESCryptKey == NULL) return (Node *) NULL;
6620
6621 // now have the array, need to get the start address of the K array
6622 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6623 return k_start;
6624 }
6625
6626 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6627 // Return node representing slow path of predicate check.
6628 // the pseudo code we want to emulate with this predicate is:
6629 // for encryption:
6630 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6631 // for decryption:
6632 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6633 // note cipher==plain is more conservative than the original java code but that's OK
6634 //
6635 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6636 // The receiver was checked for NULL already.
6637 Node* objCBC = argument(0);
6638
6639 Node* src = argument(1);
6640 Node* dest = argument(4);
6641
6642 // Load embeddedCipher field of CipherBlockChaining object.
6643 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6644
6645 // get AESCrypt klass for instanceOf check
6646 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6647 // will have same classloader as CipherBlockChaining object
6648 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6649 assert(tinst != NULL, "CBCobj is null");
6650 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6651
6652 // we want to do an instanceof comparison against the AESCrypt class
6653 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6654 if (!klass_AESCrypt->is_loaded()) {
6655 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6656 Node* ctrl = control();
6657 set_control(top()); // no regular fast path
6658 return ctrl;
6659 }
6660
6661 src = must_be_not_null(src, true);
6662 dest = must_be_not_null(dest, true);
6663
6664 // Resolve oops to stable for CmpP below.
6665 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6666
6667 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6668 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6669 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6670
6671 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6672
6673 // for encryption, we are done
6674 if (!decrypting)
6675 return instof_false; // even if it is NULL
6676
6677 // for decryption, we need to add a further check to avoid
6678 // taking the intrinsic path when cipher and plain are the same
6679 // see the original java code for why.
6680 RegionNode* region = new RegionNode(3);
6681 region->init_req(1, instof_false);
6682
6683 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6684 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6685 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6686 region->init_req(2, src_dest_conjoint);
6687
6688 record_for_igvn(region);
6689 return _gvn.transform(region);
6690 }
6691
6692 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6693 // Return node representing slow path of predicate check.
6694 // the pseudo code we want to emulate with this predicate is:
6695 // for encryption:
6696 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6697 // for decryption:
6698 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6699 // note cipher==plain is more conservative than the original java code but that's OK
6700 //
6701 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6702 // The receiver was checked for NULL already.
6703 Node* objECB = argument(0);
6704
6705 // Load embeddedCipher field of ElectronicCodeBook object.
6706 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6707
6708 // get AESCrypt klass for instanceOf check
6709 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6710 // will have same classloader as ElectronicCodeBook object
6711 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6712 assert(tinst != NULL, "ECBobj is null");
6713 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6714
6715 // we want to do an instanceof comparison against the AESCrypt class
6716 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6717 if (!klass_AESCrypt->is_loaded()) {
6718 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6719 Node* ctrl = control();
6720 set_control(top()); // no regular fast path
6721 return ctrl;
6722 }
6723 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6724
6725 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6726 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6727 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6728
6729 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6730
6731 // for encryption, we are done
6732 if (!decrypting)
6733 return instof_false; // even if it is NULL
6734
6735 // for decryption, we need to add a further check to avoid
6736 // taking the intrinsic path when cipher and plain are the same
6737 // see the original java code for why.
6738 RegionNode* region = new RegionNode(3);
6739 region->init_req(1, instof_false);
6740 Node* src = argument(1);
6741 Node* dest = argument(4);
6742 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6743 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6744 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6745 region->init_req(2, src_dest_conjoint);
6746
6747 record_for_igvn(region);
6748 return _gvn.transform(region);
6749 }
6750
6751 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6752 // Return node representing slow path of predicate check.
6753 // the pseudo code we want to emulate with this predicate is:
6754 // for encryption:
6755 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6756 // for decryption:
6757 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6758 // note cipher==plain is more conservative than the original java code but that's OK
6759 //
6760
6761 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6762 // The receiver was checked for NULL already.
6763 Node* objCTR = argument(0);
6764
6765 // Load embeddedCipher field of CipherBlockChaining object.
6766 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6767
6768 // get AESCrypt klass for instanceOf check
6769 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6770 // will have same classloader as CipherBlockChaining object
6771 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6772 assert(tinst != NULL, "CTRobj is null");
6773 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6774
6775 // we want to do an instanceof comparison against the AESCrypt class
6776 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6777 if (!klass_AESCrypt->is_loaded()) {
6778 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6779 Node* ctrl = control();
6780 set_control(top()); // no regular fast path
6781 return ctrl;
6782 }
6783
6784 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6785 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6786 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6787 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6788 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6789
6790 return instof_false; // even if it is NULL
6791 }
6792
6793 //------------------------------inline_ghash_processBlocks
6794 bool LibraryCallKit::inline_ghash_processBlocks() {
6795 address stubAddr;
6796 const char *stubName;
6797 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6798
6799 stubAddr = StubRoutines::ghash_processBlocks();
6800 stubName = "ghash_processBlocks";
6801
6802 Node* data = argument(0);
6803 Node* offset = argument(1);
6804 Node* len = argument(2);
6805 Node* state = argument(3);
6806 Node* subkeyH = argument(4);
6807
6808 state = must_be_not_null(state, true);
6809 subkeyH = must_be_not_null(subkeyH, true);
6810 data = must_be_not_null(data, true);
6811
6812 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6813 assert(state_start, "state is NULL");
6814 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6815 assert(subkeyH_start, "subkeyH is NULL");
6816 Node* data_start = array_element_address(data, offset, T_BYTE);
6817 assert(data_start, "data is NULL");
6818
6819 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6820 OptoRuntime::ghash_processBlocks_Type(),
6821 stubAddr, stubName, TypePtr::BOTTOM,
6822 state_start, subkeyH_start, data_start, len);
6823 return true;
6824 }
6825
6826 bool LibraryCallKit::inline_base64_encodeBlock() {
6827 address stubAddr;
6828 const char *stubName;
6829 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6830 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6831 stubAddr = StubRoutines::base64_encodeBlock();
6832 stubName = "encodeBlock";
6833
6834 if (!stubAddr) return false;
6835 Node* base64obj = argument(0);
6836 Node* src = argument(1);
6837 Node* offset = argument(2);
6838 Node* len = argument(3);
6839 Node* dest = argument(4);
6840 Node* dp = argument(5);
6841 Node* isURL = argument(6);
6842
6843 src = must_be_not_null(src, true);
6844 dest = must_be_not_null(dest, true);
6845
6846 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6847 assert(src_start, "source array is NULL");
6848 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6849 assert(dest_start, "destination array is NULL");
6850
6851 Node* base64 = make_runtime_call(RC_LEAF,
6852 OptoRuntime::base64_encodeBlock_Type(),
6853 stubAddr, stubName, TypePtr::BOTTOM,
6854 src_start, offset, len, dest_start, dp, isURL);
6855 return true;
6856 }
6857
6858 bool LibraryCallKit::inline_base64_decodeBlock() {
6859 address stubAddr;
6860 const char *stubName;
6861 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6862 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6863 stubAddr = StubRoutines::base64_decodeBlock();
6864 stubName = "decodeBlock";
6865
6866 if (!stubAddr) return false;
6867 Node* base64obj = argument(0);
6868 Node* src = argument(1);
6869 Node* src_offset = argument(2);
6870 Node* len = argument(3);
6871 Node* dest = argument(4);
6872 Node* dest_offset = argument(5);
6873 Node* isURL = argument(6);
6874 Node* isMIME = argument(7);
6875
6876 src = must_be_not_null(src, true);
6877 dest = must_be_not_null(dest, true);
6878
6879 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6880 assert(src_start, "source array is NULL");
6881 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6882 assert(dest_start, "destination array is NULL");
6883
6884 Node* call = make_runtime_call(RC_LEAF,
6885 OptoRuntime::base64_decodeBlock_Type(),
6886 stubAddr, stubName, TypePtr::BOTTOM,
6887 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6888 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6889 set_result(result);
6890 return true;
6891 }
6892
6893 //------------------------------inline_digestBase_implCompress-----------------------
6894 //
6895 // Calculate MD5 for single-block byte[] array.
6896 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6897 //
6898 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6899 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6900 //
6901 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6902 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6903 //
6904 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6905 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6906 //
6907 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6908 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6909 //
6910 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6911 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6912
6913 Node* digestBase_obj = argument(0);
6914 Node* src = argument(1); // type oop
6915 Node* ofs = argument(2); // type int
6916
6917 const Type* src_type = src->Value(&_gvn);
6918 const TypeAryPtr* top_src = src_type->isa_aryptr();
6919 if (top_src == NULL || top_src->klass() == NULL) {
6920 // failed array check
6921 return false;
6922 }
6923 // Figure out the size and type of the elements we will be copying.
6924 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6925 if (src_elem != T_BYTE) {
6926 return false;
6927 }
6928 // 'src_start' points to src array + offset
6929 src = must_be_not_null(src, true);
6930 Node* src_start = array_element_address(src, ofs, src_elem);
6931 Node* state = NULL;
6932 Node* digest_length = NULL;
6933 address stubAddr;
6934 const char *stubName;
6935
6936 switch(id) {
6937 case vmIntrinsics::_md5_implCompress:
6938 assert(UseMD5Intrinsics, "need MD5 instruction support");
6939 state = get_state_from_digest_object(digestBase_obj, "[I");
6940 stubAddr = StubRoutines::md5_implCompress();
6941 stubName = "md5_implCompress";
6942 break;
6943 case vmIntrinsics::_sha_implCompress:
6944 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6945 state = get_state_from_digest_object(digestBase_obj, "[I");
6946 stubAddr = StubRoutines::sha1_implCompress();
6947 stubName = "sha1_implCompress";
6948 break;
6949 case vmIntrinsics::_sha2_implCompress:
6950 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6951 state = get_state_from_digest_object(digestBase_obj, "[I");
6952 stubAddr = StubRoutines::sha256_implCompress();
6953 stubName = "sha256_implCompress";
6954 break;
6955 case vmIntrinsics::_sha5_implCompress:
6956 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6957 state = get_state_from_digest_object(digestBase_obj, "[J");
6958 stubAddr = StubRoutines::sha512_implCompress();
6959 stubName = "sha512_implCompress";
6960 break;
6961 case vmIntrinsics::_sha3_implCompress:
6962 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
6963 state = get_state_from_digest_object(digestBase_obj, "[B");
6964 stubAddr = StubRoutines::sha3_implCompress();
6965 stubName = "sha3_implCompress";
6966 digest_length = get_digest_length_from_digest_object(digestBase_obj);
6967 if (digest_length == NULL) return false;
6968 break;
6969 default:
6970 fatal_unexpected_iid(id);
6971 return false;
6972 }
6973 if (state == NULL) return false;
6974
6975 assert(stubAddr != NULL, "Stub is generated");
6976 if (stubAddr == NULL) return false;
6977
6978 // Call the stub.
6979 Node* call;
6980 if (digest_length == NULL) {
6981 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
6982 stubAddr, stubName, TypePtr::BOTTOM,
6983 src_start, state);
6984 } else {
6985 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
6986 stubAddr, stubName, TypePtr::BOTTOM,
6987 src_start, state, digest_length);
6988 }
6989
6990 return true;
6991 }
6992
6993 //------------------------------inline_digestBase_implCompressMB-----------------------
6994 //
6995 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
6996 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6997 //
6998 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6999 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7000 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7001 assert((uint)predicate < 5, "sanity");
7002 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7003
7004 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7005 Node* src = argument(1); // byte[] array
7006 Node* ofs = argument(2); // type int
7007 Node* limit = argument(3); // type int
7008
7009 const Type* src_type = src->Value(&_gvn);
7010 const TypeAryPtr* top_src = src_type->isa_aryptr();
7011 if (top_src == NULL || top_src->klass() == NULL) {
7012 // failed array check
7013 return false;
7014 }
7015 // Figure out the size and type of the elements we will be copying.
7016 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
7017 if (src_elem != T_BYTE) {
7018 return false;
7019 }
7020 // 'src_start' points to src array + offset
7021 src = must_be_not_null(src, false);
7022 Node* src_start = array_element_address(src, ofs, src_elem);
7023
7024 const char* klass_digestBase_name = NULL;
7025 const char* stub_name = NULL;
7026 address stub_addr = NULL;
7027 const char* state_type = "[I";
7028
7029 switch (predicate) {
7030 case 0:
7031 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7032 klass_digestBase_name = "sun/security/provider/MD5";
7033 stub_name = "md5_implCompressMB";
7034 stub_addr = StubRoutines::md5_implCompressMB();
7035 }
7036 break;
7037 case 1:
7038 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7039 klass_digestBase_name = "sun/security/provider/SHA";
7040 stub_name = "sha1_implCompressMB";
7041 stub_addr = StubRoutines::sha1_implCompressMB();
7042 }
7043 break;
7044 case 2:
7045 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7046 klass_digestBase_name = "sun/security/provider/SHA2";
7047 stub_name = "sha256_implCompressMB";
7048 stub_addr = StubRoutines::sha256_implCompressMB();
7049 }
7050 break;
7051 case 3:
7052 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7053 klass_digestBase_name = "sun/security/provider/SHA5";
7054 stub_name = "sha512_implCompressMB";
7055 stub_addr = StubRoutines::sha512_implCompressMB();
7056 state_type = "[J";
7057 }
7058 break;
7059 case 4:
7060 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7061 klass_digestBase_name = "sun/security/provider/SHA3";
7062 stub_name = "sha3_implCompressMB";
7063 stub_addr = StubRoutines::sha3_implCompressMB();
7064 state_type = "[B";
7065 }
7066 break;
7067 default:
7068 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7069 }
7070 if (klass_digestBase_name != NULL) {
7071 assert(stub_addr != NULL, "Stub is generated");
7072 if (stub_addr == NULL) return false;
7073
7074 // get DigestBase klass to lookup for SHA klass
7075 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7076 assert(tinst != NULL, "digestBase_obj is not instance???");
7077 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7078
7079 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7080 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7081 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7082 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, state_type, stub_addr, stub_name, src_start, ofs, limit);
7083 }
7084 return false;
7085 }
7086
7087 //------------------------------inline_digestBase_implCompressMB-----------------------
7088 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7089 const char* state_type, address stubAddr, const char *stubName,
7090 Node* src_start, Node* ofs, Node* limit) {
7091 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7092 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7093 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7094 digest_obj = _gvn.transform(digest_obj);
7095
7096 Node* state = get_state_from_digest_object(digest_obj, state_type);
7097 if (state == NULL) return false;
7098
7099 Node* digest_length = NULL;
7100 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7101 digest_length = get_digest_length_from_digest_object(digest_obj);
7102 if (digest_length == NULL) return false;
7103 }
7104
7105 // Call the stub.
7106 Node* call;
7107 if (digest_length == NULL) {
7108 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7109 OptoRuntime::digestBase_implCompressMB_Type(false),
7110 stubAddr, stubName, TypePtr::BOTTOM,
7111 src_start, state, ofs, limit);
7112 } else {
7113 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7114 OptoRuntime::digestBase_implCompressMB_Type(true),
7115 stubAddr, stubName, TypePtr::BOTTOM,
7116 src_start, state, digest_length, ofs, limit);
7117 }
7118
7119 // return ofs (int)
7120 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7121 set_result(result);
7122
7123 return true;
7124 }
7125
7126 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7127 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7128 assert(UseAES, "need AES instruction support");
7129 address stubAddr = NULL;
7130 const char *stubName = NULL;
7131 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7132 stubName = "galoisCounterMode_AESCrypt";
7133
7134 if (stubAddr == NULL) return false;
7135
7136 Node* in = argument(0);
7137 Node* inOfs = argument(1);
7138 Node* len = argument(2);
7139 Node* ct = argument(3);
7140 Node* ctOfs = argument(4);
7141 Node* out = argument(5);
7142 Node* outOfs = argument(6);
7143 Node* gctr_object = argument(7);
7144 Node* ghash_object = argument(8);
7145
7146 // (1) in, ct and out are arrays.
7147 const Type* in_type = in->Value(&_gvn);
7148 const Type* ct_type = ct->Value(&_gvn);
7149 const Type* out_type = out->Value(&_gvn);
7150 const TypeAryPtr* top_in = in_type->isa_aryptr();
7151 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7152 const TypeAryPtr* top_out = out_type->isa_aryptr();
7153 assert(top_in != NULL && top_in->klass() != NULL &&
7154 top_ct != NULL && top_ct->klass() != NULL &&
7155 top_out != NULL && top_out->klass() != NULL, "args are strange");
7156
7157 // checks are the responsibility of the caller
7158 Node* in_start = in;
7159 Node* ct_start = ct;
7160 Node* out_start = out;
7161 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7162 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7163 in_start = array_element_address(in, inOfs, T_BYTE);
7164 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7165 out_start = array_element_address(out, outOfs, T_BYTE);
7166 }
7167
7168 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7169 // (because of the predicated logic executed earlier).
7170 // so we cast it here safely.
7171 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7172 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7173 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7174 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7175 Node* state = load_field_from_object(ghash_object, "state", "[J");
7176
7177 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7178 return false;
7179 }
7180 // cast it to what we know it will be at runtime
7181 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7182 assert(tinst != NULL, "GCTR obj is null");
7183 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
7184 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7185 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7186 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7187 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7188 const TypeOopPtr* xtype = aklass->as_instance_type();
7189 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7190 aescrypt_object = _gvn.transform(aescrypt_object);
7191 // we need to get the start of the aescrypt_object's expanded key array
7192 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7193 if (k_start == NULL) return false;
7194 // similarly, get the start address of the r vector
7195 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7196 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7197 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7198
7199
7200 // Call the stub, passing params
7201 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7202 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7203 stubAddr, stubName, TypePtr::BOTTOM,
7204 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7205
7206 // return cipher length (int)
7207 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7208 set_result(retvalue);
7209
7210 return true;
7211 }
7212
7213 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7214 // Return node representing slow path of predicate check.
7215 // the pseudo code we want to emulate with this predicate is:
7216 // for encryption:
7217 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7218 // for decryption:
7219 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7220 // note cipher==plain is more conservative than the original java code but that's OK
7221 //
7222
7223 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7224 // The receiver was checked for NULL already.
7225 Node* objGCTR = argument(7);
7226 // Load embeddedCipher field of GCTR object.
7227 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7228 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7229
7230 // get AESCrypt klass for instanceOf check
7231 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7232 // will have same classloader as CipherBlockChaining object
7233 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7234 assert(tinst != NULL, "GCTR obj is null");
7235 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
7236
7237 // we want to do an instanceof comparison against the AESCrypt class
7238 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7239 if (!klass_AESCrypt->is_loaded()) {
7240 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7241 Node* ctrl = control();
7242 set_control(top()); // no regular fast path
7243 return ctrl;
7244 }
7245
7246 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7247 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7248 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7249 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7250 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7251
7252 return instof_false; // even if it is NULL
7253 }
7254
7255 //------------------------------get_state_from_digest_object-----------------------
7256 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, const char *state_type) {
7257 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7258 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7259 if (digest_state == NULL) return (Node *) NULL;
7260
7261 // now have the array, need to get the start address of the state array
7262 Node* state = array_element_address(digest_state, intcon(0), T_INT);
7263 return state;
7264 }
7265
7266 //------------------------------get_digest_length_from_sha3_object----------------------------------
7267 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
7268 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
7269 assert (digest_length != NULL, "sanity");
7270 return digest_length;
7271 }
7272
7273 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7274 // Return node representing slow path of predicate check.
7275 // the pseudo code we want to emulate with this predicate is:
7276 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7277 //
7278 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7279 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7280 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7281 assert((uint)predicate < 5, "sanity");
7282
7283 // The receiver was checked for NULL already.
7284 Node* digestBaseObj = argument(0);
7285
7286 // get DigestBase klass for instanceOf check
7287 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7288 assert(tinst != NULL, "digestBaseObj is null");
7289 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7290
7291 const char* klass_name = NULL;
7292 switch (predicate) {
7293 case 0:
7294 if (UseMD5Intrinsics) {
7295 // we want to do an instanceof comparison against the MD5 class
7296 klass_name = "sun/security/provider/MD5";
7297 }
7298 break;
7299 case 1:
7300 if (UseSHA1Intrinsics) {
7301 // we want to do an instanceof comparison against the SHA class
7302 klass_name = "sun/security/provider/SHA";
7303 }
7304 break;
7305 case 2:
7306 if (UseSHA256Intrinsics) {
7307 // we want to do an instanceof comparison against the SHA2 class
7308 klass_name = "sun/security/provider/SHA2";
7309 }
7310 break;
7311 case 3:
7312 if (UseSHA512Intrinsics) {
7313 // we want to do an instanceof comparison against the SHA5 class
7314 klass_name = "sun/security/provider/SHA5";
7315 }
7316 break;
7317 case 4:
7318 if (UseSHA3Intrinsics) {
7319 // we want to do an instanceof comparison against the SHA3 class
7320 klass_name = "sun/security/provider/SHA3";
7321 }
7322 break;
7323 default:
7324 fatal("unknown SHA intrinsic predicate: %d", predicate);
7325 }
7326
7327 ciKlass* klass = NULL;
7328 if (klass_name != NULL) {
7329 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
7330 }
7331 if ((klass == NULL) || !klass->is_loaded()) {
7332 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7333 Node* ctrl = control();
7334 set_control(top()); // no intrinsic path
7335 return ctrl;
7336 }
7337 ciInstanceKlass* instklass = klass->as_instance_klass();
7338
7339 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7340 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7341 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7342 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7343
7344 return instof_false; // even if it is NULL
7345 }
7346
7347 //-------------inline_fma-----------------------------------
7348 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7349 Node *a = NULL;
7350 Node *b = NULL;
7351 Node *c = NULL;
7352 Node* result = NULL;
7353 switch (id) {
7354 case vmIntrinsics::_fmaD:
7355 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7356 // no receiver since it is static method
7357 a = round_double_node(argument(0));
7358 b = round_double_node(argument(2));
7359 c = round_double_node(argument(4));
7360 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7361 break;
7362 case vmIntrinsics::_fmaF:
7363 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7364 a = argument(0);
7365 b = argument(1);
7366 c = argument(2);
7367 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7368 break;
7369 default:
7370 fatal_unexpected_iid(id); break;
7371 }
7372 set_result(result);
7373 return true;
7374 }
7375
7376 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7377 // argument(0) is receiver
7378 Node* codePoint = argument(1);
7379 Node* n = NULL;
7380
7381 switch (id) {
7382 case vmIntrinsics::_isDigit :
7383 n = new DigitNode(control(), codePoint);
7384 break;
7385 case vmIntrinsics::_isLowerCase :
7386 n = new LowerCaseNode(control(), codePoint);
7387 break;
7388 case vmIntrinsics::_isUpperCase :
7389 n = new UpperCaseNode(control(), codePoint);
7390 break;
7391 case vmIntrinsics::_isWhitespace :
7392 n = new WhitespaceNode(control(), codePoint);
7393 break;
7394 default:
7395 fatal_unexpected_iid(id);
7396 }
7397
7398 set_result(_gvn.transform(n));
7399 return true;
7400 }
7401
7402 //------------------------------inline_fp_min_max------------------------------
7403 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7404 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7405
7406 // The intrinsic should be used only when the API branches aren't predictable,
7407 // the last one performing the most important comparison. The following heuristic
7408 // uses the branch statistics to eventually bail out if necessary.
7409
7410 ciMethodData *md = callee()->method_data();
7411
7412 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7413 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7414
7415 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7416 // Bail out if the call-site didn't contribute enough to the statistics.
7417 return false;
7418 }
7419
7420 uint taken = 0, not_taken = 0;
7421
7422 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7423 if (p->is_BranchData()) {
7424 taken = ((ciBranchData*)p)->taken();
7425 not_taken = ((ciBranchData*)p)->not_taken();
7426 }
7427 }
7428
7429 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7430 balance = balance < 0 ? -balance : balance;
7431 if ( balance > 0.2 ) {
7432 // Bail out if the most important branch is predictable enough.
7433 return false;
7434 }
7435 }
7436 */
7437
7438 Node *a = NULL;
7439 Node *b = NULL;
7440 Node *n = NULL;
7441 switch (id) {
7442 case vmIntrinsics::_maxF:
7443 case vmIntrinsics::_minF:
7444 case vmIntrinsics::_maxF_strict:
7445 case vmIntrinsics::_minF_strict:
7446 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7447 a = argument(0);
7448 b = argument(1);
7449 break;
7450 case vmIntrinsics::_maxD:
7451 case vmIntrinsics::_minD:
7452 case vmIntrinsics::_maxD_strict:
7453 case vmIntrinsics::_minD_strict:
7454 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7455 a = round_double_node(argument(0));
7456 b = round_double_node(argument(2));
7457 break;
7458 default:
7459 fatal_unexpected_iid(id);
7460 break;
7461 }
7462 switch (id) {
7463 case vmIntrinsics::_maxF:
7464 case vmIntrinsics::_maxF_strict:
7465 n = new MaxFNode(a, b);
7466 break;
7467 case vmIntrinsics::_minF:
7468 case vmIntrinsics::_minF_strict:
7469 n = new MinFNode(a, b);
7470 break;
7471 case vmIntrinsics::_maxD:
7472 case vmIntrinsics::_maxD_strict:
7473 n = new MaxDNode(a, b);
7474 break;
7475 case vmIntrinsics::_minD:
7476 case vmIntrinsics::_minD_strict:
7477 n = new MinDNode(a, b);
7478 break;
7479 default:
7480 fatal_unexpected_iid(id);
7481 break;
7482 }
7483 set_result(_gvn.transform(n));
7484 return true;
7485 }
7486
7487 bool LibraryCallKit::inline_profileBoolean() {
7488 Node* counts = argument(1);
7489 const TypeAryPtr* ary = NULL;
7490 ciArray* aobj = NULL;
7491 if (counts->is_Con()
7492 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7493 && (aobj = ary->const_oop()->as_array()) != NULL
7494 && (aobj->length() == 2)) {
7495 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7496 jint false_cnt = aobj->element_value(0).as_int();
7497 jint true_cnt = aobj->element_value(1).as_int();
7498
7499 if (C->log() != NULL) {
7500 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7501 false_cnt, true_cnt);
7502 }
7503
7504 if (false_cnt + true_cnt == 0) {
7505 // According to profile, never executed.
7506 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7507 Deoptimization::Action_reinterpret);
7508 return true;
7509 }
7510
7511 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7512 // is a number of each value occurrences.
7513 Node* result = argument(0);
7514 if (false_cnt == 0 || true_cnt == 0) {
7515 // According to profile, one value has been never seen.
7516 int expected_val = (false_cnt == 0) ? 1 : 0;
7517
7518 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7519 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7520
7521 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7522 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7523 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7524
7525 { // Slow path: uncommon trap for never seen value and then reexecute
7526 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7527 // the value has been seen at least once.
7528 PreserveJVMState pjvms(this);
7529 PreserveReexecuteState preexecs(this);
7530 jvms()->set_should_reexecute(true);
7531
7532 set_control(slow_path);
7533 set_i_o(i_o());
7534
7535 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7536 Deoptimization::Action_reinterpret);
7537 }
7538 // The guard for never seen value enables sharpening of the result and
7539 // returning a constant. It allows to eliminate branches on the same value
7540 // later on.
7541 set_control(fast_path);
7542 result = intcon(expected_val);
7543 }
7544 // Stop profiling.
7545 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7546 // By replacing method body with profile data (represented as ProfileBooleanNode
7547 // on IR level) we effectively disable profiling.
7548 // It enables full speed execution once optimized code is generated.
7549 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7550 C->record_for_igvn(profile);
7551 set_result(profile);
7552 return true;
7553 } else {
7554 // Continue profiling.
7555 // Profile data isn't available at the moment. So, execute method's bytecode version.
7556 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7557 // is compiled and counters aren't available since corresponding MethodHandle
7558 // isn't a compile-time constant.
7559 return false;
7560 }
7561 }
7562
7563 bool LibraryCallKit::inline_isCompileConstant() {
7564 Node* n = argument(0);
7565 set_result(n->is_Con() ? intcon(1) : intcon(0));
7566 return true;
7567 }
7568
7569 //------------------------------- inline_getObjectSize --------------------------------------
7570 //
7571 // Calculate the runtime size of the object/array.
7572 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7573 //
7574 bool LibraryCallKit::inline_getObjectSize() {
7575 Node* obj = argument(3);
7576 Node* klass_node = load_object_klass(obj);
7577
7578 jint layout_con = Klass::_lh_neutral_value;
7579 Node* layout_val = get_layout_helper(klass_node, layout_con);
7580 int layout_is_con = (layout_val == NULL);
7581
7582 if (layout_is_con) {
7583 // Layout helper is constant, can figure out things at compile time.
7584
7585 if (Klass::layout_helper_is_instance(layout_con)) {
7586 // Instance case: layout_con contains the size itself.
7587 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7588 set_result(size);
7589 } else {
7590 // Array case: size is round(header + element_size*arraylength).
7591 // Since arraylength is different for every array instance, we have to
7592 // compute the whole thing at runtime.
7593
7594 Node* arr_length = load_array_length(obj);
7595
7596 int round_mask = MinObjAlignmentInBytes - 1;
7597 int hsize = Klass::layout_helper_header_size(layout_con);
7598 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7599
7600 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7601 round_mask = 0; // strength-reduce it if it goes away completely
7602 }
7603 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7604 Node* header_size = intcon(hsize + round_mask);
7605
7606 Node* lengthx = ConvI2X(arr_length);
7607 Node* headerx = ConvI2X(header_size);
7608
7609 Node* abody = lengthx;
7610 if (eshift != 0) {
7611 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7612 }
7613 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7614 if (round_mask != 0) {
7615 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7616 }
7617 size = ConvX2L(size);
7618 set_result(size);
7619 }
7620 } else {
7621 // Layout helper is not constant, need to test for array-ness at runtime.
7622
7623 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7624 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7625 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7626 record_for_igvn(result_reg);
7627
7628 Node* array_ctl = generate_array_guard(klass_node, NULL);
7629 if (array_ctl != NULL) {
7630 // Array case: size is round(header + element_size*arraylength).
7631 // Since arraylength is different for every array instance, we have to
7632 // compute the whole thing at runtime.
7633
7634 PreserveJVMState pjvms(this);
7635 set_control(array_ctl);
7636 Node* arr_length = load_array_length(obj);
7637
7638 int round_mask = MinObjAlignmentInBytes - 1;
7639 Node* mask = intcon(round_mask);
7640
7641 Node* hss = intcon(Klass::_lh_header_size_shift);
7642 Node* hsm = intcon(Klass::_lh_header_size_mask);
7643 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7644 header_size = _gvn.transform(new AndINode(header_size, hsm));
7645 header_size = _gvn.transform(new AddINode(header_size, mask));
7646
7647 // There is no need to mask or shift this value.
7648 // The semantics of LShiftINode include an implicit mask to 0x1F.
7649 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7650 Node* elem_shift = layout_val;
7651
7652 Node* lengthx = ConvI2X(arr_length);
7653 Node* headerx = ConvI2X(header_size);
7654
7655 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7656 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7657 if (round_mask != 0) {
7658 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7659 }
7660 size = ConvX2L(size);
7661
7662 result_reg->init_req(_array_path, control());
7663 result_val->init_req(_array_path, size);
7664 }
7665
7666 if (!stopped()) {
7667 // Instance case: the layout helper gives us instance size almost directly,
7668 // but we need to mask out the _lh_instance_slow_path_bit.
7669 Node* size = ConvI2X(layout_val);
7670 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7671 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7672 size = _gvn.transform(new AndXNode(size, mask));
7673 size = ConvX2L(size);
7674
7675 result_reg->init_req(_instance_path, control());
7676 result_val->init_req(_instance_path, size);
7677 }
7678
7679 set_result(result_reg, result_val);
7680 }
7681
7682 return true;
7683 }
7684
7685 //------------------------------- inline_blackhole --------------------------------------
7686 //
7687 // Make sure all arguments to this node are alive.
7688 // This matches methods that were requested to be blackholed through compile commands.
7689 //
7690 bool LibraryCallKit::inline_blackhole() {
7691 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7692 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7693 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7694
7695 // Bind call arguments as blackhole arguments to keep them alive
7696 Node* bh = insert_mem_bar(Op_Blackhole);
7697 uint nargs = callee()->arg_size();
7698 for (uint i = 0; i < nargs; i++) {
7699 bh->add_req(argument(i));
7700 }
7701
7702 return true;
7703 }