1 /*
2 * Copyright (c) 1999, 2022, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/unsafe.hpp"
54 #include "runtime/objectMonitor.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "utilities/macros.hpp"
58 #include "utilities/powerOfTwo.hpp"
59
60 #if INCLUDE_JFR
61 #include "jfr/jfr.hpp"
62 #endif
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return NULL;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
84 !C->directive()->is_intrinsic_disabled(mh) &&
85 !vmIntrinsics::is_disabled_by_flags(mh);
86
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return NULL;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
123 if (C->print_intrinsics() || C->print_inlining()) {
124 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
125 }
126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
127 if (C->log()) {
128 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
129 vmIntrinsics::name_at(intrinsic_id()),
130 (is_virtual() ? " virtual='1'" : ""),
131 C->unique() - nodes);
132 }
133 // Push the result from the inlined method onto the stack.
134 kit.push_result();
135 C->print_inlining_update(this);
136 return kit.transfer_exceptions_into_jvms();
137 }
138
139 // The intrinsic bailed out
140 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
141 if (jvms->has_method()) {
142 // Not a root compile.
143 const char* msg;
144 if (callee->intrinsic_candidate()) {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
146 } else {
147 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
148 : "failed to inline (intrinsic), method not annotated";
149 }
150 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
151 if (C->print_intrinsics() || C->print_inlining()) {
152 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
153 }
154 } else {
155 // Root compile
156 ResourceMark rm;
157 stringStream msg_stream;
158 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
159 vmIntrinsics::name_at(intrinsic_id()),
160 is_virtual() ? " (virtual)" : "", bci);
161 const char *msg = msg_stream.as_string();
162 log_debug(jit, inlining)("%s", msg);
163 if (C->print_intrinsics() || C->print_inlining()) {
164 tty->print("%s", msg);
165 }
166 }
167 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
168 C->print_inlining_update(this);
169
170 return NULL;
171 }
172
173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
174 LibraryCallKit kit(jvms, this);
175 Compile* C = kit.C;
176 int nodes = C->unique();
177 _last_predicate = predicate;
178 #ifndef PRODUCT
179 assert(is_predicated() && predicate < predicates_count(), "sanity");
180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
181 char buf[1000];
182 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
183 tty->print_cr("Predicate for intrinsic %s", str);
184 }
185 #endif
186 ciMethod* callee = kit.callee();
187 const int bci = kit.bci();
188
189 Node* slow_ctl = kit.try_to_predicate(predicate);
190 if (!kit.failing()) {
191 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
192 : "(intrinsic, predicate)";
193 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
194 if (C->print_intrinsics() || C->print_inlining()) {
195 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
196 }
197 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
198 if (C->log()) {
199 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
200 vmIntrinsics::name_at(intrinsic_id()),
201 (is_virtual() ? " virtual='1'" : ""),
202 C->unique() - nodes);
203 }
204 return slow_ctl; // Could be NULL if the check folds.
205 }
206
207 // The intrinsic bailed out
208 if (jvms->has_method()) {
209 // Not a root compile.
210 const char* msg = "failed to generate predicate for intrinsic";
211 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
212 if (C->print_intrinsics() || C->print_inlining()) {
213 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
214 }
215 } else {
216 // Root compile
217 ResourceMark rm;
218 stringStream msg_stream;
219 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
220 vmIntrinsics::name_at(intrinsic_id()),
221 is_virtual() ? " (virtual)" : "", bci);
222 const char *msg = msg_stream.as_string();
223 log_debug(jit, inlining)("%s", msg);
224 if (C->print_intrinsics() || C->print_inlining()) {
225 C->print_inlining_stream()->print("%s", msg);
226 }
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return NULL;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dexp:
265 case vmIntrinsics::_dlog:
266 case vmIntrinsics::_dlog10:
267 case vmIntrinsics::_dpow:
268 case vmIntrinsics::_dcopySign:
269 case vmIntrinsics::_fcopySign:
270 case vmIntrinsics::_dsignum:
271 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
272
273 case vmIntrinsics::_min:
274 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
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::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
324 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
325 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
326 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
327 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
328 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
329 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
330 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
331 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
332
333 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
334 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
335 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
336 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
337 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
338 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
339 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
340 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
341 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
342
343 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
344 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
345 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
346 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
347 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
348 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
349 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
350 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
351 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
352
353 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
354 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
355 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
356 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
357 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
358 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
359 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
360 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
361 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
362
363 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
364 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
365 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
366 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
367
368 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
374 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
375 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
376 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
377 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
378 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
379 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
380 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
381 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
382
383 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
384 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
385 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
386 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
387 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
388 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
389 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
390 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
391 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
392
393 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
394 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
395 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
396 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
397 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
398 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
399 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
400 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
401 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
402
403 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
404 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
405 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
406 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
407 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
408 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
409 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
410 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
411 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
412
413 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
417 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
418
419 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
420 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
421 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
422 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
423 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
435 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
439
440 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
441 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
442 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
443 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
444 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
445 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
446 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
452 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
453 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
454 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
455
456 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
459 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
460
461 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
465 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
466
467 case vmIntrinsics::_loadFence:
468 case vmIntrinsics::_storeFence:
469 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
470
471 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
472
473 case vmIntrinsics::_currentThread: return inline_native_currentThread();
474
475 #ifdef JFR_HAVE_INTRINSICS
476 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
477 case vmIntrinsics::_getClassId: return inline_native_classID();
478 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
479 #endif
480 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
481 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
482 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
483 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
484 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
485 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
486 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
487 case vmIntrinsics::_getLength: return inline_native_getLength();
488 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
489 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
490 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
491 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
492 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
493 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
494 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
495
496 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
497 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
498
499 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
500
501 case vmIntrinsics::_isInstance:
502 case vmIntrinsics::_getModifiers:
503 case vmIntrinsics::_isInterface:
504 case vmIntrinsics::_isArray:
505 case vmIntrinsics::_isPrimitive:
506 case vmIntrinsics::_isHidden:
507 case vmIntrinsics::_getSuperclass:
508 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
509
510 case vmIntrinsics::_floatToRawIntBits:
511 case vmIntrinsics::_floatToIntBits:
512 case vmIntrinsics::_intBitsToFloat:
513 case vmIntrinsics::_doubleToRawLongBits:
514 case vmIntrinsics::_doubleToLongBits:
515 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
516
517 case vmIntrinsics::_numberOfLeadingZeros_i:
518 case vmIntrinsics::_numberOfLeadingZeros_l:
519 case vmIntrinsics::_numberOfTrailingZeros_i:
520 case vmIntrinsics::_numberOfTrailingZeros_l:
521 case vmIntrinsics::_bitCount_i:
522 case vmIntrinsics::_bitCount_l:
523 case vmIntrinsics::_reverseBytes_i:
524 case vmIntrinsics::_reverseBytes_l:
525 case vmIntrinsics::_reverseBytes_s:
526 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
527
528 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
529
530 case vmIntrinsics::_Reference_get: return inline_reference_get();
531 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
532 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
533
534 case vmIntrinsics::_Class_cast: return inline_Class_cast();
535
536 case vmIntrinsics::_aescrypt_encryptBlock:
537 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
538
539 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
540 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
541 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
542
543 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
544 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
545 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
546
547 case vmIntrinsics::_counterMode_AESCrypt:
548 return inline_counterMode_AESCrypt(intrinsic_id());
549
550 case vmIntrinsics::_md5_implCompress:
551 case vmIntrinsics::_sha_implCompress:
552 case vmIntrinsics::_sha2_implCompress:
553 case vmIntrinsics::_sha5_implCompress:
554 case vmIntrinsics::_sha3_implCompress:
555 return inline_digestBase_implCompress(intrinsic_id());
556
557 case vmIntrinsics::_digestBase_implCompressMB:
558 return inline_digestBase_implCompressMB(predicate);
559
560 case vmIntrinsics::_multiplyToLen:
561 return inline_multiplyToLen();
562
563 case vmIntrinsics::_squareToLen:
564 return inline_squareToLen();
565
566 case vmIntrinsics::_mulAdd:
567 return inline_mulAdd();
568
569 case vmIntrinsics::_montgomeryMultiply:
570 return inline_montgomeryMultiply();
571 case vmIntrinsics::_montgomerySquare:
572 return inline_montgomerySquare();
573
574 case vmIntrinsics::_bigIntegerRightShiftWorker:
575 return inline_bigIntegerShift(true);
576 case vmIntrinsics::_bigIntegerLeftShiftWorker:
577 return inline_bigIntegerShift(false);
578
579 case vmIntrinsics::_vectorizedMismatch:
580 return inline_vectorizedMismatch();
581
582 case vmIntrinsics::_ghash_processBlocks:
583 return inline_ghash_processBlocks();
584 case vmIntrinsics::_base64_encodeBlock:
585 return inline_base64_encodeBlock();
586 case vmIntrinsics::_base64_decodeBlock:
587 return inline_base64_decodeBlock();
588
589 case vmIntrinsics::_encodeISOArray:
590 case vmIntrinsics::_encodeByteISOArray:
591 return inline_encodeISOArray(false);
592 case vmIntrinsics::_encodeAsciiArray:
593 return inline_encodeISOArray(true);
594
595 case vmIntrinsics::_updateCRC32:
596 return inline_updateCRC32();
597 case vmIntrinsics::_updateBytesCRC32:
598 return inline_updateBytesCRC32();
599 case vmIntrinsics::_updateByteBufferCRC32:
600 return inline_updateByteBufferCRC32();
601
602 case vmIntrinsics::_updateBytesCRC32C:
603 return inline_updateBytesCRC32C();
604 case vmIntrinsics::_updateDirectByteBufferCRC32C:
605 return inline_updateDirectByteBufferCRC32C();
606
607 case vmIntrinsics::_updateBytesAdler32:
608 return inline_updateBytesAdler32();
609 case vmIntrinsics::_updateByteBufferAdler32:
610 return inline_updateByteBufferAdler32();
611
612 case vmIntrinsics::_profileBoolean:
613 return inline_profileBoolean();
614 case vmIntrinsics::_isCompileConstant:
615 return inline_isCompileConstant();
616
617 case vmIntrinsics::_hasNegatives:
618 return inline_hasNegatives();
619
620 case vmIntrinsics::_fmaD:
621 case vmIntrinsics::_fmaF:
622 return inline_fma(intrinsic_id());
623
624 case vmIntrinsics::_isDigit:
625 case vmIntrinsics::_isLowerCase:
626 case vmIntrinsics::_isUpperCase:
627 case vmIntrinsics::_isWhitespace:
628 return inline_character_compare(intrinsic_id());
629
630 case vmIntrinsics::_maxF:
631 case vmIntrinsics::_minF:
632 case vmIntrinsics::_maxD:
633 case vmIntrinsics::_minD:
634 return inline_fp_min_max(intrinsic_id());
635
636 case vmIntrinsics::_VectorUnaryOp:
637 return inline_vector_nary_operation(1);
638 case vmIntrinsics::_VectorBinaryOp:
639 return inline_vector_nary_operation(2);
640 case vmIntrinsics::_VectorTernaryOp:
641 return inline_vector_nary_operation(3);
642 case vmIntrinsics::_VectorBroadcastCoerced:
643 return inline_vector_broadcast_coerced();
644 case vmIntrinsics::_VectorShuffleIota:
645 return inline_vector_shuffle_iota();
646 case vmIntrinsics::_VectorMaskOp:
647 return inline_vector_mask_operation();
648 case vmIntrinsics::_VectorShuffleToVector:
649 return inline_vector_shuffle_to_vector();
650 case vmIntrinsics::_VectorLoadOp:
651 return inline_vector_mem_operation(/*is_store=*/false);
652 case vmIntrinsics::_VectorStoreOp:
653 return inline_vector_mem_operation(/*is_store=*/true);
654 case vmIntrinsics::_VectorGatherOp:
655 return inline_vector_gather_scatter(/*is_scatter*/ false);
656 case vmIntrinsics::_VectorScatterOp:
657 return inline_vector_gather_scatter(/*is_scatter*/ true);
658 case vmIntrinsics::_VectorReductionCoerced:
659 return inline_vector_reduction();
660 case vmIntrinsics::_VectorTest:
661 return inline_vector_test();
662 case vmIntrinsics::_VectorBlend:
663 return inline_vector_blend();
664 case vmIntrinsics::_VectorRearrange:
665 return inline_vector_rearrange();
666 case vmIntrinsics::_VectorCompare:
667 return inline_vector_compare();
668 case vmIntrinsics::_VectorBroadcastInt:
669 return inline_vector_broadcast_int();
670 case vmIntrinsics::_VectorConvert:
671 return inline_vector_convert();
672 case vmIntrinsics::_VectorInsert:
673 return inline_vector_insert();
674 case vmIntrinsics::_VectorExtract:
675 return inline_vector_extract();
676
677 case vmIntrinsics::_getObjectSize:
678 return inline_getObjectSize();
679
680 case vmIntrinsics::_blackhole:
681 return inline_blackhole();
682
683 default:
684 // If you get here, it may be that someone has added a new intrinsic
685 // to the list in vmIntrinsics.hpp without implementing it here.
686 #ifndef PRODUCT
687 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
688 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
689 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
690 }
691 #endif
692 return false;
693 }
694 }
695
696 Node* LibraryCallKit::try_to_predicate(int predicate) {
697 if (!jvms()->has_method()) {
698 // Root JVMState has a null method.
699 assert(map()->memory()->Opcode() == Op_Parm, "");
700 // Insert the memory aliasing node
701 set_all_memory(reset_memory());
702 }
703 assert(merged_memory(), "");
704
705 switch (intrinsic_id()) {
706 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
707 return inline_cipherBlockChaining_AESCrypt_predicate(false);
708 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
709 return inline_cipherBlockChaining_AESCrypt_predicate(true);
710 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
711 return inline_electronicCodeBook_AESCrypt_predicate(false);
712 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
713 return inline_electronicCodeBook_AESCrypt_predicate(true);
714 case vmIntrinsics::_counterMode_AESCrypt:
715 return inline_counterMode_AESCrypt_predicate();
716 case vmIntrinsics::_digestBase_implCompressMB:
717 return inline_digestBase_implCompressMB_predicate(predicate);
718
719 default:
720 // If you get here, it may be that someone has added a new intrinsic
721 // to the list in vmIntrinsics.hpp without implementing it here.
722 #ifndef PRODUCT
723 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
724 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
725 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
726 }
727 #endif
728 Node* slow_ctl = control();
729 set_control(top()); // No fast path instrinsic
730 return slow_ctl;
731 }
732 }
733
734 //------------------------------set_result-------------------------------
735 // Helper function for finishing intrinsics.
736 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
737 record_for_igvn(region);
738 set_control(_gvn.transform(region));
739 set_result( _gvn.transform(value));
740 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
741 }
742
743 //------------------------------generate_guard---------------------------
744 // Helper function for generating guarded fast-slow graph structures.
745 // The given 'test', if true, guards a slow path. If the test fails
746 // then a fast path can be taken. (We generally hope it fails.)
747 // In all cases, GraphKit::control() is updated to the fast path.
748 // The returned value represents the control for the slow path.
749 // The return value is never 'top'; it is either a valid control
750 // or NULL if it is obvious that the slow path can never be taken.
751 // Also, if region and the slow control are not NULL, the slow edge
752 // is appended to the region.
753 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
754 if (stopped()) {
755 // Already short circuited.
756 return NULL;
757 }
758
759 // Build an if node and its projections.
760 // If test is true we take the slow path, which we assume is uncommon.
761 if (_gvn.type(test) == TypeInt::ZERO) {
762 // The slow branch is never taken. No need to build this guard.
763 return NULL;
764 }
765
766 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
767
768 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
769 if (if_slow == top()) {
770 // The slow branch is never taken. No need to build this guard.
771 return NULL;
772 }
773
774 if (region != NULL)
775 region->add_req(if_slow);
776
777 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
778 set_control(if_fast);
779
780 return if_slow;
781 }
782
783 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
784 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
785 }
786 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
787 return generate_guard(test, region, PROB_FAIR);
788 }
789
790 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
791 Node* *pos_index) {
792 if (stopped())
793 return NULL; // already stopped
794 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
795 return NULL; // index is already adequately typed
796 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
797 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
798 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
799 if (is_neg != NULL && pos_index != NULL) {
800 // Emulate effect of Parse::adjust_map_after_if.
801 Node* ccast = new CastIINode(index, TypeInt::POS);
802 ccast->set_req(0, control());
803 (*pos_index) = _gvn.transform(ccast);
804 }
805 return is_neg;
806 }
807
808 // Make sure that 'position' is a valid limit index, in [0..length].
809 // There are two equivalent plans for checking this:
810 // A. (offset + copyLength) unsigned<= arrayLength
811 // B. offset <= (arrayLength - copyLength)
812 // We require that all of the values above, except for the sum and
813 // difference, are already known to be non-negative.
814 // Plan A is robust in the face of overflow, if offset and copyLength
815 // are both hugely positive.
816 //
817 // Plan B is less direct and intuitive, but it does not overflow at
818 // all, since the difference of two non-negatives is always
819 // representable. Whenever Java methods must perform the equivalent
820 // check they generally use Plan B instead of Plan A.
821 // For the moment we use Plan A.
822 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
823 Node* subseq_length,
824 Node* array_length,
825 RegionNode* region) {
826 if (stopped())
827 return NULL; // already stopped
828 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
829 if (zero_offset && subseq_length->eqv_uncast(array_length))
830 return NULL; // common case of whole-array copy
831 Node* last = subseq_length;
832 if (!zero_offset) // last += offset
833 last = _gvn.transform(new AddINode(last, offset));
834 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
835 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
836 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
837 return is_over;
838 }
839
840 // Emit range checks for the given String.value byte array
841 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
842 if (stopped()) {
843 return; // already stopped
844 }
845 RegionNode* bailout = new RegionNode(1);
846 record_for_igvn(bailout);
847 if (char_count) {
848 // Convert char count to byte count
849 count = _gvn.transform(new LShiftINode(count, intcon(1)));
850 }
851
852 // Offset and count must not be negative
853 generate_negative_guard(offset, bailout);
854 generate_negative_guard(count, bailout);
855 // Offset + count must not exceed length of array
856 generate_limit_guard(offset, count, load_array_length(array), bailout);
857
858 if (bailout->req() > 1) {
859 PreserveJVMState pjvms(this);
860 set_control(_gvn.transform(bailout));
861 uncommon_trap(Deoptimization::Reason_intrinsic,
862 Deoptimization::Action_maybe_recompile);
863 }
864 }
865
866 //--------------------------generate_current_thread--------------------
867 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
868 ciKlass* thread_klass = env()->Thread_klass();
869 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
870 Node* thread = _gvn.transform(new ThreadLocalNode());
871 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
872 tls_output = thread;
873 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
874 thread_obj_handle = _gvn.transform(thread_obj_handle);
875 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
876 }
877
878
879 //------------------------------make_string_method_node------------------------
880 // Helper method for String intrinsic functions. This version is called with
881 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
882 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
883 // containing the lengths of str1 and str2.
884 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
885 Node* result = NULL;
886 switch (opcode) {
887 case Op_StrIndexOf:
888 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
889 str1_start, cnt1, str2_start, cnt2, ae);
890 break;
891 case Op_StrComp:
892 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
893 str1_start, cnt1, str2_start, cnt2, ae);
894 break;
895 case Op_StrEquals:
896 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
897 // Use the constant length if there is one because optimized match rule may exist.
898 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
899 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
900 break;
901 default:
902 ShouldNotReachHere();
903 return NULL;
904 }
905
906 // All these intrinsics have checks.
907 C->set_has_split_ifs(true); // Has chance for split-if optimization
908 clear_upper_avx();
909
910 return _gvn.transform(result);
911 }
912
913 //------------------------------inline_string_compareTo------------------------
914 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
915 Node* arg1 = argument(0);
916 Node* arg2 = argument(1);
917
918 arg1 = must_be_not_null(arg1, true);
919 arg2 = must_be_not_null(arg2, true);
920
921 // Get start addr and length of first argument
922 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
923 Node* arg1_cnt = load_array_length(arg1);
924
925 // Get start addr and length of second argument
926 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
927 Node* arg2_cnt = load_array_length(arg2);
928
929 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
930 set_result(result);
931 return true;
932 }
933
934 //------------------------------inline_string_equals------------------------
935 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
936 Node* arg1 = argument(0);
937 Node* arg2 = argument(1);
938
939 // paths (plus control) merge
940 RegionNode* region = new RegionNode(3);
941 Node* phi = new PhiNode(region, TypeInt::BOOL);
942
943 if (!stopped()) {
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 // Check for arg1_cnt != arg2_cnt
957 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
958 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
959 Node* if_ne = generate_slow_guard(bol, NULL);
960 if (if_ne != NULL) {
961 phi->init_req(2, intcon(0));
962 region->init_req(2, if_ne);
963 }
964
965 // Check for count == 0 is done by assembler code for StrEquals.
966
967 if (!stopped()) {
968 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
969 phi->init_req(1, equals);
970 region->init_req(1, control());
971 }
972 }
973
974 // post merge
975 set_control(_gvn.transform(region));
976 record_for_igvn(region);
977
978 set_result(_gvn.transform(phi));
979 return true;
980 }
981
982 //------------------------------inline_array_equals----------------------------
983 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
984 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
985 Node* arg1 = argument(0);
986 Node* arg2 = argument(1);
987
988 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
989 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
990 clear_upper_avx();
991
992 return true;
993 }
994
995 //------------------------------inline_hasNegatives------------------------------
996 bool LibraryCallKit::inline_hasNegatives() {
997 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
998 return false;
999 }
1000
1001 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1002 // no receiver since it is static method
1003 Node* ba = argument(0);
1004 Node* offset = argument(1);
1005 Node* len = argument(2);
1006
1007 ba = must_be_not_null(ba, true);
1008
1009 // Range checks
1010 generate_string_range_check(ba, offset, len, false);
1011 if (stopped()) {
1012 return true;
1013 }
1014 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1015 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1016 set_result(_gvn.transform(result));
1017 return true;
1018 }
1019
1020 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1021 Node* index = argument(0);
1022 Node* length = bt == T_INT ? argument(1) : argument(2);
1023 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1024 return false;
1025 }
1026
1027 // check that length is positive
1028 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1029 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1030
1031 {
1032 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1033 uncommon_trap(Deoptimization::Reason_intrinsic,
1034 Deoptimization::Action_make_not_entrant);
1035 }
1036
1037 if (stopped()) {
1038 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1039 return true;
1040 }
1041
1042 // length is now known postive, add a cast node to make this explicit
1043 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1044 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1045 casted_length = _gvn.transform(casted_length);
1046 replace_in_map(length, casted_length);
1047 length = casted_length;
1048
1049 // Use an unsigned comparison for the range check itself
1050 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1051 BoolTest::mask btest = BoolTest::lt;
1052 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1053 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1054 _gvn.set_type(rc, rc->Value(&_gvn));
1055 if (!rc_bool->is_Con()) {
1056 record_for_igvn(rc);
1057 }
1058 set_control(_gvn.transform(new IfTrueNode(rc)));
1059 {
1060 PreserveJVMState pjvms(this);
1061 set_control(_gvn.transform(new IfFalseNode(rc)));
1062 uncommon_trap(Deoptimization::Reason_range_check,
1063 Deoptimization::Action_make_not_entrant);
1064 }
1065
1066 if (stopped()) {
1067 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1068 return true;
1069 }
1070
1071 // index is now known to be >= 0 and < length, cast it
1072 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1073 result = _gvn.transform(result);
1074 set_result(result);
1075 replace_in_map(index, result);
1076 return true;
1077 }
1078
1079 //------------------------------inline_string_indexOf------------------------
1080 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1081 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1082 return false;
1083 }
1084 Node* src = argument(0);
1085 Node* tgt = argument(1);
1086
1087 // Make the merge point
1088 RegionNode* result_rgn = new RegionNode(4);
1089 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1090
1091 src = must_be_not_null(src, true);
1092 tgt = must_be_not_null(tgt, true);
1093
1094 // Get start addr and length of source string
1095 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1096 Node* src_count = load_array_length(src);
1097
1098 // Get start addr and length of substring
1099 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1100 Node* tgt_count = load_array_length(tgt);
1101
1102 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1103 // Divide src size by 2 if String is UTF16 encoded
1104 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1105 }
1106 if (ae == StrIntrinsicNode::UU) {
1107 // Divide substring size by 2 if String is UTF16 encoded
1108 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1109 }
1110
1111 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1112 if (result != NULL) {
1113 result_phi->init_req(3, result);
1114 result_rgn->init_req(3, control());
1115 }
1116 set_control(_gvn.transform(result_rgn));
1117 record_for_igvn(result_rgn);
1118 set_result(_gvn.transform(result_phi));
1119
1120 return true;
1121 }
1122
1123 //-----------------------------inline_string_indexOf-----------------------
1124 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1125 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1126 return false;
1127 }
1128 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1129 return false;
1130 }
1131 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1132 Node* src = argument(0); // byte[]
1133 Node* src_count = argument(1); // char count
1134 Node* tgt = argument(2); // byte[]
1135 Node* tgt_count = argument(3); // char count
1136 Node* from_index = argument(4); // char index
1137
1138 src = must_be_not_null(src, true);
1139 tgt = must_be_not_null(tgt, true);
1140
1141 // Multiply byte array index by 2 if String is UTF16 encoded
1142 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1143 src_count = _gvn.transform(new SubINode(src_count, from_index));
1144 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1145 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1146
1147 // Range checks
1148 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1149 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1150 if (stopped()) {
1151 return true;
1152 }
1153
1154 RegionNode* region = new RegionNode(5);
1155 Node* phi = new PhiNode(region, TypeInt::INT);
1156
1157 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1158 if (result != NULL) {
1159 // The result is index relative to from_index if substring was found, -1 otherwise.
1160 // Generate code which will fold into cmove.
1161 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1162 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1163
1164 Node* if_lt = generate_slow_guard(bol, NULL);
1165 if (if_lt != NULL) {
1166 // result == -1
1167 phi->init_req(3, result);
1168 region->init_req(3, if_lt);
1169 }
1170 if (!stopped()) {
1171 result = _gvn.transform(new AddINode(result, from_index));
1172 phi->init_req(4, result);
1173 region->init_req(4, control());
1174 }
1175 }
1176
1177 set_control(_gvn.transform(region));
1178 record_for_igvn(region);
1179 set_result(_gvn.transform(phi));
1180 clear_upper_avx();
1181
1182 return true;
1183 }
1184
1185 // Create StrIndexOfNode with fast path checks
1186 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1187 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1188 // Check for substr count > string count
1189 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1190 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1191 Node* if_gt = generate_slow_guard(bol, NULL);
1192 if (if_gt != NULL) {
1193 phi->init_req(1, intcon(-1));
1194 region->init_req(1, if_gt);
1195 }
1196 if (!stopped()) {
1197 // Check for substr count == 0
1198 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1199 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1200 Node* if_zero = generate_slow_guard(bol, NULL);
1201 if (if_zero != NULL) {
1202 phi->init_req(2, intcon(0));
1203 region->init_req(2, if_zero);
1204 }
1205 }
1206 if (!stopped()) {
1207 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1208 }
1209 return NULL;
1210 }
1211
1212 //-----------------------------inline_string_indexOfChar-----------------------
1213 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1214 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1215 return false;
1216 }
1217 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1218 return false;
1219 }
1220 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1221 Node* src = argument(0); // byte[]
1222 Node* tgt = argument(1); // tgt is int ch
1223 Node* from_index = argument(2);
1224 Node* max = argument(3);
1225
1226 src = must_be_not_null(src, true);
1227
1228 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1229 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1230 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1231
1232 // Range checks
1233 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1234 if (stopped()) {
1235 return true;
1236 }
1237
1238 RegionNode* region = new RegionNode(3);
1239 Node* phi = new PhiNode(region, TypeInt::INT);
1240
1241 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1242 C->set_has_split_ifs(true); // Has chance for split-if optimization
1243 _gvn.transform(result);
1244
1245 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1246 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1247
1248 Node* if_lt = generate_slow_guard(bol, NULL);
1249 if (if_lt != NULL) {
1250 // result == -1
1251 phi->init_req(2, result);
1252 region->init_req(2, if_lt);
1253 }
1254 if (!stopped()) {
1255 result = _gvn.transform(new AddINode(result, from_index));
1256 phi->init_req(1, result);
1257 region->init_req(1, control());
1258 }
1259 set_control(_gvn.transform(region));
1260 record_for_igvn(region);
1261 set_result(_gvn.transform(phi));
1262
1263 return true;
1264 }
1265 //---------------------------inline_string_copy---------------------
1266 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1267 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1268 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1269 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1270 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1271 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1272 bool LibraryCallKit::inline_string_copy(bool compress) {
1273 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1274 return false;
1275 }
1276 int nargs = 5; // 2 oops, 3 ints
1277 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1278
1279 Node* src = argument(0);
1280 Node* src_offset = argument(1);
1281 Node* dst = argument(2);
1282 Node* dst_offset = argument(3);
1283 Node* length = argument(4);
1284
1285 // Check for allocation before we add nodes that would confuse
1286 // tightly_coupled_allocation()
1287 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1288
1289 // Figure out the size and type of the elements we will be copying.
1290 const Type* src_type = src->Value(&_gvn);
1291 const Type* dst_type = dst->Value(&_gvn);
1292 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1293 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1294 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1295 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1296 "Unsupported array types for inline_string_copy");
1297
1298 src = must_be_not_null(src, true);
1299 dst = must_be_not_null(dst, true);
1300
1301 // Convert char[] offsets to byte[] offsets
1302 bool convert_src = (compress && src_elem == T_BYTE);
1303 bool convert_dst = (!compress && dst_elem == T_BYTE);
1304 if (convert_src) {
1305 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1306 } else if (convert_dst) {
1307 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1308 }
1309
1310 // Range checks
1311 generate_string_range_check(src, src_offset, length, convert_src);
1312 generate_string_range_check(dst, dst_offset, length, convert_dst);
1313 if (stopped()) {
1314 return true;
1315 }
1316
1317 Node* src_start = array_element_address(src, src_offset, src_elem);
1318 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1319 // 'src_start' points to src array + scaled offset
1320 // 'dst_start' points to dst array + scaled offset
1321 Node* count = NULL;
1322 if (compress) {
1323 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1324 } else {
1325 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1326 }
1327
1328 if (alloc != NULL) {
1329 if (alloc->maybe_set_complete(&_gvn)) {
1330 // "You break it, you buy it."
1331 InitializeNode* init = alloc->initialization();
1332 assert(init->is_complete(), "we just did this");
1333 init->set_complete_with_arraycopy();
1334 assert(dst->is_CheckCastPP(), "sanity");
1335 assert(dst->in(0)->in(0) == init, "dest pinned");
1336 }
1337 // Do not let stores that initialize this object be reordered with
1338 // a subsequent store that would make this object accessible by
1339 // other threads.
1340 // Record what AllocateNode this StoreStore protects so that
1341 // escape analysis can go from the MemBarStoreStoreNode to the
1342 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1343 // based on the escape status of the AllocateNode.
1344 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1345 }
1346 if (compress) {
1347 set_result(_gvn.transform(count));
1348 }
1349 clear_upper_avx();
1350
1351 return true;
1352 }
1353
1354 #ifdef _LP64
1355 #define XTOP ,top() /*additional argument*/
1356 #else //_LP64
1357 #define XTOP /*no additional argument*/
1358 #endif //_LP64
1359
1360 //------------------------inline_string_toBytesU--------------------------
1361 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1362 bool LibraryCallKit::inline_string_toBytesU() {
1363 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1364 return false;
1365 }
1366 // Get the arguments.
1367 Node* value = argument(0);
1368 Node* offset = argument(1);
1369 Node* length = argument(2);
1370
1371 Node* newcopy = NULL;
1372
1373 // Set the original stack and the reexecute bit for the interpreter to reexecute
1374 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1375 { PreserveReexecuteState preexecs(this);
1376 jvms()->set_should_reexecute(true);
1377
1378 // Check if a null path was taken unconditionally.
1379 value = null_check(value);
1380
1381 RegionNode* bailout = new RegionNode(1);
1382 record_for_igvn(bailout);
1383
1384 // Range checks
1385 generate_negative_guard(offset, bailout);
1386 generate_negative_guard(length, bailout);
1387 generate_limit_guard(offset, length, load_array_length(value), bailout);
1388 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1389 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1390
1391 if (bailout->req() > 1) {
1392 PreserveJVMState pjvms(this);
1393 set_control(_gvn.transform(bailout));
1394 uncommon_trap(Deoptimization::Reason_intrinsic,
1395 Deoptimization::Action_maybe_recompile);
1396 }
1397 if (stopped()) {
1398 return true;
1399 }
1400
1401 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1402 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1403 newcopy = new_array(klass_node, size, 0); // no arguments to push
1404 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1405 guarantee(alloc != NULL, "created above");
1406
1407 // Calculate starting addresses.
1408 Node* src_start = array_element_address(value, offset, T_CHAR);
1409 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1410
1411 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1412 const TypeInt* toffset = gvn().type(offset)->is_int();
1413 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1414
1415 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1416 const char* copyfunc_name = "arraycopy";
1417 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1418 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1419 OptoRuntime::fast_arraycopy_Type(),
1420 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1421 src_start, dst_start, ConvI2X(length) XTOP);
1422 // Do not let reads from the cloned object float above the arraycopy.
1423 if (alloc->maybe_set_complete(&_gvn)) {
1424 // "You break it, you buy it."
1425 InitializeNode* init = alloc->initialization();
1426 assert(init->is_complete(), "we just did this");
1427 init->set_complete_with_arraycopy();
1428 assert(newcopy->is_CheckCastPP(), "sanity");
1429 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1430 }
1431 // Do not let stores that initialize this object be reordered with
1432 // a subsequent store that would make this object accessible by
1433 // other threads.
1434 // Record what AllocateNode this StoreStore protects so that
1435 // escape analysis can go from the MemBarStoreStoreNode to the
1436 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1437 // based on the escape status of the AllocateNode.
1438 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1439 } // original reexecute is set back here
1440
1441 C->set_has_split_ifs(true); // Has chance for split-if optimization
1442 if (!stopped()) {
1443 set_result(newcopy);
1444 }
1445 clear_upper_avx();
1446
1447 return true;
1448 }
1449
1450 //------------------------inline_string_getCharsU--------------------------
1451 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1452 bool LibraryCallKit::inline_string_getCharsU() {
1453 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1454 return false;
1455 }
1456
1457 // Get the arguments.
1458 Node* src = argument(0);
1459 Node* src_begin = argument(1);
1460 Node* src_end = argument(2); // exclusive offset (i < src_end)
1461 Node* dst = argument(3);
1462 Node* dst_begin = argument(4);
1463
1464 // Check for allocation before we add nodes that would confuse
1465 // tightly_coupled_allocation()
1466 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1467
1468 // Check if a null path was taken unconditionally.
1469 src = null_check(src);
1470 dst = null_check(dst);
1471 if (stopped()) {
1472 return true;
1473 }
1474
1475 // Get length and convert char[] offset to byte[] offset
1476 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1477 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1478
1479 // Range checks
1480 generate_string_range_check(src, src_begin, length, true);
1481 generate_string_range_check(dst, dst_begin, length, false);
1482 if (stopped()) {
1483 return true;
1484 }
1485
1486 if (!stopped()) {
1487 // Calculate starting addresses.
1488 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1489 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1490
1491 // Check if array addresses are aligned to HeapWordSize
1492 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1493 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1494 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1495 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1496
1497 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1498 const char* copyfunc_name = "arraycopy";
1499 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1500 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1501 OptoRuntime::fast_arraycopy_Type(),
1502 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1503 src_start, dst_start, ConvI2X(length) XTOP);
1504 // Do not let reads from the cloned object float above the arraycopy.
1505 if (alloc != NULL) {
1506 if (alloc->maybe_set_complete(&_gvn)) {
1507 // "You break it, you buy it."
1508 InitializeNode* init = alloc->initialization();
1509 assert(init->is_complete(), "we just did this");
1510 init->set_complete_with_arraycopy();
1511 assert(dst->is_CheckCastPP(), "sanity");
1512 assert(dst->in(0)->in(0) == init, "dest pinned");
1513 }
1514 // Do not let stores that initialize this object be reordered with
1515 // a subsequent store that would make this object accessible by
1516 // other threads.
1517 // Record what AllocateNode this StoreStore protects so that
1518 // escape analysis can go from the MemBarStoreStoreNode to the
1519 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1520 // based on the escape status of the AllocateNode.
1521 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1522 } else {
1523 insert_mem_bar(Op_MemBarCPUOrder);
1524 }
1525 }
1526
1527 C->set_has_split_ifs(true); // Has chance for split-if optimization
1528 return true;
1529 }
1530
1531 //----------------------inline_string_char_access----------------------------
1532 // Store/Load char to/from byte[] array.
1533 // static void StringUTF16.putChar(byte[] val, int index, int c)
1534 // static char StringUTF16.getChar(byte[] val, int index)
1535 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1536 Node* value = argument(0);
1537 Node* index = argument(1);
1538 Node* ch = is_store ? argument(2) : NULL;
1539
1540 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1541 // correctly requires matched array shapes.
1542 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1543 "sanity: byte[] and char[] bases agree");
1544 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1545 "sanity: byte[] and char[] scales agree");
1546
1547 // Bail when getChar over constants is requested: constant folding would
1548 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1549 // Java method would constant fold nicely instead.
1550 if (!is_store && value->is_Con() && index->is_Con()) {
1551 return false;
1552 }
1553
1554 // Save state and restore on bailout
1555 uint old_sp = sp();
1556 SafePointNode* old_map = clone_map();
1557
1558 value = must_be_not_null(value, true);
1559
1560 Node* adr = array_element_address(value, index, T_CHAR);
1561 if (adr->is_top()) {
1562 set_map(old_map);
1563 set_sp(old_sp);
1564 return false;
1565 }
1566 old_map->destruct(&_gvn);
1567 if (is_store) {
1568 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1569 } else {
1570 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
1571 set_result(ch);
1572 }
1573 return true;
1574 }
1575
1576 //--------------------------round_double_node--------------------------------
1577 // Round a double node if necessary.
1578 Node* LibraryCallKit::round_double_node(Node* n) {
1579 if (Matcher::strict_fp_requires_explicit_rounding) {
1580 #ifdef IA32
1581 if (UseSSE < 2) {
1582 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1583 }
1584 #else
1585 Unimplemented();
1586 #endif // IA32
1587 }
1588 return n;
1589 }
1590
1591 //------------------------------inline_math-----------------------------------
1592 // public static double Math.abs(double)
1593 // public static double Math.sqrt(double)
1594 // public static double Math.log(double)
1595 // public static double Math.log10(double)
1596 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1597 Node* arg = round_double_node(argument(0));
1598 Node* n = NULL;
1599 switch (id) {
1600 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1601 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1602 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1603 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1604 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1605 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1606 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1607 default: fatal_unexpected_iid(id); break;
1608 }
1609 set_result(_gvn.transform(n));
1610 return true;
1611 }
1612
1613 //------------------------------inline_math-----------------------------------
1614 // public static float Math.abs(float)
1615 // public static int Math.abs(int)
1616 // public static long Math.abs(long)
1617 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1618 Node* arg = argument(0);
1619 Node* n = NULL;
1620 switch (id) {
1621 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1622 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1623 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1624 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1625 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1626 default: fatal_unexpected_iid(id); break;
1627 }
1628 set_result(_gvn.transform(n));
1629 return true;
1630 }
1631
1632 //------------------------------runtime_math-----------------------------
1633 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1634 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1635 "must be (DD)D or (D)D type");
1636
1637 // Inputs
1638 Node* a = round_double_node(argument(0));
1639 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1640
1641 const TypePtr* no_memory_effects = NULL;
1642 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1643 no_memory_effects,
1644 a, top(), b, b ? top() : NULL);
1645 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1646 #ifdef ASSERT
1647 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1648 assert(value_top == top(), "second value must be top");
1649 #endif
1650
1651 set_result(value);
1652 return true;
1653 }
1654
1655 //------------------------------inline_math_pow-----------------------------
1656 bool LibraryCallKit::inline_math_pow() {
1657 Node* exp = round_double_node(argument(2));
1658 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1659 if (d != NULL) {
1660 if (d->getd() == 2.0) {
1661 // Special case: pow(x, 2.0) => x * x
1662 Node* base = round_double_node(argument(0));
1663 set_result(_gvn.transform(new MulDNode(base, base)));
1664 return true;
1665 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1666 // Special case: pow(x, 0.5) => sqrt(x)
1667 Node* base = round_double_node(argument(0));
1668 Node* zero = _gvn.zerocon(T_DOUBLE);
1669
1670 RegionNode* region = new RegionNode(3);
1671 Node* phi = new PhiNode(region, Type::DOUBLE);
1672
1673 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1674 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1675 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1676 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1677 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1678
1679 Node* if_pow = generate_slow_guard(test, NULL);
1680 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1681 phi->init_req(1, value_sqrt);
1682 region->init_req(1, control());
1683
1684 if (if_pow != NULL) {
1685 set_control(if_pow);
1686 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1687 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1688 const TypePtr* no_memory_effects = NULL;
1689 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1690 no_memory_effects, base, top(), exp, top());
1691 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1692 #ifdef ASSERT
1693 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1694 assert(value_top == top(), "second value must be top");
1695 #endif
1696 phi->init_req(2, value_pow);
1697 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1698 }
1699
1700 C->set_has_split_ifs(true); // Has chance for split-if optimization
1701 set_control(_gvn.transform(region));
1702 record_for_igvn(region);
1703 set_result(_gvn.transform(phi));
1704
1705 return true;
1706 }
1707 }
1708
1709 return StubRoutines::dpow() != NULL ?
1710 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1711 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1712 }
1713
1714 //------------------------------inline_math_native-----------------------------
1715 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1716 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1717 switch (id) {
1718 // These intrinsics are not properly supported on all hardware
1719 case vmIntrinsics::_dsin:
1720 return StubRoutines::dsin() != NULL ?
1721 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1722 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1723 case vmIntrinsics::_dcos:
1724 return StubRoutines::dcos() != NULL ?
1725 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1726 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1727 case vmIntrinsics::_dtan:
1728 return StubRoutines::dtan() != NULL ?
1729 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1730 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1731 case vmIntrinsics::_dlog:
1732 return StubRoutines::dlog() != NULL ?
1733 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1734 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1735 case vmIntrinsics::_dlog10:
1736 return StubRoutines::dlog10() != NULL ?
1737 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1738 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1739
1740 // These intrinsics are supported on all hardware
1741 case vmIntrinsics::_ceil:
1742 case vmIntrinsics::_floor:
1743 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1744 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1745 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1746 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1747 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1748 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1749
1750 case vmIntrinsics::_dexp:
1751 return StubRoutines::dexp() != NULL ?
1752 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1753 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1754 #undef FN_PTR
1755
1756 case vmIntrinsics::_dpow: return inline_math_pow();
1757 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1758 case vmIntrinsics::_fcopySign: return inline_math(id);
1759 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1760 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1761
1762 // These intrinsics are not yet correctly implemented
1763 case vmIntrinsics::_datan2:
1764 return false;
1765
1766 default:
1767 fatal_unexpected_iid(id);
1768 return false;
1769 }
1770 }
1771
1772 static bool is_simple_name(Node* n) {
1773 return (n->req() == 1 // constant
1774 || (n->is_Type() && n->as_Type()->type()->singleton())
1775 || n->is_Proj() // parameter or return value
1776 || n->is_Phi() // local of some sort
1777 );
1778 }
1779
1780 //----------------------------inline_notify-----------------------------------*
1781 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1782 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1783 address func;
1784 if (id == vmIntrinsics::_notify) {
1785 func = OptoRuntime::monitor_notify_Java();
1786 } else {
1787 func = OptoRuntime::monitor_notifyAll_Java();
1788 }
1789 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1790 make_slow_call_ex(call, env()->Throwable_klass(), false);
1791 return true;
1792 }
1793
1794
1795 //----------------------------inline_min_max-----------------------------------
1796 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1797 set_result(generate_min_max(id, argument(0), argument(1)));
1798 return true;
1799 }
1800
1801 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1802 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1803 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1804 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1805 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1806
1807 {
1808 PreserveJVMState pjvms(this);
1809 PreserveReexecuteState preexecs(this);
1810 jvms()->set_should_reexecute(true);
1811
1812 set_control(slow_path);
1813 set_i_o(i_o());
1814
1815 uncommon_trap(Deoptimization::Reason_intrinsic,
1816 Deoptimization::Action_none);
1817 }
1818
1819 set_control(fast_path);
1820 set_result(math);
1821 }
1822
1823 template <typename OverflowOp>
1824 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1825 typedef typename OverflowOp::MathOp MathOp;
1826
1827 MathOp* mathOp = new MathOp(arg1, arg2);
1828 Node* operation = _gvn.transform( mathOp );
1829 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1830 inline_math_mathExact(operation, ofcheck);
1831 return true;
1832 }
1833
1834 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1835 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1836 }
1837
1838 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1839 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1840 }
1841
1842 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1843 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1844 }
1845
1846 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1847 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1848 }
1849
1850 bool LibraryCallKit::inline_math_negateExactI() {
1851 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1852 }
1853
1854 bool LibraryCallKit::inline_math_negateExactL() {
1855 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1856 }
1857
1858 bool LibraryCallKit::inline_math_multiplyExactI() {
1859 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1860 }
1861
1862 bool LibraryCallKit::inline_math_multiplyExactL() {
1863 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1864 }
1865
1866 bool LibraryCallKit::inline_math_multiplyHigh() {
1867 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1868 return true;
1869 }
1870
1871 Node*
1872 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1873 // These are the candidate return value:
1874 Node* xvalue = x0;
1875 Node* yvalue = y0;
1876
1877 if (xvalue == yvalue) {
1878 return xvalue;
1879 }
1880
1881 bool want_max = (id == vmIntrinsics::_max);
1882
1883 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1884 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1885 if (txvalue == NULL || tyvalue == NULL) return top();
1886 // This is not really necessary, but it is consistent with a
1887 // hypothetical MaxINode::Value method:
1888 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1889
1890 // %%% This folding logic should (ideally) be in a different place.
1891 // Some should be inside IfNode, and there to be a more reliable
1892 // transformation of ?: style patterns into cmoves. We also want
1893 // more powerful optimizations around cmove and min/max.
1894
1895 // Try to find a dominating comparison of these guys.
1896 // It can simplify the index computation for Arrays.copyOf
1897 // and similar uses of System.arraycopy.
1898 // First, compute the normalized version of CmpI(x, y).
1899 int cmp_op = Op_CmpI;
1900 Node* xkey = xvalue;
1901 Node* ykey = yvalue;
1902 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1903 if (ideal_cmpxy->is_Cmp()) {
1904 // E.g., if we have CmpI(length - offset, count),
1905 // it might idealize to CmpI(length, count + offset)
1906 cmp_op = ideal_cmpxy->Opcode();
1907 xkey = ideal_cmpxy->in(1);
1908 ykey = ideal_cmpxy->in(2);
1909 }
1910
1911 // Start by locating any relevant comparisons.
1912 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1913 Node* cmpxy = NULL;
1914 Node* cmpyx = NULL;
1915 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1916 Node* cmp = start_from->fast_out(k);
1917 if (cmp->outcnt() > 0 && // must have prior uses
1918 cmp->in(0) == NULL && // must be context-independent
1919 cmp->Opcode() == cmp_op) { // right kind of compare
1920 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1921 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1922 }
1923 }
1924
1925 const int NCMPS = 2;
1926 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1927 int cmpn;
1928 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1929 if (cmps[cmpn] != NULL) break; // find a result
1930 }
1931 if (cmpn < NCMPS) {
1932 // Look for a dominating test that tells us the min and max.
1933 int depth = 0; // Limit search depth for speed
1934 Node* dom = control();
1935 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1936 if (++depth >= 100) break;
1937 Node* ifproj = dom;
1938 if (!ifproj->is_Proj()) continue;
1939 Node* iff = ifproj->in(0);
1940 if (!iff->is_If()) continue;
1941 Node* bol = iff->in(1);
1942 if (!bol->is_Bool()) continue;
1943 Node* cmp = bol->in(1);
1944 if (cmp == NULL) continue;
1945 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1946 if (cmps[cmpn] == cmp) break;
1947 if (cmpn == NCMPS) continue;
1948 BoolTest::mask btest = bol->as_Bool()->_test._test;
1949 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1950 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1951 // At this point, we know that 'x btest y' is true.
1952 switch (btest) {
1953 case BoolTest::eq:
1954 // They are proven equal, so we can collapse the min/max.
1955 // Either value is the answer. Choose the simpler.
1956 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1957 return yvalue;
1958 return xvalue;
1959 case BoolTest::lt: // x < y
1960 case BoolTest::le: // x <= y
1961 return (want_max ? yvalue : xvalue);
1962 case BoolTest::gt: // x > y
1963 case BoolTest::ge: // x >= y
1964 return (want_max ? xvalue : yvalue);
1965 default:
1966 break;
1967 }
1968 }
1969 }
1970
1971 // We failed to find a dominating test.
1972 // Let's pick a test that might GVN with prior tests.
1973 Node* best_bol = NULL;
1974 BoolTest::mask best_btest = BoolTest::illegal;
1975 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1976 Node* cmp = cmps[cmpn];
1977 if (cmp == NULL) continue;
1978 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
1979 Node* bol = cmp->fast_out(j);
1980 if (!bol->is_Bool()) continue;
1981 BoolTest::mask btest = bol->as_Bool()->_test._test;
1982 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
1983 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1984 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
1985 best_bol = bol->as_Bool();
1986 best_btest = btest;
1987 }
1988 }
1989 }
1990
1991 Node* answer_if_true = NULL;
1992 Node* answer_if_false = NULL;
1993 switch (best_btest) {
1994 default:
1995 if (cmpxy == NULL)
1996 cmpxy = ideal_cmpxy;
1997 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
1998 // and fall through:
1999 case BoolTest::lt: // x < y
2000 case BoolTest::le: // x <= y
2001 answer_if_true = (want_max ? yvalue : xvalue);
2002 answer_if_false = (want_max ? xvalue : yvalue);
2003 break;
2004 case BoolTest::gt: // x > y
2005 case BoolTest::ge: // x >= y
2006 answer_if_true = (want_max ? xvalue : yvalue);
2007 answer_if_false = (want_max ? yvalue : xvalue);
2008 break;
2009 }
2010
2011 jint hi, lo;
2012 if (want_max) {
2013 // We can sharpen the minimum.
2014 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2015 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2016 } else {
2017 // We can sharpen the maximum.
2018 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2019 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2020 }
2021
2022 // Use a flow-free graph structure, to avoid creating excess control edges
2023 // which could hinder other optimizations.
2024 // Since Math.min/max is often used with arraycopy, we want
2025 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2026 Node* cmov = CMoveNode::make(NULL, best_bol,
2027 answer_if_false, answer_if_true,
2028 TypeInt::make(lo, hi, widen));
2029
2030 return _gvn.transform(cmov);
2031
2032 /*
2033 // This is not as desirable as it may seem, since Min and Max
2034 // nodes do not have a full set of optimizations.
2035 // And they would interfere, anyway, with 'if' optimizations
2036 // and with CMoveI canonical forms.
2037 switch (id) {
2038 case vmIntrinsics::_min:
2039 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2040 case vmIntrinsics::_max:
2041 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2042 default:
2043 ShouldNotReachHere();
2044 }
2045 */
2046 }
2047
2048 inline int
2049 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2050 const TypePtr* base_type = TypePtr::NULL_PTR;
2051 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2052 if (base_type == NULL) {
2053 // Unknown type.
2054 return Type::AnyPtr;
2055 } else if (base_type == TypePtr::NULL_PTR) {
2056 // Since this is a NULL+long form, we have to switch to a rawptr.
2057 base = _gvn.transform(new CastX2PNode(offset));
2058 offset = MakeConX(0);
2059 return Type::RawPtr;
2060 } else if (base_type->base() == Type::RawPtr) {
2061 return Type::RawPtr;
2062 } else if (base_type->isa_oopptr()) {
2063 // Base is never null => always a heap address.
2064 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2065 return Type::OopPtr;
2066 }
2067 // Offset is small => always a heap address.
2068 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2069 if (offset_type != NULL &&
2070 base_type->offset() == 0 && // (should always be?)
2071 offset_type->_lo >= 0 &&
2072 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2073 return Type::OopPtr;
2074 } else if (type == T_OBJECT) {
2075 // off heap access to an oop doesn't make any sense. Has to be on
2076 // heap.
2077 return Type::OopPtr;
2078 }
2079 // Otherwise, it might either be oop+off or NULL+addr.
2080 return Type::AnyPtr;
2081 } else {
2082 // No information:
2083 return Type::AnyPtr;
2084 }
2085 }
2086
2087 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2088 Node* uncasted_base = base;
2089 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2090 if (kind == Type::RawPtr) {
2091 return basic_plus_adr(top(), uncasted_base, offset);
2092 } else if (kind == Type::AnyPtr) {
2093 assert(base == uncasted_base, "unexpected base change");
2094 if (can_cast) {
2095 if (!_gvn.type(base)->speculative_maybe_null() &&
2096 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2097 // According to profiling, this access is always on
2098 // heap. Casting the base to not null and thus avoiding membars
2099 // around the access should allow better optimizations
2100 Node* null_ctl = top();
2101 base = null_check_oop(base, &null_ctl, true, true, true);
2102 assert(null_ctl->is_top(), "no null control here");
2103 return basic_plus_adr(base, offset);
2104 } else if (_gvn.type(base)->speculative_always_null() &&
2105 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2106 // According to profiling, this access is always off
2107 // heap.
2108 base = null_assert(base);
2109 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2110 offset = MakeConX(0);
2111 return basic_plus_adr(top(), raw_base, offset);
2112 }
2113 }
2114 // We don't know if it's an on heap or off heap access. Fall back
2115 // to raw memory access.
2116 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2117 return basic_plus_adr(top(), raw, offset);
2118 } else {
2119 assert(base == uncasted_base, "unexpected base change");
2120 // We know it's an on heap access so base can't be null
2121 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2122 base = must_be_not_null(base, true);
2123 }
2124 return basic_plus_adr(base, offset);
2125 }
2126 }
2127
2128 //--------------------------inline_number_methods-----------------------------
2129 // inline int Integer.numberOfLeadingZeros(int)
2130 // inline int Long.numberOfLeadingZeros(long)
2131 //
2132 // inline int Integer.numberOfTrailingZeros(int)
2133 // inline int Long.numberOfTrailingZeros(long)
2134 //
2135 // inline int Integer.bitCount(int)
2136 // inline int Long.bitCount(long)
2137 //
2138 // inline char Character.reverseBytes(char)
2139 // inline short Short.reverseBytes(short)
2140 // inline int Integer.reverseBytes(int)
2141 // inline long Long.reverseBytes(long)
2142 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2143 Node* arg = argument(0);
2144 Node* n = NULL;
2145 switch (id) {
2146 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2147 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2148 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2149 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2150 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2151 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2152 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2153 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2154 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2155 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2156 default: fatal_unexpected_iid(id); break;
2157 }
2158 set_result(_gvn.transform(n));
2159 return true;
2160 }
2161
2162 //----------------------------inline_unsafe_access----------------------------
2163
2164 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2165 // Attempt to infer a sharper value type from the offset and base type.
2166 ciKlass* sharpened_klass = NULL;
2167
2168 // See if it is an instance field, with an object type.
2169 if (alias_type->field() != NULL) {
2170 if (alias_type->field()->type()->is_klass()) {
2171 sharpened_klass = alias_type->field()->type()->as_klass();
2172 }
2173 }
2174
2175 // See if it is a narrow oop array.
2176 if (adr_type->isa_aryptr()) {
2177 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2178 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2179 if (elem_type != NULL) {
2180 sharpened_klass = elem_type->klass();
2181 }
2182 }
2183 }
2184
2185 // The sharpened class might be unloaded if there is no class loader
2186 // contraint in place.
2187 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2188 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2189
2190 #ifndef PRODUCT
2191 if (C->print_intrinsics() || C->print_inlining()) {
2192 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2193 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2194 }
2195 #endif
2196 // Sharpen the value type.
2197 return tjp;
2198 }
2199 return NULL;
2200 }
2201
2202 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2203 switch (kind) {
2204 case Relaxed:
2205 return MO_UNORDERED;
2206 case Opaque:
2207 return MO_RELAXED;
2208 case Acquire:
2209 return MO_ACQUIRE;
2210 case Release:
2211 return MO_RELEASE;
2212 case Volatile:
2213 return MO_SEQ_CST;
2214 default:
2215 ShouldNotReachHere();
2216 return 0;
2217 }
2218 }
2219
2220 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2221 if (callee()->is_static()) return false; // caller must have the capability!
2222 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2223 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2224 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2225 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2226
2227 if (is_reference_type(type)) {
2228 decorators |= ON_UNKNOWN_OOP_REF;
2229 }
2230
2231 if (unaligned) {
2232 decorators |= C2_UNALIGNED;
2233 }
2234
2235 #ifndef PRODUCT
2236 {
2237 ResourceMark rm;
2238 // Check the signatures.
2239 ciSignature* sig = callee()->signature();
2240 #ifdef ASSERT
2241 if (!is_store) {
2242 // Object getReference(Object base, int/long offset), etc.
2243 BasicType rtype = sig->return_type()->basic_type();
2244 assert(rtype == type, "getter must return the expected value");
2245 assert(sig->count() == 2, "oop getter has 2 arguments");
2246 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2247 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2248 } else {
2249 // void putReference(Object base, int/long offset, Object x), etc.
2250 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2251 assert(sig->count() == 3, "oop putter has 3 arguments");
2252 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2253 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2254 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2255 assert(vtype == type, "putter must accept the expected value");
2256 }
2257 #endif // ASSERT
2258 }
2259 #endif //PRODUCT
2260
2261 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2262
2263 Node* receiver = argument(0); // type: oop
2264
2265 // Build address expression.
2266 Node* heap_base_oop = top();
2267
2268 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2269 Node* base = argument(1); // type: oop
2270 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2271 Node* offset = argument(2); // type: long
2272 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2273 // to be plain byte offsets, which are also the same as those accepted
2274 // by oopDesc::field_addr.
2275 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2276 "fieldOffset must be byte-scaled");
2277 // 32-bit machines ignore the high half!
2278 offset = ConvL2X(offset);
2279
2280 // Save state and restore on bailout
2281 uint old_sp = sp();
2282 SafePointNode* old_map = clone_map();
2283
2284 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2285
2286 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2287 if (type != T_OBJECT) {
2288 decorators |= IN_NATIVE; // off-heap primitive access
2289 } else {
2290 set_map(old_map);
2291 set_sp(old_sp);
2292 return false; // off-heap oop accesses are not supported
2293 }
2294 } else {
2295 heap_base_oop = base; // on-heap or mixed access
2296 }
2297
2298 // Can base be NULL? Otherwise, always on-heap access.
2299 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2300
2301 if (!can_access_non_heap) {
2302 decorators |= IN_HEAP;
2303 }
2304
2305 Node* val = is_store ? argument(4) : NULL;
2306
2307 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2308 if (adr_type == TypePtr::NULL_PTR) {
2309 set_map(old_map);
2310 set_sp(old_sp);
2311 return false; // off-heap access with zero address
2312 }
2313
2314 // Try to categorize the address.
2315 Compile::AliasType* alias_type = C->alias_type(adr_type);
2316 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2317
2318 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2319 alias_type->adr_type() == TypeAryPtr::RANGE) {
2320 set_map(old_map);
2321 set_sp(old_sp);
2322 return false; // not supported
2323 }
2324
2325 bool mismatched = false;
2326 BasicType bt = alias_type->basic_type();
2327 if (bt != T_ILLEGAL) {
2328 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2329 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2330 // Alias type doesn't differentiate between byte[] and boolean[]).
2331 // Use address type to get the element type.
2332 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2333 }
2334 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2335 // accessing an array field with getReference is not a mismatch
2336 bt = T_OBJECT;
2337 }
2338 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2339 // Don't intrinsify mismatched object accesses
2340 set_map(old_map);
2341 set_sp(old_sp);
2342 return false;
2343 }
2344 mismatched = (bt != type);
2345 } else if (alias_type->adr_type()->isa_oopptr()) {
2346 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2347 }
2348
2349 old_map->destruct(&_gvn);
2350 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2351
2352 if (mismatched) {
2353 decorators |= C2_MISMATCHED;
2354 }
2355
2356 // First guess at the value type.
2357 const Type *value_type = Type::get_const_basic_type(type);
2358
2359 // Figure out the memory ordering.
2360 decorators |= mo_decorator_for_access_kind(kind);
2361
2362 if (!is_store && type == T_OBJECT) {
2363 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2364 if (tjp != NULL) {
2365 value_type = tjp;
2366 }
2367 }
2368
2369 receiver = null_check(receiver);
2370 if (stopped()) {
2371 return true;
2372 }
2373 // Heap pointers get a null-check from the interpreter,
2374 // as a courtesy. However, this is not guaranteed by Unsafe,
2375 // and it is not possible to fully distinguish unintended nulls
2376 // from intended ones in this API.
2377
2378 if (!is_store) {
2379 Node* p = NULL;
2380 // Try to constant fold a load from a constant field
2381 ciField* field = alias_type->field();
2382 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2383 // final or stable field
2384 p = make_constant_from_field(field, heap_base_oop);
2385 }
2386
2387 if (p == NULL) { // Could not constant fold the load
2388 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2389 // Normalize the value returned by getBoolean in the following cases
2390 if (type == T_BOOLEAN &&
2391 (mismatched ||
2392 heap_base_oop == top() || // - heap_base_oop is NULL or
2393 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2394 // and the unsafe access is made to large offset
2395 // (i.e., larger than the maximum offset necessary for any
2396 // field access)
2397 ) {
2398 IdealKit ideal = IdealKit(this);
2399 #define __ ideal.
2400 IdealVariable normalized_result(ideal);
2401 __ declarations_done();
2402 __ set(normalized_result, p);
2403 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2404 __ set(normalized_result, ideal.ConI(1));
2405 ideal.end_if();
2406 final_sync(ideal);
2407 p = __ value(normalized_result);
2408 #undef __
2409 }
2410 }
2411 if (type == T_ADDRESS) {
2412 p = gvn().transform(new CastP2XNode(NULL, p));
2413 p = ConvX2UL(p);
2414 }
2415 // The load node has the control of the preceding MemBarCPUOrder. All
2416 // following nodes will have the control of the MemBarCPUOrder inserted at
2417 // the end of this method. So, pushing the load onto the stack at a later
2418 // point is fine.
2419 set_result(p);
2420 } else {
2421 if (bt == T_ADDRESS) {
2422 // Repackage the long as a pointer.
2423 val = ConvL2X(val);
2424 val = gvn().transform(new CastX2PNode(val));
2425 }
2426 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2427 }
2428
2429 return true;
2430 }
2431
2432 //----------------------------inline_unsafe_load_store----------------------------
2433 // This method serves a couple of different customers (depending on LoadStoreKind):
2434 //
2435 // LS_cmp_swap:
2436 //
2437 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2438 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2439 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2440 //
2441 // LS_cmp_swap_weak:
2442 //
2443 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2444 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2445 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2446 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2447 //
2448 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2449 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2450 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2451 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2452 //
2453 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2454 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2455 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2456 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2457 //
2458 // LS_cmp_exchange:
2459 //
2460 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2461 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2462 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2463 //
2464 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2465 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2466 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2467 //
2468 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2469 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2470 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2471 //
2472 // LS_get_add:
2473 //
2474 // int getAndAddInt( Object o, long offset, int delta)
2475 // long getAndAddLong(Object o, long offset, long delta)
2476 //
2477 // LS_get_set:
2478 //
2479 // int getAndSet(Object o, long offset, int newValue)
2480 // long getAndSet(Object o, long offset, long newValue)
2481 // Object getAndSet(Object o, long offset, Object newValue)
2482 //
2483 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2484 // This basic scheme here is the same as inline_unsafe_access, but
2485 // differs in enough details that combining them would make the code
2486 // overly confusing. (This is a true fact! I originally combined
2487 // them, but even I was confused by it!) As much code/comments as
2488 // possible are retained from inline_unsafe_access though to make
2489 // the correspondences clearer. - dl
2490
2491 if (callee()->is_static()) return false; // caller must have the capability!
2492
2493 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2494 decorators |= mo_decorator_for_access_kind(access_kind);
2495
2496 #ifndef PRODUCT
2497 BasicType rtype;
2498 {
2499 ResourceMark rm;
2500 // Check the signatures.
2501 ciSignature* sig = callee()->signature();
2502 rtype = sig->return_type()->basic_type();
2503 switch(kind) {
2504 case LS_get_add:
2505 case LS_get_set: {
2506 // Check the signatures.
2507 #ifdef ASSERT
2508 assert(rtype == type, "get and set must return the expected type");
2509 assert(sig->count() == 3, "get and set has 3 arguments");
2510 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2511 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2512 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2513 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2514 #endif // ASSERT
2515 break;
2516 }
2517 case LS_cmp_swap:
2518 case LS_cmp_swap_weak: {
2519 // Check the signatures.
2520 #ifdef ASSERT
2521 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2522 assert(sig->count() == 4, "CAS has 4 arguments");
2523 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2524 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2525 #endif // ASSERT
2526 break;
2527 }
2528 case LS_cmp_exchange: {
2529 // Check the signatures.
2530 #ifdef ASSERT
2531 assert(rtype == type, "CAS must return the expected type");
2532 assert(sig->count() == 4, "CAS has 4 arguments");
2533 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2534 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2535 #endif // ASSERT
2536 break;
2537 }
2538 default:
2539 ShouldNotReachHere();
2540 }
2541 }
2542 #endif //PRODUCT
2543
2544 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2545
2546 // Get arguments:
2547 Node* receiver = NULL;
2548 Node* base = NULL;
2549 Node* offset = NULL;
2550 Node* oldval = NULL;
2551 Node* newval = NULL;
2552 switch(kind) {
2553 case LS_cmp_swap:
2554 case LS_cmp_swap_weak:
2555 case LS_cmp_exchange: {
2556 const bool two_slot_type = type2size[type] == 2;
2557 receiver = argument(0); // type: oop
2558 base = argument(1); // type: oop
2559 offset = argument(2); // type: long
2560 oldval = argument(4); // type: oop, int, or long
2561 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2562 break;
2563 }
2564 case LS_get_add:
2565 case LS_get_set: {
2566 receiver = argument(0); // type: oop
2567 base = argument(1); // type: oop
2568 offset = argument(2); // type: long
2569 oldval = NULL;
2570 newval = argument(4); // type: oop, int, or long
2571 break;
2572 }
2573 default:
2574 ShouldNotReachHere();
2575 }
2576
2577 // Build field offset expression.
2578 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2579 // to be plain byte offsets, which are also the same as those accepted
2580 // by oopDesc::field_addr.
2581 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2582 // 32-bit machines ignore the high half of long offsets
2583 offset = ConvL2X(offset);
2584 // Save state and restore on bailout
2585 uint old_sp = sp();
2586 SafePointNode* old_map = clone_map();
2587 Node* adr = make_unsafe_address(base, offset,type, false);
2588 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2589
2590 Compile::AliasType* alias_type = C->alias_type(adr_type);
2591 BasicType bt = alias_type->basic_type();
2592 if (bt != T_ILLEGAL &&
2593 (is_reference_type(bt) != (type == T_OBJECT))) {
2594 // Don't intrinsify mismatched object accesses.
2595 set_map(old_map);
2596 set_sp(old_sp);
2597 return false;
2598 }
2599
2600 old_map->destruct(&_gvn);
2601
2602 // For CAS, unlike inline_unsafe_access, there seems no point in
2603 // trying to refine types. Just use the coarse types here.
2604 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2605 const Type *value_type = Type::get_const_basic_type(type);
2606
2607 switch (kind) {
2608 case LS_get_set:
2609 case LS_cmp_exchange: {
2610 if (type == T_OBJECT) {
2611 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2612 if (tjp != NULL) {
2613 value_type = tjp;
2614 }
2615 }
2616 break;
2617 }
2618 case LS_cmp_swap:
2619 case LS_cmp_swap_weak:
2620 case LS_get_add:
2621 break;
2622 default:
2623 ShouldNotReachHere();
2624 }
2625
2626 // Null check receiver.
2627 receiver = null_check(receiver);
2628 if (stopped()) {
2629 return true;
2630 }
2631
2632 int alias_idx = C->get_alias_index(adr_type);
2633
2634 if (is_reference_type(type)) {
2635 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2636
2637 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2638 // could be delayed during Parse (for example, in adjust_map_after_if()).
2639 // Execute transformation here to avoid barrier generation in such case.
2640 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2641 newval = _gvn.makecon(TypePtr::NULL_PTR);
2642
2643 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2644 // Refine the value to a null constant, when it is known to be null
2645 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2646 }
2647 }
2648
2649 Node* result = NULL;
2650 switch (kind) {
2651 case LS_cmp_exchange: {
2652 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2653 oldval, newval, value_type, type, decorators);
2654 break;
2655 }
2656 case LS_cmp_swap_weak:
2657 decorators |= C2_WEAK_CMPXCHG;
2658 case LS_cmp_swap: {
2659 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2660 oldval, newval, value_type, type, decorators);
2661 break;
2662 }
2663 case LS_get_set: {
2664 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2665 newval, value_type, type, decorators);
2666 break;
2667 }
2668 case LS_get_add: {
2669 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2670 newval, value_type, type, decorators);
2671 break;
2672 }
2673 default:
2674 ShouldNotReachHere();
2675 }
2676
2677 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2678 set_result(result);
2679 return true;
2680 }
2681
2682 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2683 // Regardless of form, don't allow previous ld/st to move down,
2684 // then issue acquire, release, or volatile mem_bar.
2685 insert_mem_bar(Op_MemBarCPUOrder);
2686 switch(id) {
2687 case vmIntrinsics::_loadFence:
2688 insert_mem_bar(Op_LoadFence);
2689 return true;
2690 case vmIntrinsics::_storeFence:
2691 insert_mem_bar(Op_StoreFence);
2692 return true;
2693 case vmIntrinsics::_fullFence:
2694 insert_mem_bar(Op_MemBarVolatile);
2695 return true;
2696 default:
2697 fatal_unexpected_iid(id);
2698 return false;
2699 }
2700 }
2701
2702 bool LibraryCallKit::inline_onspinwait() {
2703 insert_mem_bar(Op_OnSpinWait);
2704 return true;
2705 }
2706
2707 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2708 if (!kls->is_Con()) {
2709 return true;
2710 }
2711 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2712 if (klsptr == NULL) {
2713 return true;
2714 }
2715 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2716 // don't need a guard for a klass that is already initialized
2717 return !ik->is_initialized();
2718 }
2719
2720 //----------------------------inline_unsafe_writeback0-------------------------
2721 // public native void Unsafe.writeback0(long address)
2722 bool LibraryCallKit::inline_unsafe_writeback0() {
2723 if (!Matcher::has_match_rule(Op_CacheWB)) {
2724 return false;
2725 }
2726 #ifndef PRODUCT
2727 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2728 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2729 ciSignature* sig = callee()->signature();
2730 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2731 #endif
2732 null_check_receiver(); // null-check, then ignore
2733 Node *addr = argument(1);
2734 addr = new CastX2PNode(addr);
2735 addr = _gvn.transform(addr);
2736 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2737 flush = _gvn.transform(flush);
2738 set_memory(flush, TypeRawPtr::BOTTOM);
2739 return true;
2740 }
2741
2742 //----------------------------inline_unsafe_writeback0-------------------------
2743 // public native void Unsafe.writeback0(long address)
2744 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2745 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2746 return false;
2747 }
2748 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2749 return false;
2750 }
2751 #ifndef PRODUCT
2752 assert(Matcher::has_match_rule(Op_CacheWB),
2753 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2754 : "found match rule for CacheWBPostSync but not CacheWB"));
2755
2756 #endif
2757 null_check_receiver(); // null-check, then ignore
2758 Node *sync;
2759 if (is_pre) {
2760 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2761 } else {
2762 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2763 }
2764 sync = _gvn.transform(sync);
2765 set_memory(sync, TypeRawPtr::BOTTOM);
2766 return true;
2767 }
2768
2769 //----------------------------inline_unsafe_allocate---------------------------
2770 // public native Object Unsafe.allocateInstance(Class<?> cls);
2771 bool LibraryCallKit::inline_unsafe_allocate() {
2772 if (callee()->is_static()) return false; // caller must have the capability!
2773
2774 null_check_receiver(); // null-check, then ignore
2775 Node* cls = null_check(argument(1));
2776 if (stopped()) return true;
2777
2778 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2779 kls = null_check(kls);
2780 if (stopped()) return true; // argument was like int.class
2781
2782 Node* test = NULL;
2783 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2784 // Note: The argument might still be an illegal value like
2785 // Serializable.class or Object[].class. The runtime will handle it.
2786 // But we must make an explicit check for initialization.
2787 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2788 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2789 // can generate code to load it as unsigned byte.
2790 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2791 Node* bits = intcon(InstanceKlass::fully_initialized);
2792 test = _gvn.transform(new SubINode(inst, bits));
2793 // The 'test' is non-zero if we need to take a slow path.
2794 }
2795
2796 Node* obj = new_instance(kls, test);
2797 set_result(obj);
2798 return true;
2799 }
2800
2801 //------------------------inline_native_time_funcs--------------
2802 // inline code for System.currentTimeMillis() and System.nanoTime()
2803 // these have the same type and signature
2804 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2805 const TypeFunc* tf = OptoRuntime::void_long_Type();
2806 const TypePtr* no_memory_effects = NULL;
2807 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2808 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2809 #ifdef ASSERT
2810 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2811 assert(value_top == top(), "second value must be top");
2812 #endif
2813 set_result(value);
2814 return true;
2815 }
2816
2817 #ifdef JFR_HAVE_INTRINSICS
2818
2819 /**
2820 * if oop->klass != null
2821 * // normal class
2822 * epoch = _epoch_state ? 2 : 1
2823 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2824 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2825 * }
2826 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2827 * else
2828 * // primitive class
2829 * if oop->array_klass != null
2830 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2831 * else
2832 * id = LAST_TYPE_ID + 1 // void class path
2833 * if (!signaled)
2834 * signaled = true
2835 */
2836 bool LibraryCallKit::inline_native_classID() {
2837 Node* cls = argument(0);
2838
2839 IdealKit ideal(this);
2840 #define __ ideal.
2841 IdealVariable result(ideal); __ declarations_done();
2842 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2843 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2844 TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
2845
2846
2847 __ if_then(kls, BoolTest::ne, null()); {
2848 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2849 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2850
2851 Node* epoch_address = makecon(TypeRawPtr::make(Jfr::epoch_address()));
2852 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2853 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2854 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2855 mask = _gvn.transform(new OrLNode(mask, epoch));
2856 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2857
2858 float unlikely = PROB_UNLIKELY(0.999);
2859 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
2860 sync_kit(ideal);
2861 make_runtime_call(RC_LEAF,
2862 OptoRuntime::get_class_id_intrinsic_Type(),
2863 CAST_FROM_FN_PTR(address, Jfr::get_class_id_intrinsic),
2864 "get_class_id_intrinsic",
2865 TypePtr::BOTTOM,
2866 kls);
2867 ideal.sync_kit(this);
2868 } __ end_if();
2869
2870 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
2871 } __ else_(); {
2872 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2873 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
2874 TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
2875 __ if_then(array_kls, BoolTest::ne, null()); {
2876 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2877 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2878 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
2879 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
2880 } __ else_(); {
2881 // void class case
2882 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
2883 } __ end_if();
2884
2885 Node* signaled_flag_address = makecon(TypeRawPtr::make(Jfr::signal_address()));
2886 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
2887 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
2888 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
2889 } __ end_if();
2890 } __ end_if();
2891
2892 final_sync(ideal);
2893 set_result(ideal.value(result));
2894 #undef __
2895 return true;
2896 }
2897
2898 bool LibraryCallKit::inline_native_getEventWriter() {
2899 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2900
2901 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2902 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2903
2904 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2905
2906 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2907 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2908
2909 IfNode* iff_jobj_null =
2910 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2911
2912 enum { _normal_path = 1,
2913 _null_path = 2,
2914 PATH_LIMIT };
2915
2916 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2917 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
2918
2919 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2920 result_rgn->init_req(_null_path, jobj_is_null);
2921 result_val->init_req(_null_path, null());
2922
2923 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2924 set_control(jobj_is_not_null);
2925 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
2926 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
2927 result_rgn->init_req(_normal_path, control());
2928 result_val->init_req(_normal_path, res);
2929
2930 set_result(result_rgn, result_val);
2931
2932 return true;
2933 }
2934
2935 #endif // JFR_HAVE_INTRINSICS
2936
2937 //------------------------inline_native_currentThread------------------
2938 bool LibraryCallKit::inline_native_currentThread() {
2939 Node* junk = NULL;
2940 set_result(generate_current_thread(junk));
2941 return true;
2942 }
2943
2944 //---------------------------load_mirror_from_klass----------------------------
2945 // Given a klass oop, load its java mirror (a java.lang.Class oop).
2946 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
2947 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
2948 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
2949 // mirror = ((OopHandle)mirror)->resolve();
2950 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
2951 }
2952
2953 //-----------------------load_klass_from_mirror_common-------------------------
2954 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
2955 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
2956 // and branch to the given path on the region.
2957 // If never_see_null, take an uncommon trap on null, so we can optimistically
2958 // compile for the non-null case.
2959 // If the region is NULL, force never_see_null = true.
2960 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
2961 bool never_see_null,
2962 RegionNode* region,
2963 int null_path,
2964 int offset) {
2965 if (region == NULL) never_see_null = true;
2966 Node* p = basic_plus_adr(mirror, offset);
2967 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
2968 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
2969 Node* null_ctl = top();
2970 kls = null_check_oop(kls, &null_ctl, never_see_null);
2971 if (region != NULL) {
2972 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
2973 region->init_req(null_path, null_ctl);
2974 } else {
2975 assert(null_ctl == top(), "no loose ends");
2976 }
2977 return kls;
2978 }
2979
2980 //--------------------(inline_native_Class_query helpers)---------------------
2981 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
2982 // Fall through if (mods & mask) == bits, take the guard otherwise.
2983 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
2984 // Branch around if the given klass has the given modifier bit set.
2985 // Like generate_guard, adds a new path onto the region.
2986 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
2987 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
2988 Node* mask = intcon(modifier_mask);
2989 Node* bits = intcon(modifier_bits);
2990 Node* mbit = _gvn.transform(new AndINode(mods, mask));
2991 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
2992 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
2993 return generate_fair_guard(bol, region);
2994 }
2995 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
2996 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
2997 }
2998 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
2999 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3000 }
3001
3002 //-------------------------inline_native_Class_query-------------------
3003 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3004 const Type* return_type = TypeInt::BOOL;
3005 Node* prim_return_value = top(); // what happens if it's a primitive class?
3006 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3007 bool expect_prim = false; // most of these guys expect to work on refs
3008
3009 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3010
3011 Node* mirror = argument(0);
3012 Node* obj = top();
3013
3014 switch (id) {
3015 case vmIntrinsics::_isInstance:
3016 // nothing is an instance of a primitive type
3017 prim_return_value = intcon(0);
3018 obj = argument(1);
3019 break;
3020 case vmIntrinsics::_getModifiers:
3021 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3022 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3023 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3024 break;
3025 case vmIntrinsics::_isInterface:
3026 prim_return_value = intcon(0);
3027 break;
3028 case vmIntrinsics::_isArray:
3029 prim_return_value = intcon(0);
3030 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3031 break;
3032 case vmIntrinsics::_isPrimitive:
3033 prim_return_value = intcon(1);
3034 expect_prim = true; // obviously
3035 break;
3036 case vmIntrinsics::_isHidden:
3037 prim_return_value = intcon(0);
3038 break;
3039 case vmIntrinsics::_getSuperclass:
3040 prim_return_value = null();
3041 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3042 break;
3043 case vmIntrinsics::_getClassAccessFlags:
3044 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3045 return_type = TypeInt::INT; // not bool! 6297094
3046 break;
3047 default:
3048 fatal_unexpected_iid(id);
3049 break;
3050 }
3051
3052 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3053 if (mirror_con == NULL) return false; // cannot happen?
3054
3055 #ifndef PRODUCT
3056 if (C->print_intrinsics() || C->print_inlining()) {
3057 ciType* k = mirror_con->java_mirror_type();
3058 if (k) {
3059 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3060 k->print_name();
3061 tty->cr();
3062 }
3063 }
3064 #endif
3065
3066 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3067 RegionNode* region = new RegionNode(PATH_LIMIT);
3068 record_for_igvn(region);
3069 PhiNode* phi = new PhiNode(region, return_type);
3070
3071 // The mirror will never be null of Reflection.getClassAccessFlags, however
3072 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3073 // if it is. See bug 4774291.
3074
3075 // For Reflection.getClassAccessFlags(), the null check occurs in
3076 // the wrong place; see inline_unsafe_access(), above, for a similar
3077 // situation.
3078 mirror = null_check(mirror);
3079 // If mirror or obj is dead, only null-path is taken.
3080 if (stopped()) return true;
3081
3082 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3083
3084 // Now load the mirror's klass metaobject, and null-check it.
3085 // Side-effects region with the control path if the klass is null.
3086 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3087 // If kls is null, we have a primitive mirror.
3088 phi->init_req(_prim_path, prim_return_value);
3089 if (stopped()) { set_result(region, phi); return true; }
3090 bool safe_for_replace = (region->in(_prim_path) == top());
3091
3092 Node* p; // handy temp
3093 Node* null_ctl;
3094
3095 // Now that we have the non-null klass, we can perform the real query.
3096 // For constant classes, the query will constant-fold in LoadNode::Value.
3097 Node* query_value = top();
3098 switch (id) {
3099 case vmIntrinsics::_isInstance:
3100 // nothing is an instance of a primitive type
3101 query_value = gen_instanceof(obj, kls, safe_for_replace);
3102 break;
3103
3104 case vmIntrinsics::_getModifiers:
3105 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3106 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3107 break;
3108
3109 case vmIntrinsics::_isInterface:
3110 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3111 if (generate_interface_guard(kls, region) != NULL)
3112 // A guard was added. If the guard is taken, it was an interface.
3113 phi->add_req(intcon(1));
3114 // If we fall through, it's a plain class.
3115 query_value = intcon(0);
3116 break;
3117
3118 case vmIntrinsics::_isArray:
3119 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3120 if (generate_array_guard(kls, region) != NULL)
3121 // A guard was added. If the guard is taken, it was an array.
3122 phi->add_req(intcon(1));
3123 // If we fall through, it's a plain class.
3124 query_value = intcon(0);
3125 break;
3126
3127 case vmIntrinsics::_isPrimitive:
3128 query_value = intcon(0); // "normal" path produces false
3129 break;
3130
3131 case vmIntrinsics::_isHidden:
3132 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3133 if (generate_hidden_class_guard(kls, region) != NULL)
3134 // A guard was added. If the guard is taken, it was an hidden class.
3135 phi->add_req(intcon(1));
3136 // If we fall through, it's a plain class.
3137 query_value = intcon(0);
3138 break;
3139
3140
3141 case vmIntrinsics::_getSuperclass:
3142 // The rules here are somewhat unfortunate, but we can still do better
3143 // with random logic than with a JNI call.
3144 // Interfaces store null or Object as _super, but must report null.
3145 // Arrays store an intermediate super as _super, but must report Object.
3146 // Other types can report the actual _super.
3147 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3148 if (generate_interface_guard(kls, region) != NULL)
3149 // A guard was added. If the guard is taken, it was an interface.
3150 phi->add_req(null());
3151 if (generate_array_guard(kls, region) != NULL)
3152 // A guard was added. If the guard is taken, it was an array.
3153 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3154 // If we fall through, it's a plain class. Get its _super.
3155 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3156 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3157 null_ctl = top();
3158 kls = null_check_oop(kls, &null_ctl);
3159 if (null_ctl != top()) {
3160 // If the guard is taken, Object.superClass is null (both klass and mirror).
3161 region->add_req(null_ctl);
3162 phi ->add_req(null());
3163 }
3164 if (!stopped()) {
3165 query_value = load_mirror_from_klass(kls);
3166 }
3167 break;
3168
3169 case vmIntrinsics::_getClassAccessFlags:
3170 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3171 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3172 break;
3173
3174 default:
3175 fatal_unexpected_iid(id);
3176 break;
3177 }
3178
3179 // Fall-through is the normal case of a query to a real class.
3180 phi->init_req(1, query_value);
3181 region->init_req(1, control());
3182
3183 C->set_has_split_ifs(true); // Has chance for split-if optimization
3184 set_result(region, phi);
3185 return true;
3186 }
3187
3188 //-------------------------inline_Class_cast-------------------
3189 bool LibraryCallKit::inline_Class_cast() {
3190 Node* mirror = argument(0); // Class
3191 Node* obj = argument(1);
3192 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3193 if (mirror_con == NULL) {
3194 return false; // dead path (mirror->is_top()).
3195 }
3196 if (obj == NULL || obj->is_top()) {
3197 return false; // dead path
3198 }
3199 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3200
3201 // First, see if Class.cast() can be folded statically.
3202 // java_mirror_type() returns non-null for compile-time Class constants.
3203 ciType* tm = mirror_con->java_mirror_type();
3204 if (tm != NULL && tm->is_klass() &&
3205 tp != NULL && tp->klass() != NULL) {
3206 if (!tp->klass()->is_loaded()) {
3207 // Don't use intrinsic when class is not loaded.
3208 return false;
3209 } else {
3210 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3211 if (static_res == Compile::SSC_always_true) {
3212 // isInstance() is true - fold the code.
3213 set_result(obj);
3214 return true;
3215 } else if (static_res == Compile::SSC_always_false) {
3216 // Don't use intrinsic, have to throw ClassCastException.
3217 // If the reference is null, the non-intrinsic bytecode will
3218 // be optimized appropriately.
3219 return false;
3220 }
3221 }
3222 }
3223
3224 // Bailout intrinsic and do normal inlining if exception path is frequent.
3225 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3226 return false;
3227 }
3228
3229 // Generate dynamic checks.
3230 // Class.cast() is java implementation of _checkcast bytecode.
3231 // Do checkcast (Parse::do_checkcast()) optimizations here.
3232
3233 mirror = null_check(mirror);
3234 // If mirror is dead, only null-path is taken.
3235 if (stopped()) {
3236 return true;
3237 }
3238
3239 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3240 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3241 RegionNode* region = new RegionNode(PATH_LIMIT);
3242 record_for_igvn(region);
3243
3244 // Now load the mirror's klass metaobject, and null-check it.
3245 // If kls is null, we have a primitive mirror and
3246 // nothing is an instance of a primitive type.
3247 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3248
3249 Node* res = top();
3250 if (!stopped()) {
3251 Node* bad_type_ctrl = top();
3252 // Do checkcast optimizations.
3253 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3254 region->init_req(_bad_type_path, bad_type_ctrl);
3255 }
3256 if (region->in(_prim_path) != top() ||
3257 region->in(_bad_type_path) != top()) {
3258 // Let Interpreter throw ClassCastException.
3259 PreserveJVMState pjvms(this);
3260 set_control(_gvn.transform(region));
3261 uncommon_trap(Deoptimization::Reason_intrinsic,
3262 Deoptimization::Action_maybe_recompile);
3263 }
3264 if (!stopped()) {
3265 set_result(res);
3266 }
3267 return true;
3268 }
3269
3270
3271 //--------------------------inline_native_subtype_check------------------------
3272 // This intrinsic takes the JNI calls out of the heart of
3273 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3274 bool LibraryCallKit::inline_native_subtype_check() {
3275 // Pull both arguments off the stack.
3276 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3277 args[0] = argument(0);
3278 args[1] = argument(1);
3279 Node* klasses[2]; // corresponding Klasses: superk, subk
3280 klasses[0] = klasses[1] = top();
3281
3282 enum {
3283 // A full decision tree on {superc is prim, subc is prim}:
3284 _prim_0_path = 1, // {P,N} => false
3285 // {P,P} & superc!=subc => false
3286 _prim_same_path, // {P,P} & superc==subc => true
3287 _prim_1_path, // {N,P} => false
3288 _ref_subtype_path, // {N,N} & subtype check wins => true
3289 _both_ref_path, // {N,N} & subtype check loses => false
3290 PATH_LIMIT
3291 };
3292
3293 RegionNode* region = new RegionNode(PATH_LIMIT);
3294 Node* phi = new PhiNode(region, TypeInt::BOOL);
3295 record_for_igvn(region);
3296
3297 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3298 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3299 int class_klass_offset = java_lang_Class::klass_offset();
3300
3301 // First null-check both mirrors and load each mirror's klass metaobject.
3302 int which_arg;
3303 for (which_arg = 0; which_arg <= 1; which_arg++) {
3304 Node* arg = args[which_arg];
3305 arg = null_check(arg);
3306 if (stopped()) break;
3307 args[which_arg] = arg;
3308
3309 Node* p = basic_plus_adr(arg, class_klass_offset);
3310 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3311 klasses[which_arg] = _gvn.transform(kls);
3312 }
3313
3314 // Having loaded both klasses, test each for null.
3315 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3316 for (which_arg = 0; which_arg <= 1; which_arg++) {
3317 Node* kls = klasses[which_arg];
3318 Node* null_ctl = top();
3319 kls = null_check_oop(kls, &null_ctl, never_see_null);
3320 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3321 region->init_req(prim_path, null_ctl);
3322 if (stopped()) break;
3323 klasses[which_arg] = kls;
3324 }
3325
3326 if (!stopped()) {
3327 // now we have two reference types, in klasses[0..1]
3328 Node* subk = klasses[1]; // the argument to isAssignableFrom
3329 Node* superk = klasses[0]; // the receiver
3330 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3331 // now we have a successful reference subtype check
3332 region->set_req(_ref_subtype_path, control());
3333 }
3334
3335 // If both operands are primitive (both klasses null), then
3336 // we must return true when they are identical primitives.
3337 // It is convenient to test this after the first null klass check.
3338 set_control(region->in(_prim_0_path)); // go back to first null check
3339 if (!stopped()) {
3340 // Since superc is primitive, make a guard for the superc==subc case.
3341 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3342 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3343 generate_guard(bol_eq, region, PROB_FAIR);
3344 if (region->req() == PATH_LIMIT+1) {
3345 // A guard was added. If the added guard is taken, superc==subc.
3346 region->swap_edges(PATH_LIMIT, _prim_same_path);
3347 region->del_req(PATH_LIMIT);
3348 }
3349 region->set_req(_prim_0_path, control()); // Not equal after all.
3350 }
3351
3352 // these are the only paths that produce 'true':
3353 phi->set_req(_prim_same_path, intcon(1));
3354 phi->set_req(_ref_subtype_path, intcon(1));
3355
3356 // pull together the cases:
3357 assert(region->req() == PATH_LIMIT, "sane region");
3358 for (uint i = 1; i < region->req(); i++) {
3359 Node* ctl = region->in(i);
3360 if (ctl == NULL || ctl == top()) {
3361 region->set_req(i, top());
3362 phi ->set_req(i, top());
3363 } else if (phi->in(i) == NULL) {
3364 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3365 }
3366 }
3367
3368 set_control(_gvn.transform(region));
3369 set_result(_gvn.transform(phi));
3370 return true;
3371 }
3372
3373 //---------------------generate_array_guard_common------------------------
3374 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3375 bool obj_array, bool not_array) {
3376
3377 if (stopped()) {
3378 return NULL;
3379 }
3380
3381 // If obj_array/non_array==false/false:
3382 // Branch around if the given klass is in fact an array (either obj or prim).
3383 // If obj_array/non_array==false/true:
3384 // Branch around if the given klass is not an array klass of any kind.
3385 // If obj_array/non_array==true/true:
3386 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3387 // If obj_array/non_array==true/false:
3388 // Branch around if the kls is an oop array (Object[] or subtype)
3389 //
3390 // Like generate_guard, adds a new path onto the region.
3391 jint layout_con = 0;
3392 Node* layout_val = get_layout_helper(kls, layout_con);
3393 if (layout_val == NULL) {
3394 bool query = (obj_array
3395 ? Klass::layout_helper_is_objArray(layout_con)
3396 : Klass::layout_helper_is_array(layout_con));
3397 if (query == not_array) {
3398 return NULL; // never a branch
3399 } else { // always a branch
3400 Node* always_branch = control();
3401 if (region != NULL)
3402 region->add_req(always_branch);
3403 set_control(top());
3404 return always_branch;
3405 }
3406 }
3407 // Now test the correct condition.
3408 jint nval = (obj_array
3409 ? (jint)(Klass::_lh_array_tag_type_value
3410 << Klass::_lh_array_tag_shift)
3411 : Klass::_lh_neutral_value);
3412 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3413 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3414 // invert the test if we are looking for a non-array
3415 if (not_array) btest = BoolTest(btest).negate();
3416 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3417 return generate_fair_guard(bol, region);
3418 }
3419
3420
3421 //-----------------------inline_native_newArray--------------------------
3422 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3423 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3424 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3425 Node* mirror;
3426 Node* count_val;
3427 if (uninitialized) {
3428 mirror = argument(1);
3429 count_val = argument(2);
3430 } else {
3431 mirror = argument(0);
3432 count_val = argument(1);
3433 }
3434
3435 mirror = null_check(mirror);
3436 // If mirror or obj is dead, only null-path is taken.
3437 if (stopped()) return true;
3438
3439 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3440 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3441 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3442 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3443 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3444
3445 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3446 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3447 result_reg, _slow_path);
3448 Node* normal_ctl = control();
3449 Node* no_array_ctl = result_reg->in(_slow_path);
3450
3451 // Generate code for the slow case. We make a call to newArray().
3452 set_control(no_array_ctl);
3453 if (!stopped()) {
3454 // Either the input type is void.class, or else the
3455 // array klass has not yet been cached. Either the
3456 // ensuing call will throw an exception, or else it
3457 // will cache the array klass for next time.
3458 PreserveJVMState pjvms(this);
3459 CallJavaNode* slow_call = NULL;
3460 if (uninitialized) {
3461 // Generate optimized virtual call (holder class 'Unsafe' is final)
3462 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3463 } else {
3464 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3465 }
3466 Node* slow_result = set_results_for_java_call(slow_call);
3467 // this->control() comes from set_results_for_java_call
3468 result_reg->set_req(_slow_path, control());
3469 result_val->set_req(_slow_path, slow_result);
3470 result_io ->set_req(_slow_path, i_o());
3471 result_mem->set_req(_slow_path, reset_memory());
3472 }
3473
3474 set_control(normal_ctl);
3475 if (!stopped()) {
3476 // Normal case: The array type has been cached in the java.lang.Class.
3477 // The following call works fine even if the array type is polymorphic.
3478 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3479 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3480 result_reg->init_req(_normal_path, control());
3481 result_val->init_req(_normal_path, obj);
3482 result_io ->init_req(_normal_path, i_o());
3483 result_mem->init_req(_normal_path, reset_memory());
3484
3485 if (uninitialized) {
3486 // Mark the allocation so that zeroing is skipped
3487 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3488 alloc->maybe_set_complete(&_gvn);
3489 }
3490 }
3491
3492 // Return the combined state.
3493 set_i_o( _gvn.transform(result_io) );
3494 set_all_memory( _gvn.transform(result_mem));
3495
3496 C->set_has_split_ifs(true); // Has chance for split-if optimization
3497 set_result(result_reg, result_val);
3498 return true;
3499 }
3500
3501 //----------------------inline_native_getLength--------------------------
3502 // public static native int java.lang.reflect.Array.getLength(Object array);
3503 bool LibraryCallKit::inline_native_getLength() {
3504 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3505
3506 Node* array = null_check(argument(0));
3507 // If array is dead, only null-path is taken.
3508 if (stopped()) return true;
3509
3510 // Deoptimize if it is a non-array.
3511 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3512
3513 if (non_array != NULL) {
3514 PreserveJVMState pjvms(this);
3515 set_control(non_array);
3516 uncommon_trap(Deoptimization::Reason_intrinsic,
3517 Deoptimization::Action_maybe_recompile);
3518 }
3519
3520 // If control is dead, only non-array-path is taken.
3521 if (stopped()) return true;
3522
3523 // The works fine even if the array type is polymorphic.
3524 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3525 Node* result = load_array_length(array);
3526
3527 C->set_has_split_ifs(true); // Has chance for split-if optimization
3528 set_result(result);
3529 return true;
3530 }
3531
3532 //------------------------inline_array_copyOf----------------------------
3533 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3534 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3535 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3536 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3537
3538 // Get the arguments.
3539 Node* original = argument(0);
3540 Node* start = is_copyOfRange? argument(1): intcon(0);
3541 Node* end = is_copyOfRange? argument(2): argument(1);
3542 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3543
3544 Node* newcopy = NULL;
3545
3546 // Set the original stack and the reexecute bit for the interpreter to reexecute
3547 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3548 { PreserveReexecuteState preexecs(this);
3549 jvms()->set_should_reexecute(true);
3550
3551 array_type_mirror = null_check(array_type_mirror);
3552 original = null_check(original);
3553
3554 // Check if a null path was taken unconditionally.
3555 if (stopped()) return true;
3556
3557 Node* orig_length = load_array_length(original);
3558
3559 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3560 klass_node = null_check(klass_node);
3561
3562 RegionNode* bailout = new RegionNode(1);
3563 record_for_igvn(bailout);
3564
3565 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3566 // Bail out if that is so.
3567 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3568 if (not_objArray != NULL) {
3569 // Improve the klass node's type from the new optimistic assumption:
3570 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3571 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3572 Node* cast = new CastPPNode(klass_node, akls);
3573 cast->init_req(0, control());
3574 klass_node = _gvn.transform(cast);
3575 }
3576
3577 // Bail out if either start or end is negative.
3578 generate_negative_guard(start, bailout, &start);
3579 generate_negative_guard(end, bailout, &end);
3580
3581 Node* length = end;
3582 if (_gvn.type(start) != TypeInt::ZERO) {
3583 length = _gvn.transform(new SubINode(end, start));
3584 }
3585
3586 // Bail out if length is negative.
3587 // Without this the new_array would throw
3588 // NegativeArraySizeException but IllegalArgumentException is what
3589 // should be thrown
3590 generate_negative_guard(length, bailout, &length);
3591
3592 if (bailout->req() > 1) {
3593 PreserveJVMState pjvms(this);
3594 set_control(_gvn.transform(bailout));
3595 uncommon_trap(Deoptimization::Reason_intrinsic,
3596 Deoptimization::Action_maybe_recompile);
3597 }
3598
3599 if (!stopped()) {
3600 // How many elements will we copy from the original?
3601 // The answer is MinI(orig_length - start, length).
3602 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3603 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3604
3605 // Generate a direct call to the right arraycopy function(s).
3606 // We know the copy is disjoint but we might not know if the
3607 // oop stores need checking.
3608 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3609 // This will fail a store-check if x contains any non-nulls.
3610
3611 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3612 // loads/stores but it is legal only if we're sure the
3613 // Arrays.copyOf would succeed. So we need all input arguments
3614 // to the copyOf to be validated, including that the copy to the
3615 // new array won't trigger an ArrayStoreException. That subtype
3616 // check can be optimized if we know something on the type of
3617 // the input array from type speculation.
3618 if (_gvn.type(klass_node)->singleton()) {
3619 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3620 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3621
3622 int test = C->static_subtype_check(superk, subk);
3623 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3624 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3625 if (t_original->speculative_type() != NULL) {
3626 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3627 }
3628 }
3629 }
3630
3631 bool validated = false;
3632 // Reason_class_check rather than Reason_intrinsic because we
3633 // want to intrinsify even if this traps.
3634 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3635 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3636
3637 if (not_subtype_ctrl != top()) {
3638 PreserveJVMState pjvms(this);
3639 set_control(not_subtype_ctrl);
3640 uncommon_trap(Deoptimization::Reason_class_check,
3641 Deoptimization::Action_make_not_entrant);
3642 assert(stopped(), "Should be stopped");
3643 }
3644 validated = true;
3645 }
3646
3647 if (!stopped()) {
3648 newcopy = new_array(klass_node, length, 0); // no arguments to push
3649
3650 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3651 load_object_klass(original), klass_node);
3652 if (!is_copyOfRange) {
3653 ac->set_copyof(validated);
3654 } else {
3655 ac->set_copyofrange(validated);
3656 }
3657 Node* n = _gvn.transform(ac);
3658 if (n == ac) {
3659 ac->connect_outputs(this);
3660 } else {
3661 assert(validated, "shouldn't transform if all arguments not validated");
3662 set_all_memory(n);
3663 }
3664 }
3665 }
3666 } // original reexecute is set back here
3667
3668 C->set_has_split_ifs(true); // Has chance for split-if optimization
3669 if (!stopped()) {
3670 set_result(newcopy);
3671 }
3672 return true;
3673 }
3674
3675
3676 //----------------------generate_virtual_guard---------------------------
3677 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3678 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3679 RegionNode* slow_region) {
3680 ciMethod* method = callee();
3681 int vtable_index = method->vtable_index();
3682 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3683 "bad index %d", vtable_index);
3684 // Get the Method* out of the appropriate vtable entry.
3685 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3686 vtable_index*vtableEntry::size_in_bytes() +
3687 vtableEntry::method_offset_in_bytes();
3688 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3689 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3690
3691 // Compare the target method with the expected method (e.g., Object.hashCode).
3692 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3693
3694 Node* native_call = makecon(native_call_addr);
3695 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3696 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3697
3698 return generate_slow_guard(test_native, slow_region);
3699 }
3700
3701 //-----------------------generate_method_call----------------------------
3702 // Use generate_method_call to make a slow-call to the real
3703 // method if the fast path fails. An alternative would be to
3704 // use a stub like OptoRuntime::slow_arraycopy_Java.
3705 // This only works for expanding the current library call,
3706 // not another intrinsic. (E.g., don't use this for making an
3707 // arraycopy call inside of the copyOf intrinsic.)
3708 CallJavaNode*
3709 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3710 // When compiling the intrinsic method itself, do not use this technique.
3711 guarantee(callee() != C->method(), "cannot make slow-call to self");
3712
3713 ciMethod* method = callee();
3714 // ensure the JVMS we have will be correct for this call
3715 guarantee(method_id == method->intrinsic_id(), "must match");
3716
3717 const TypeFunc* tf = TypeFunc::make(method);
3718 CallJavaNode* slow_call;
3719 if (is_static) {
3720 assert(!is_virtual, "");
3721 slow_call = new CallStaticJavaNode(C, tf,
3722 SharedRuntime::get_resolve_static_call_stub(), method);
3723 } else if (is_virtual) {
3724 null_check_receiver();
3725 int vtable_index = Method::invalid_vtable_index;
3726 if (UseInlineCaches) {
3727 // Suppress the vtable call
3728 } else {
3729 // hashCode and clone are not a miranda methods,
3730 // so the vtable index is fixed.
3731 // No need to use the linkResolver to get it.
3732 vtable_index = method->vtable_index();
3733 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3734 "bad index %d", vtable_index);
3735 }
3736 slow_call = new CallDynamicJavaNode(tf,
3737 SharedRuntime::get_resolve_virtual_call_stub(),
3738 method, vtable_index);
3739 } else { // neither virtual nor static: opt_virtual
3740 null_check_receiver();
3741 slow_call = new CallStaticJavaNode(C, tf,
3742 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
3743 slow_call->set_optimized_virtual(true);
3744 }
3745 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3746 // To be able to issue a direct call (optimized virtual or virtual)
3747 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3748 // about the method being invoked should be attached to the call site to
3749 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3750 slow_call->set_override_symbolic_info(true);
3751 }
3752 set_arguments_for_java_call(slow_call);
3753 set_edges_for_java_call(slow_call);
3754 return slow_call;
3755 }
3756
3757
3758 /**
3759 * Build special case code for calls to hashCode on an object. This call may
3760 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3761 * slightly different code.
3762 */
3763 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3764 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3765 assert(!(is_virtual && is_static), "either virtual, special, or static");
3766
3767 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3768
3769 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3770 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3771 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3772 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3773 Node* obj = NULL;
3774 if (!is_static) {
3775 // Check for hashing null object
3776 obj = null_check_receiver();
3777 if (stopped()) return true; // unconditionally null
3778 result_reg->init_req(_null_path, top());
3779 result_val->init_req(_null_path, top());
3780 } else {
3781 // Do a null check, and return zero if null.
3782 // System.identityHashCode(null) == 0
3783 obj = argument(0);
3784 Node* null_ctl = top();
3785 obj = null_check_oop(obj, &null_ctl);
3786 result_reg->init_req(_null_path, null_ctl);
3787 result_val->init_req(_null_path, _gvn.intcon(0));
3788 }
3789
3790 // Unconditionally null? Then return right away.
3791 if (stopped()) {
3792 set_control( result_reg->in(_null_path));
3793 if (!stopped())
3794 set_result(result_val->in(_null_path));
3795 return true;
3796 }
3797
3798 // We only go to the fast case code if we pass a number of guards. The
3799 // paths which do not pass are accumulated in the slow_region.
3800 RegionNode* slow_region = new RegionNode(1);
3801 record_for_igvn(slow_region);
3802
3803 // If this is a virtual call, we generate a funny guard. We pull out
3804 // the vtable entry corresponding to hashCode() from the target object.
3805 // If the target method which we are calling happens to be the native
3806 // Object hashCode() method, we pass the guard. We do not need this
3807 // guard for non-virtual calls -- the caller is known to be the native
3808 // Object hashCode().
3809 if (is_virtual) {
3810 // After null check, get the object's klass.
3811 Node* obj_klass = load_object_klass(obj);
3812 generate_virtual_guard(obj_klass, slow_region);
3813 }
3814
3815 // Get the header out of the object, use LoadMarkNode when available
3816 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3817 // The control of the load must be NULL. Otherwise, the load can move before
3818 // the null check after castPP removal.
3819 Node* no_ctrl = NULL;
3820 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3821
3822 // Test the header to see if it is unlocked.
3823 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3824 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3825 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3826 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3827 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3828
3829 generate_slow_guard(test_unlocked, slow_region);
3830
3831 // Get the hash value and check to see that it has been properly assigned.
3832 // We depend on hash_mask being at most 32 bits and avoid the use of
3833 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3834 // vm: see markWord.hpp.
3835 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3836 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3837 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3838 // This hack lets the hash bits live anywhere in the mark object now, as long
3839 // as the shift drops the relevant bits into the low 32 bits. Note that
3840 // Java spec says that HashCode is an int so there's no point in capturing
3841 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3842 hshifted_header = ConvX2I(hshifted_header);
3843 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3844
3845 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3846 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3847 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3848
3849 generate_slow_guard(test_assigned, slow_region);
3850
3851 Node* init_mem = reset_memory();
3852 // fill in the rest of the null path:
3853 result_io ->init_req(_null_path, i_o());
3854 result_mem->init_req(_null_path, init_mem);
3855
3856 result_val->init_req(_fast_path, hash_val);
3857 result_reg->init_req(_fast_path, control());
3858 result_io ->init_req(_fast_path, i_o());
3859 result_mem->init_req(_fast_path, init_mem);
3860
3861 // Generate code for the slow case. We make a call to hashCode().
3862 set_control(_gvn.transform(slow_region));
3863 if (!stopped()) {
3864 // No need for PreserveJVMState, because we're using up the present state.
3865 set_all_memory(init_mem);
3866 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3867 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3868 Node* slow_result = set_results_for_java_call(slow_call);
3869 // this->control() comes from set_results_for_java_call
3870 result_reg->init_req(_slow_path, control());
3871 result_val->init_req(_slow_path, slow_result);
3872 result_io ->set_req(_slow_path, i_o());
3873 result_mem ->set_req(_slow_path, reset_memory());
3874 }
3875
3876 // Return the combined state.
3877 set_i_o( _gvn.transform(result_io) );
3878 set_all_memory( _gvn.transform(result_mem));
3879
3880 set_result(result_reg, result_val);
3881 return true;
3882 }
3883
3884 //---------------------------inline_native_getClass----------------------------
3885 // public final native Class<?> java.lang.Object.getClass();
3886 //
3887 // Build special case code for calls to getClass on an object.
3888 bool LibraryCallKit::inline_native_getClass() {
3889 Node* obj = null_check_receiver();
3890 if (stopped()) return true;
3891 set_result(load_mirror_from_klass(load_object_klass(obj)));
3892 return true;
3893 }
3894
3895 //-----------------inline_native_Reflection_getCallerClass---------------------
3896 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3897 //
3898 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3899 //
3900 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3901 // in that it must skip particular security frames and checks for
3902 // caller sensitive methods.
3903 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3904 #ifndef PRODUCT
3905 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3906 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3907 }
3908 #endif
3909
3910 if (!jvms()->has_method()) {
3911 #ifndef PRODUCT
3912 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3913 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
3914 }
3915 #endif
3916 return false;
3917 }
3918
3919 // Walk back up the JVM state to find the caller at the required
3920 // depth.
3921 JVMState* caller_jvms = jvms();
3922
3923 // Cf. JVM_GetCallerClass
3924 // NOTE: Start the loop at depth 1 because the current JVM state does
3925 // not include the Reflection.getCallerClass() frame.
3926 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
3927 ciMethod* m = caller_jvms->method();
3928 switch (n) {
3929 case 0:
3930 fatal("current JVM state does not include the Reflection.getCallerClass frame");
3931 break;
3932 case 1:
3933 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
3934 if (!m->caller_sensitive()) {
3935 #ifndef PRODUCT
3936 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3937 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
3938 }
3939 #endif
3940 return false; // bail-out; let JVM_GetCallerClass do the work
3941 }
3942 break;
3943 default:
3944 if (!m->is_ignored_by_security_stack_walk()) {
3945 // We have reached the desired frame; return the holder class.
3946 // Acquire method holder as java.lang.Class and push as constant.
3947 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
3948 ciInstance* caller_mirror = caller_klass->java_mirror();
3949 set_result(makecon(TypeInstPtr::make(caller_mirror)));
3950
3951 #ifndef PRODUCT
3952 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3953 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());
3954 tty->print_cr(" JVM state at this point:");
3955 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
3956 ciMethod* m = jvms()->of_depth(i)->method();
3957 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
3958 }
3959 }
3960 #endif
3961 return true;
3962 }
3963 break;
3964 }
3965 }
3966
3967 #ifndef PRODUCT
3968 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3969 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
3970 tty->print_cr(" JVM state at this point:");
3971 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
3972 ciMethod* m = jvms()->of_depth(i)->method();
3973 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
3974 }
3975 }
3976 #endif
3977
3978 return false; // bail-out; let JVM_GetCallerClass do the work
3979 }
3980
3981 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
3982 Node* arg = argument(0);
3983 Node* result = NULL;
3984
3985 switch (id) {
3986 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
3987 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
3988 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
3989 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
3990
3991 case vmIntrinsics::_doubleToLongBits: {
3992 // two paths (plus control) merge in a wood
3993 RegionNode *r = new RegionNode(3);
3994 Node *phi = new PhiNode(r, TypeLong::LONG);
3995
3996 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
3997 // Build the boolean node
3998 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
3999
4000 // Branch either way.
4001 // NaN case is less traveled, which makes all the difference.
4002 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4003 Node *opt_isnan = _gvn.transform(ifisnan);
4004 assert( opt_isnan->is_If(), "Expect an IfNode");
4005 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4006 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4007
4008 set_control(iftrue);
4009
4010 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4011 Node *slow_result = longcon(nan_bits); // return NaN
4012 phi->init_req(1, _gvn.transform( slow_result ));
4013 r->init_req(1, iftrue);
4014
4015 // Else fall through
4016 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4017 set_control(iffalse);
4018
4019 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4020 r->init_req(2, iffalse);
4021
4022 // Post merge
4023 set_control(_gvn.transform(r));
4024 record_for_igvn(r);
4025
4026 C->set_has_split_ifs(true); // Has chance for split-if optimization
4027 result = phi;
4028 assert(result->bottom_type()->isa_long(), "must be");
4029 break;
4030 }
4031
4032 case vmIntrinsics::_floatToIntBits: {
4033 // two paths (plus control) merge in a wood
4034 RegionNode *r = new RegionNode(3);
4035 Node *phi = new PhiNode(r, TypeInt::INT);
4036
4037 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4038 // Build the boolean node
4039 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4040
4041 // Branch either way.
4042 // NaN case is less traveled, which makes all the difference.
4043 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4044 Node *opt_isnan = _gvn.transform(ifisnan);
4045 assert( opt_isnan->is_If(), "Expect an IfNode");
4046 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4047 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4048
4049 set_control(iftrue);
4050
4051 static const jint nan_bits = 0x7fc00000;
4052 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4053 phi->init_req(1, _gvn.transform( slow_result ));
4054 r->init_req(1, iftrue);
4055
4056 // Else fall through
4057 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4058 set_control(iffalse);
4059
4060 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4061 r->init_req(2, iffalse);
4062
4063 // Post merge
4064 set_control(_gvn.transform(r));
4065 record_for_igvn(r);
4066
4067 C->set_has_split_ifs(true); // Has chance for split-if optimization
4068 result = phi;
4069 assert(result->bottom_type()->isa_int(), "must be");
4070 break;
4071 }
4072
4073 default:
4074 fatal_unexpected_iid(id);
4075 break;
4076 }
4077 set_result(_gvn.transform(result));
4078 return true;
4079 }
4080
4081 //----------------------inline_unsafe_copyMemory-------------------------
4082 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4083 bool LibraryCallKit::inline_unsafe_copyMemory() {
4084 if (callee()->is_static()) return false; // caller must have the capability!
4085 null_check_receiver(); // null-check receiver
4086 if (stopped()) return true;
4087
4088 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4089
4090 Node* src_ptr = argument(1); // type: oop
4091 Node* src_off = ConvL2X(argument(2)); // type: long
4092 Node* dst_ptr = argument(4); // type: oop
4093 Node* dst_off = ConvL2X(argument(5)); // type: long
4094 Node* size = ConvL2X(argument(7)); // type: long
4095
4096 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4097 "fieldOffset must be byte-scaled");
4098
4099 Node* src = make_unsafe_address(src_ptr, src_off);
4100 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4101
4102 // Conservatively insert a memory barrier on all memory slices.
4103 // Do not let writes of the copy source or destination float below the copy.
4104 insert_mem_bar(Op_MemBarCPUOrder);
4105
4106 Node* thread = _gvn.transform(new ThreadLocalNode());
4107 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4108 BasicType doing_unsafe_access_bt = T_BYTE;
4109 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4110
4111 // update volatile field
4112 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4113
4114 // Call it. Note that the length argument is not scaled.
4115 make_runtime_call(RC_LEAF|RC_NO_FP,
4116 OptoRuntime::fast_arraycopy_Type(),
4117 StubRoutines::unsafe_arraycopy(),
4118 "unsafe_arraycopy",
4119 TypeRawPtr::BOTTOM,
4120 src, dst, size XTOP);
4121
4122 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4123
4124 // Do not let reads of the copy destination float above the copy.
4125 insert_mem_bar(Op_MemBarCPUOrder);
4126
4127 return true;
4128 }
4129
4130 #undef XTOP
4131
4132 //------------------------clone_coping-----------------------------------
4133 // Helper function for inline_native_clone.
4134 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4135 assert(obj_size != NULL, "");
4136 Node* raw_obj = alloc_obj->in(1);
4137 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4138
4139 AllocateNode* alloc = NULL;
4140 if (ReduceBulkZeroing) {
4141 // We will be completely responsible for initializing this object -
4142 // mark Initialize node as complete.
4143 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4144 // The object was just allocated - there should be no any stores!
4145 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4146 // Mark as complete_with_arraycopy so that on AllocateNode
4147 // expansion, we know this AllocateNode is initialized by an array
4148 // copy and a StoreStore barrier exists after the array copy.
4149 alloc->initialization()->set_complete_with_arraycopy();
4150 }
4151
4152 Node* size = _gvn.transform(obj_size);
4153 access_clone(obj, alloc_obj, size, is_array);
4154
4155 // Do not let reads from the cloned object float above the arraycopy.
4156 if (alloc != NULL) {
4157 // Do not let stores that initialize this object be reordered with
4158 // a subsequent store that would make this object accessible by
4159 // other threads.
4160 // Record what AllocateNode this StoreStore protects so that
4161 // escape analysis can go from the MemBarStoreStoreNode to the
4162 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4163 // based on the escape status of the AllocateNode.
4164 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4165 } else {
4166 insert_mem_bar(Op_MemBarCPUOrder);
4167 }
4168 }
4169
4170 //------------------------inline_native_clone----------------------------
4171 // protected native Object java.lang.Object.clone();
4172 //
4173 // Here are the simple edge cases:
4174 // null receiver => normal trap
4175 // virtual and clone was overridden => slow path to out-of-line clone
4176 // not cloneable or finalizer => slow path to out-of-line Object.clone
4177 //
4178 // The general case has two steps, allocation and copying.
4179 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4180 //
4181 // Copying also has two cases, oop arrays and everything else.
4182 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4183 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4184 //
4185 // These steps fold up nicely if and when the cloned object's klass
4186 // can be sharply typed as an object array, a type array, or an instance.
4187 //
4188 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4189 PhiNode* result_val;
4190
4191 // Set the reexecute bit for the interpreter to reexecute
4192 // the bytecode that invokes Object.clone if deoptimization happens.
4193 { PreserveReexecuteState preexecs(this);
4194 jvms()->set_should_reexecute(true);
4195
4196 Node* obj = null_check_receiver();
4197 if (stopped()) return true;
4198
4199 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4200
4201 // If we are going to clone an instance, we need its exact type to
4202 // know the number and types of fields to convert the clone to
4203 // loads/stores. Maybe a speculative type can help us.
4204 if (!obj_type->klass_is_exact() &&
4205 obj_type->speculative_type() != NULL &&
4206 obj_type->speculative_type()->is_instance_klass()) {
4207 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4208 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4209 !spec_ik->has_injected_fields()) {
4210 ciKlass* k = obj_type->klass();
4211 if (!k->is_instance_klass() ||
4212 k->as_instance_klass()->is_interface() ||
4213 k->as_instance_klass()->has_subklass()) {
4214 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4215 }
4216 }
4217 }
4218
4219 // Conservatively insert a memory barrier on all memory slices.
4220 // Do not let writes into the original float below the clone.
4221 insert_mem_bar(Op_MemBarCPUOrder);
4222
4223 // paths into result_reg:
4224 enum {
4225 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4226 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4227 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4228 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4229 PATH_LIMIT
4230 };
4231 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4232 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4233 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4234 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4235 record_for_igvn(result_reg);
4236
4237 Node* obj_klass = load_object_klass(obj);
4238 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4239 if (array_ctl != NULL) {
4240 // It's an array.
4241 PreserveJVMState pjvms(this);
4242 set_control(array_ctl);
4243 Node* obj_length = load_array_length(obj);
4244 Node* obj_size = NULL;
4245 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4246
4247 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4248 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4249 // If it is an oop array, it requires very special treatment,
4250 // because gc barriers are required when accessing the array.
4251 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4252 if (is_obja != NULL) {
4253 PreserveJVMState pjvms2(this);
4254 set_control(is_obja);
4255 // Generate a direct call to the right arraycopy function(s).
4256 // Clones are always tightly coupled.
4257 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4258 ac->set_clone_oop_array();
4259 Node* n = _gvn.transform(ac);
4260 assert(n == ac, "cannot disappear");
4261 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4262
4263 result_reg->init_req(_objArray_path, control());
4264 result_val->init_req(_objArray_path, alloc_obj);
4265 result_i_o ->set_req(_objArray_path, i_o());
4266 result_mem ->set_req(_objArray_path, reset_memory());
4267 }
4268 }
4269 // Otherwise, there are no barriers to worry about.
4270 // (We can dispense with card marks if we know the allocation
4271 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4272 // causes the non-eden paths to take compensating steps to
4273 // simulate a fresh allocation, so that no further
4274 // card marks are required in compiled code to initialize
4275 // the object.)
4276
4277 if (!stopped()) {
4278 copy_to_clone(obj, alloc_obj, obj_size, true);
4279
4280 // Present the results of the copy.
4281 result_reg->init_req(_array_path, control());
4282 result_val->init_req(_array_path, alloc_obj);
4283 result_i_o ->set_req(_array_path, i_o());
4284 result_mem ->set_req(_array_path, reset_memory());
4285 }
4286 }
4287
4288 // We only go to the instance fast case code if we pass a number of guards.
4289 // The paths which do not pass are accumulated in the slow_region.
4290 RegionNode* slow_region = new RegionNode(1);
4291 record_for_igvn(slow_region);
4292 if (!stopped()) {
4293 // It's an instance (we did array above). Make the slow-path tests.
4294 // If this is a virtual call, we generate a funny guard. We grab
4295 // the vtable entry corresponding to clone() from the target object.
4296 // If the target method which we are calling happens to be the
4297 // Object clone() method, we pass the guard. We do not need this
4298 // guard for non-virtual calls; the caller is known to be the native
4299 // Object clone().
4300 if (is_virtual) {
4301 generate_virtual_guard(obj_klass, slow_region);
4302 }
4303
4304 // The object must be easily cloneable and must not have a finalizer.
4305 // Both of these conditions may be checked in a single test.
4306 // We could optimize the test further, but we don't care.
4307 generate_access_flags_guard(obj_klass,
4308 // Test both conditions:
4309 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4310 // Must be cloneable but not finalizer:
4311 JVM_ACC_IS_CLONEABLE_FAST,
4312 slow_region);
4313 }
4314
4315 if (!stopped()) {
4316 // It's an instance, and it passed the slow-path tests.
4317 PreserveJVMState pjvms(this);
4318 Node* obj_size = NULL;
4319 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4320 // is reexecuted if deoptimization occurs and there could be problems when merging
4321 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4322 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4323
4324 copy_to_clone(obj, alloc_obj, obj_size, false);
4325
4326 // Present the results of the slow call.
4327 result_reg->init_req(_instance_path, control());
4328 result_val->init_req(_instance_path, alloc_obj);
4329 result_i_o ->set_req(_instance_path, i_o());
4330 result_mem ->set_req(_instance_path, reset_memory());
4331 }
4332
4333 // Generate code for the slow case. We make a call to clone().
4334 set_control(_gvn.transform(slow_region));
4335 if (!stopped()) {
4336 PreserveJVMState pjvms(this);
4337 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4338 // We need to deoptimize on exception (see comment above)
4339 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4340 // this->control() comes from set_results_for_java_call
4341 result_reg->init_req(_slow_path, control());
4342 result_val->init_req(_slow_path, slow_result);
4343 result_i_o ->set_req(_slow_path, i_o());
4344 result_mem ->set_req(_slow_path, reset_memory());
4345 }
4346
4347 // Return the combined state.
4348 set_control( _gvn.transform(result_reg));
4349 set_i_o( _gvn.transform(result_i_o));
4350 set_all_memory( _gvn.transform(result_mem));
4351 } // original reexecute is set back here
4352
4353 set_result(_gvn.transform(result_val));
4354 return true;
4355 }
4356
4357 // If we have a tightly coupled allocation, the arraycopy may take care
4358 // of the array initialization. If one of the guards we insert between
4359 // the allocation and the arraycopy causes a deoptimization, an
4360 // unitialized array will escape the compiled method. To prevent that
4361 // we set the JVM state for uncommon traps between the allocation and
4362 // the arraycopy to the state before the allocation so, in case of
4363 // deoptimization, we'll reexecute the allocation and the
4364 // initialization.
4365 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4366 if (alloc != NULL) {
4367 ciMethod* trap_method = alloc->jvms()->method();
4368 int trap_bci = alloc->jvms()->bci();
4369
4370 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4371 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4372 // Make sure there's no store between the allocation and the
4373 // arraycopy otherwise visible side effects could be rexecuted
4374 // in case of deoptimization and cause incorrect execution.
4375 bool no_interfering_store = true;
4376 Node* mem = alloc->in(TypeFunc::Memory);
4377 if (mem->is_MergeMem()) {
4378 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4379 Node* n = mms.memory();
4380 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4381 assert(n->is_Store(), "what else?");
4382 no_interfering_store = false;
4383 break;
4384 }
4385 }
4386 } else {
4387 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4388 Node* n = mms.memory();
4389 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4390 assert(n->is_Store(), "what else?");
4391 no_interfering_store = false;
4392 break;
4393 }
4394 }
4395 }
4396
4397 if (no_interfering_store) {
4398 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4399 uint size = alloc->req();
4400 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4401 old_jvms->set_map(sfpt);
4402 for (uint i = 0; i < size; i++) {
4403 sfpt->init_req(i, alloc->in(i));
4404 }
4405 // re-push array length for deoptimization
4406 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4407 old_jvms->set_sp(old_jvms->sp()+1);
4408 old_jvms->set_monoff(old_jvms->monoff()+1);
4409 old_jvms->set_scloff(old_jvms->scloff()+1);
4410 old_jvms->set_endoff(old_jvms->endoff()+1);
4411 old_jvms->set_should_reexecute(true);
4412
4413 sfpt->set_i_o(map()->i_o());
4414 sfpt->set_memory(map()->memory());
4415 sfpt->set_control(map()->control());
4416
4417 JVMState* saved_jvms = jvms();
4418 saved_reexecute_sp = _reexecute_sp;
4419
4420 set_jvms(sfpt->jvms());
4421 _reexecute_sp = jvms()->sp();
4422
4423 return saved_jvms;
4424 }
4425 }
4426 }
4427 return NULL;
4428 }
4429
4430 // In case of a deoptimization, we restart execution at the
4431 // allocation, allocating a new array. We would leave an uninitialized
4432 // array in the heap that GCs wouldn't expect. Move the allocation
4433 // after the traps so we don't allocate the array if we
4434 // deoptimize. This is possible because tightly_coupled_allocation()
4435 // guarantees there's no observer of the allocated array at this point
4436 // and the control flow is simple enough.
4437 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4438 int saved_reexecute_sp, uint new_idx) {
4439 if (saved_jvms != NULL && !stopped()) {
4440 assert(alloc != NULL, "only with a tightly coupled allocation");
4441 // restore JVM state to the state at the arraycopy
4442 saved_jvms->map()->set_control(map()->control());
4443 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4444 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4445 // If we've improved the types of some nodes (null check) while
4446 // emitting the guards, propagate them to the current state
4447 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4448 set_jvms(saved_jvms);
4449 _reexecute_sp = saved_reexecute_sp;
4450
4451 // Remove the allocation from above the guards
4452 CallProjections callprojs;
4453 alloc->extract_projections(&callprojs, true);
4454 InitializeNode* init = alloc->initialization();
4455 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4456 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4457 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4458
4459 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4460 // the allocation (i.e. is only valid if the allocation succeeds):
4461 // 1) replace CastIINode with AllocateArrayNode's length here
4462 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4463 //
4464 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4465 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4466 Node* init_control = init->proj_out(TypeFunc::Control);
4467 Node* alloc_length = alloc->Ideal_length();
4468 #ifdef ASSERT
4469 Node* prev_cast = NULL;
4470 #endif
4471 for (uint i = 0; i < init_control->outcnt(); i++) {
4472 Node* init_out = init_control->raw_out(i);
4473 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4474 #ifdef ASSERT
4475 if (prev_cast == NULL) {
4476 prev_cast = init_out;
4477 } else {
4478 if (prev_cast->cmp(*init_out) == false) {
4479 prev_cast->dump();
4480 init_out->dump();
4481 assert(false, "not equal CastIINode");
4482 }
4483 }
4484 #endif
4485 C->gvn_replace_by(init_out, alloc_length);
4486 }
4487 }
4488 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4489
4490 // move the allocation here (after the guards)
4491 _gvn.hash_delete(alloc);
4492 alloc->set_req(TypeFunc::Control, control());
4493 alloc->set_req(TypeFunc::I_O, i_o());
4494 Node *mem = reset_memory();
4495 set_all_memory(mem);
4496 alloc->set_req(TypeFunc::Memory, mem);
4497 set_control(init->proj_out_or_null(TypeFunc::Control));
4498 set_i_o(callprojs.fallthrough_ioproj);
4499
4500 // Update memory as done in GraphKit::set_output_for_allocation()
4501 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4502 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4503 if (ary_type->isa_aryptr() && length_type != NULL) {
4504 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4505 }
4506 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4507 int elemidx = C->get_alias_index(telemref);
4508 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4509 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4510
4511 Node* allocx = _gvn.transform(alloc);
4512 assert(allocx == alloc, "where has the allocation gone?");
4513 assert(dest->is_CheckCastPP(), "not an allocation result?");
4514
4515 _gvn.hash_delete(dest);
4516 dest->set_req(0, control());
4517 Node* destx = _gvn.transform(dest);
4518 assert(destx == dest, "where has the allocation result gone?");
4519
4520 array_ideal_length(alloc, ary_type, true);
4521 }
4522 }
4523
4524
4525 //------------------------------inline_arraycopy-----------------------
4526 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4527 // Object dest, int destPos,
4528 // int length);
4529 bool LibraryCallKit::inline_arraycopy() {
4530 // Get the arguments.
4531 Node* src = argument(0); // type: oop
4532 Node* src_offset = argument(1); // type: int
4533 Node* dest = argument(2); // type: oop
4534 Node* dest_offset = argument(3); // type: int
4535 Node* length = argument(4); // type: int
4536
4537 uint new_idx = C->unique();
4538
4539 // Check for allocation before we add nodes that would confuse
4540 // tightly_coupled_allocation()
4541 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
4542
4543 int saved_reexecute_sp = -1;
4544 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4545 // See arraycopy_restore_alloc_state() comment
4546 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4547 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4548 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4549 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4550
4551 // The following tests must be performed
4552 // (1) src and dest are arrays.
4553 // (2) src and dest arrays must have elements of the same BasicType
4554 // (3) src and dest must not be null.
4555 // (4) src_offset must not be negative.
4556 // (5) dest_offset must not be negative.
4557 // (6) length must not be negative.
4558 // (7) src_offset + length must not exceed length of src.
4559 // (8) dest_offset + length must not exceed length of dest.
4560 // (9) each element of an oop array must be assignable
4561
4562 // (3) src and dest must not be null.
4563 // always do this here because we need the JVM state for uncommon traps
4564 Node* null_ctl = top();
4565 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4566 assert(null_ctl->is_top(), "no null control here");
4567 dest = null_check(dest, T_ARRAY);
4568
4569 if (!can_emit_guards) {
4570 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4571 // guards but the arraycopy node could still take advantage of a
4572 // tightly allocated allocation. tightly_coupled_allocation() is
4573 // called again to make sure it takes the null check above into
4574 // account: the null check is mandatory and if it caused an
4575 // uncommon trap to be emitted then the allocation can't be
4576 // considered tightly coupled in this context.
4577 alloc = tightly_coupled_allocation(dest);
4578 }
4579
4580 bool validated = false;
4581
4582 const Type* src_type = _gvn.type(src);
4583 const Type* dest_type = _gvn.type(dest);
4584 const TypeAryPtr* top_src = src_type->isa_aryptr();
4585 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4586
4587 // Do we have the type of src?
4588 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4589 // Do we have the type of dest?
4590 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4591 // Is the type for src from speculation?
4592 bool src_spec = false;
4593 // Is the type for dest from speculation?
4594 bool dest_spec = false;
4595
4596 if ((!has_src || !has_dest) && can_emit_guards) {
4597 // We don't have sufficient type information, let's see if
4598 // speculative types can help. We need to have types for both src
4599 // and dest so that it pays off.
4600
4601 // Do we already have or could we have type information for src
4602 bool could_have_src = has_src;
4603 // Do we already have or could we have type information for dest
4604 bool could_have_dest = has_dest;
4605
4606 ciKlass* src_k = NULL;
4607 if (!has_src) {
4608 src_k = src_type->speculative_type_not_null();
4609 if (src_k != NULL && src_k->is_array_klass()) {
4610 could_have_src = true;
4611 }
4612 }
4613
4614 ciKlass* dest_k = NULL;
4615 if (!has_dest) {
4616 dest_k = dest_type->speculative_type_not_null();
4617 if (dest_k != NULL && dest_k->is_array_klass()) {
4618 could_have_dest = true;
4619 }
4620 }
4621
4622 if (could_have_src && could_have_dest) {
4623 // This is going to pay off so emit the required guards
4624 if (!has_src) {
4625 src = maybe_cast_profiled_obj(src, src_k, true);
4626 src_type = _gvn.type(src);
4627 top_src = src_type->isa_aryptr();
4628 has_src = (top_src != NULL && top_src->klass() != NULL);
4629 src_spec = true;
4630 }
4631 if (!has_dest) {
4632 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4633 dest_type = _gvn.type(dest);
4634 top_dest = dest_type->isa_aryptr();
4635 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4636 dest_spec = true;
4637 }
4638 }
4639 }
4640
4641 if (has_src && has_dest && can_emit_guards) {
4642 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4643 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4644 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4645 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4646
4647 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4648 // If both arrays are object arrays then having the exact types
4649 // for both will remove the need for a subtype check at runtime
4650 // before the call and may make it possible to pick a faster copy
4651 // routine (without a subtype check on every element)
4652 // Do we have the exact type of src?
4653 bool could_have_src = src_spec;
4654 // Do we have the exact type of dest?
4655 bool could_have_dest = dest_spec;
4656 ciKlass* src_k = top_src->klass();
4657 ciKlass* dest_k = top_dest->klass();
4658 if (!src_spec) {
4659 src_k = src_type->speculative_type_not_null();
4660 if (src_k != NULL && src_k->is_array_klass()) {
4661 could_have_src = true;
4662 }
4663 }
4664 if (!dest_spec) {
4665 dest_k = dest_type->speculative_type_not_null();
4666 if (dest_k != NULL && dest_k->is_array_klass()) {
4667 could_have_dest = true;
4668 }
4669 }
4670 if (could_have_src && could_have_dest) {
4671 // If we can have both exact types, emit the missing guards
4672 if (could_have_src && !src_spec) {
4673 src = maybe_cast_profiled_obj(src, src_k, true);
4674 }
4675 if (could_have_dest && !dest_spec) {
4676 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4677 }
4678 }
4679 }
4680 }
4681
4682 ciMethod* trap_method = method();
4683 int trap_bci = bci();
4684 if (saved_jvms != NULL) {
4685 trap_method = alloc->jvms()->method();
4686 trap_bci = alloc->jvms()->bci();
4687 }
4688
4689 bool negative_length_guard_generated = false;
4690
4691 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4692 can_emit_guards &&
4693 !src->is_top() && !dest->is_top()) {
4694 // validate arguments: enables transformation the ArrayCopyNode
4695 validated = true;
4696
4697 RegionNode* slow_region = new RegionNode(1);
4698 record_for_igvn(slow_region);
4699
4700 // (1) src and dest are arrays.
4701 generate_non_array_guard(load_object_klass(src), slow_region);
4702 generate_non_array_guard(load_object_klass(dest), slow_region);
4703
4704 // (2) src and dest arrays must have elements of the same BasicType
4705 // done at macro expansion or at Ideal transformation time
4706
4707 // (4) src_offset must not be negative.
4708 generate_negative_guard(src_offset, slow_region);
4709
4710 // (5) dest_offset must not be negative.
4711 generate_negative_guard(dest_offset, slow_region);
4712
4713 // (7) src_offset + length must not exceed length of src.
4714 generate_limit_guard(src_offset, length,
4715 load_array_length(src),
4716 slow_region);
4717
4718 // (8) dest_offset + length must not exceed length of dest.
4719 generate_limit_guard(dest_offset, length,
4720 load_array_length(dest),
4721 slow_region);
4722
4723 // (6) length must not be negative.
4724 // This is also checked in generate_arraycopy() during macro expansion, but
4725 // we also have to check it here for the case where the ArrayCopyNode will
4726 // be eliminated by Escape Analysis.
4727 if (EliminateAllocations) {
4728 generate_negative_guard(length, slow_region);
4729 negative_length_guard_generated = true;
4730 }
4731
4732 // (9) each element of an oop array must be assignable
4733 Node* dest_klass = load_object_klass(dest);
4734 if (src != dest) {
4735 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4736
4737 if (not_subtype_ctrl != top()) {
4738 PreserveJVMState pjvms(this);
4739 set_control(not_subtype_ctrl);
4740 uncommon_trap(Deoptimization::Reason_intrinsic,
4741 Deoptimization::Action_make_not_entrant);
4742 assert(stopped(), "Should be stopped");
4743 }
4744 }
4745 {
4746 PreserveJVMState pjvms(this);
4747 set_control(_gvn.transform(slow_region));
4748 uncommon_trap(Deoptimization::Reason_intrinsic,
4749 Deoptimization::Action_make_not_entrant);
4750 assert(stopped(), "Should be stopped");
4751 }
4752
4753 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4754 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4755 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4756 }
4757
4758 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4759
4760 if (stopped()) {
4761 return true;
4762 }
4763
4764 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4765 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4766 // so the compiler has a chance to eliminate them: during macro expansion,
4767 // we have to set their control (CastPP nodes are eliminated).
4768 load_object_klass(src), load_object_klass(dest),
4769 load_array_length(src), load_array_length(dest));
4770
4771 ac->set_arraycopy(validated);
4772
4773 Node* n = _gvn.transform(ac);
4774 if (n == ac) {
4775 ac->connect_outputs(this);
4776 } else {
4777 assert(validated, "shouldn't transform if all arguments not validated");
4778 set_all_memory(n);
4779 }
4780 clear_upper_avx();
4781
4782
4783 return true;
4784 }
4785
4786
4787 // Helper function which determines if an arraycopy immediately follows
4788 // an allocation, with no intervening tests or other escapes for the object.
4789 AllocateArrayNode*
4790 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
4791 if (stopped()) return NULL; // no fast path
4792 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4793
4794 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4795 if (alloc == NULL) return NULL;
4796
4797 Node* rawmem = memory(Compile::AliasIdxRaw);
4798 // Is the allocation's memory state untouched?
4799 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4800 // Bail out if there have been raw-memory effects since the allocation.
4801 // (Example: There might have been a call or safepoint.)
4802 return NULL;
4803 }
4804 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4805 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4806 return NULL;
4807 }
4808
4809 // There must be no unexpected observers of this allocation.
4810 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4811 Node* obs = ptr->fast_out(i);
4812 if (obs != this->map()) {
4813 return NULL;
4814 }
4815 }
4816
4817 // This arraycopy must unconditionally follow the allocation of the ptr.
4818 Node* alloc_ctl = ptr->in(0);
4819 Node* ctl = control();
4820 while (ctl != alloc_ctl) {
4821 // There may be guards which feed into the slow_region.
4822 // Any other control flow means that we might not get a chance
4823 // to finish initializing the allocated object.
4824 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4825 IfNode* iff = ctl->in(0)->as_If();
4826 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4827 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4828 // One more try: Various low-level checks bottom out in
4829 // uncommon traps. If the debug-info of the trap omits
4830 // any reference to the allocation, as we've already
4831 // observed, then there can be no objection to the trap.
4832 bool found_trap = false;
4833 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4834 Node* obs = not_ctl->fast_out(j);
4835 if (obs->in(0) == not_ctl && obs->is_Call() &&
4836 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4837 found_trap = true; break;
4838 }
4839 }
4840 if (found_trap) {
4841 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4842 continue;
4843 }
4844 }
4845 return NULL;
4846 }
4847
4848 // If we get this far, we have an allocation which immediately
4849 // precedes the arraycopy, and we can take over zeroing the new object.
4850 // The arraycopy will finish the initialization, and provide
4851 // a new control state to which we will anchor the destination pointer.
4852
4853 return alloc;
4854 }
4855
4856 //-------------inline_encodeISOArray-----------------------------------
4857 // encode char[] to byte[] in ISO_8859_1 or ASCII
4858 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
4859 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4860 // no receiver since it is static method
4861 Node *src = argument(0);
4862 Node *src_offset = argument(1);
4863 Node *dst = argument(2);
4864 Node *dst_offset = argument(3);
4865 Node *length = argument(4);
4866
4867 src = must_be_not_null(src, true);
4868 dst = must_be_not_null(dst, true);
4869
4870 const Type* src_type = src->Value(&_gvn);
4871 const Type* dst_type = dst->Value(&_gvn);
4872 const TypeAryPtr* top_src = src_type->isa_aryptr();
4873 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4874 if (top_src == NULL || top_src->klass() == NULL ||
4875 top_dest == NULL || top_dest->klass() == NULL) {
4876 // failed array check
4877 return false;
4878 }
4879
4880 // Figure out the size and type of the elements we will be copying.
4881 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4882 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4883 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4884 return false;
4885 }
4886
4887 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4888 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4889 // 'src_start' points to src array + scaled offset
4890 // 'dst_start' points to dst array + scaled offset
4891
4892 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4893 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
4894 enc = _gvn.transform(enc);
4895 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4896 set_memory(res_mem, mtype);
4897 set_result(enc);
4898 clear_upper_avx();
4899
4900 return true;
4901 }
4902
4903 //-------------inline_multiplyToLen-----------------------------------
4904 bool LibraryCallKit::inline_multiplyToLen() {
4905 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4906
4907 address stubAddr = StubRoutines::multiplyToLen();
4908 if (stubAddr == NULL) {
4909 return false; // Intrinsic's stub is not implemented on this platform
4910 }
4911 const char* stubName = "multiplyToLen";
4912
4913 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4914
4915 // no receiver because it is a static method
4916 Node* x = argument(0);
4917 Node* xlen = argument(1);
4918 Node* y = argument(2);
4919 Node* ylen = argument(3);
4920 Node* z = argument(4);
4921
4922 x = must_be_not_null(x, true);
4923 y = must_be_not_null(y, true);
4924
4925 const Type* x_type = x->Value(&_gvn);
4926 const Type* y_type = y->Value(&_gvn);
4927 const TypeAryPtr* top_x = x_type->isa_aryptr();
4928 const TypeAryPtr* top_y = y_type->isa_aryptr();
4929 if (top_x == NULL || top_x->klass() == NULL ||
4930 top_y == NULL || top_y->klass() == NULL) {
4931 // failed array check
4932 return false;
4933 }
4934
4935 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4936 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4937 if (x_elem != T_INT || y_elem != T_INT) {
4938 return false;
4939 }
4940
4941 // Set the original stack and the reexecute bit for the interpreter to reexecute
4942 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
4943 // on the return from z array allocation in runtime.
4944 { PreserveReexecuteState preexecs(this);
4945 jvms()->set_should_reexecute(true);
4946
4947 Node* x_start = array_element_address(x, intcon(0), x_elem);
4948 Node* y_start = array_element_address(y, intcon(0), y_elem);
4949 // 'x_start' points to x array + scaled xlen
4950 // 'y_start' points to y array + scaled ylen
4951
4952 // Allocate the result array
4953 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
4954 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
4955 Node* klass_node = makecon(TypeKlassPtr::make(klass));
4956
4957 IdealKit ideal(this);
4958
4959 #define __ ideal.
4960 Node* one = __ ConI(1);
4961 Node* zero = __ ConI(0);
4962 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
4963 __ set(need_alloc, zero);
4964 __ set(z_alloc, z);
4965 __ if_then(z, BoolTest::eq, null()); {
4966 __ increment (need_alloc, one);
4967 } __ else_(); {
4968 // Update graphKit memory and control from IdealKit.
4969 sync_kit(ideal);
4970 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
4971 cast->init_req(0, control());
4972 _gvn.set_type(cast, cast->bottom_type());
4973 C->record_for_igvn(cast);
4974
4975 Node* zlen_arg = load_array_length(cast);
4976 // Update IdealKit memory and control from graphKit.
4977 __ sync_kit(this);
4978 __ if_then(zlen_arg, BoolTest::lt, zlen); {
4979 __ increment (need_alloc, one);
4980 } __ end_if();
4981 } __ end_if();
4982
4983 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
4984 // Update graphKit memory and control from IdealKit.
4985 sync_kit(ideal);
4986 Node * narr = new_array(klass_node, zlen, 1);
4987 // Update IdealKit memory and control from graphKit.
4988 __ sync_kit(this);
4989 __ set(z_alloc, narr);
4990 } __ end_if();
4991
4992 sync_kit(ideal);
4993 z = __ value(z_alloc);
4994 // Can't use TypeAryPtr::INTS which uses Bottom offset.
4995 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
4996 // Final sync IdealKit and GraphKit.
4997 final_sync(ideal);
4998 #undef __
4999
5000 Node* z_start = array_element_address(z, intcon(0), T_INT);
5001
5002 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5003 OptoRuntime::multiplyToLen_Type(),
5004 stubAddr, stubName, TypePtr::BOTTOM,
5005 x_start, xlen, y_start, ylen, z_start, zlen);
5006 } // original reexecute is set back here
5007
5008 C->set_has_split_ifs(true); // Has chance for split-if optimization
5009 set_result(z);
5010 return true;
5011 }
5012
5013 //-------------inline_squareToLen------------------------------------
5014 bool LibraryCallKit::inline_squareToLen() {
5015 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5016
5017 address stubAddr = StubRoutines::squareToLen();
5018 if (stubAddr == NULL) {
5019 return false; // Intrinsic's stub is not implemented on this platform
5020 }
5021 const char* stubName = "squareToLen";
5022
5023 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5024
5025 Node* x = argument(0);
5026 Node* len = argument(1);
5027 Node* z = argument(2);
5028 Node* zlen = argument(3);
5029
5030 x = must_be_not_null(x, true);
5031 z = must_be_not_null(z, true);
5032
5033 const Type* x_type = x->Value(&_gvn);
5034 const Type* z_type = z->Value(&_gvn);
5035 const TypeAryPtr* top_x = x_type->isa_aryptr();
5036 const TypeAryPtr* top_z = z_type->isa_aryptr();
5037 if (top_x == NULL || top_x->klass() == NULL ||
5038 top_z == NULL || top_z->klass() == NULL) {
5039 // failed array check
5040 return false;
5041 }
5042
5043 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5044 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5045 if (x_elem != T_INT || z_elem != T_INT) {
5046 return false;
5047 }
5048
5049
5050 Node* x_start = array_element_address(x, intcon(0), x_elem);
5051 Node* z_start = array_element_address(z, intcon(0), z_elem);
5052
5053 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5054 OptoRuntime::squareToLen_Type(),
5055 stubAddr, stubName, TypePtr::BOTTOM,
5056 x_start, len, z_start, zlen);
5057
5058 set_result(z);
5059 return true;
5060 }
5061
5062 //-------------inline_mulAdd------------------------------------------
5063 bool LibraryCallKit::inline_mulAdd() {
5064 assert(UseMulAddIntrinsic, "not implemented on this platform");
5065
5066 address stubAddr = StubRoutines::mulAdd();
5067 if (stubAddr == NULL) {
5068 return false; // Intrinsic's stub is not implemented on this platform
5069 }
5070 const char* stubName = "mulAdd";
5071
5072 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5073
5074 Node* out = argument(0);
5075 Node* in = argument(1);
5076 Node* offset = argument(2);
5077 Node* len = argument(3);
5078 Node* k = argument(4);
5079
5080 out = must_be_not_null(out, true);
5081
5082 const Type* out_type = out->Value(&_gvn);
5083 const Type* in_type = in->Value(&_gvn);
5084 const TypeAryPtr* top_out = out_type->isa_aryptr();
5085 const TypeAryPtr* top_in = in_type->isa_aryptr();
5086 if (top_out == NULL || top_out->klass() == NULL ||
5087 top_in == NULL || top_in->klass() == NULL) {
5088 // failed array check
5089 return false;
5090 }
5091
5092 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5093 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5094 if (out_elem != T_INT || in_elem != T_INT) {
5095 return false;
5096 }
5097
5098 Node* outlen = load_array_length(out);
5099 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5100 Node* out_start = array_element_address(out, intcon(0), out_elem);
5101 Node* in_start = array_element_address(in, intcon(0), in_elem);
5102
5103 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5104 OptoRuntime::mulAdd_Type(),
5105 stubAddr, stubName, TypePtr::BOTTOM,
5106 out_start,in_start, new_offset, len, k);
5107 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5108 set_result(result);
5109 return true;
5110 }
5111
5112 //-------------inline_montgomeryMultiply-----------------------------------
5113 bool LibraryCallKit::inline_montgomeryMultiply() {
5114 address stubAddr = StubRoutines::montgomeryMultiply();
5115 if (stubAddr == NULL) {
5116 return false; // Intrinsic's stub is not implemented on this platform
5117 }
5118
5119 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5120 const char* stubName = "montgomery_multiply";
5121
5122 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5123
5124 Node* a = argument(0);
5125 Node* b = argument(1);
5126 Node* n = argument(2);
5127 Node* len = argument(3);
5128 Node* inv = argument(4);
5129 Node* m = argument(6);
5130
5131 const Type* a_type = a->Value(&_gvn);
5132 const TypeAryPtr* top_a = a_type->isa_aryptr();
5133 const Type* b_type = b->Value(&_gvn);
5134 const TypeAryPtr* top_b = b_type->isa_aryptr();
5135 const Type* n_type = a->Value(&_gvn);
5136 const TypeAryPtr* top_n = n_type->isa_aryptr();
5137 const Type* m_type = a->Value(&_gvn);
5138 const TypeAryPtr* top_m = m_type->isa_aryptr();
5139 if (top_a == NULL || top_a->klass() == NULL ||
5140 top_b == NULL || top_b->klass() == NULL ||
5141 top_n == NULL || top_n->klass() == NULL ||
5142 top_m == NULL || top_m->klass() == NULL) {
5143 // failed array check
5144 return false;
5145 }
5146
5147 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5148 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5149 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5150 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5151 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5152 return false;
5153 }
5154
5155 // Make the call
5156 {
5157 Node* a_start = array_element_address(a, intcon(0), a_elem);
5158 Node* b_start = array_element_address(b, intcon(0), b_elem);
5159 Node* n_start = array_element_address(n, intcon(0), n_elem);
5160 Node* m_start = array_element_address(m, intcon(0), m_elem);
5161
5162 Node* call = make_runtime_call(RC_LEAF,
5163 OptoRuntime::montgomeryMultiply_Type(),
5164 stubAddr, stubName, TypePtr::BOTTOM,
5165 a_start, b_start, n_start, len, inv, top(),
5166 m_start);
5167 set_result(m);
5168 }
5169
5170 return true;
5171 }
5172
5173 bool LibraryCallKit::inline_montgomerySquare() {
5174 address stubAddr = StubRoutines::montgomerySquare();
5175 if (stubAddr == NULL) {
5176 return false; // Intrinsic's stub is not implemented on this platform
5177 }
5178
5179 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5180 const char* stubName = "montgomery_square";
5181
5182 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5183
5184 Node* a = argument(0);
5185 Node* n = argument(1);
5186 Node* len = argument(2);
5187 Node* inv = argument(3);
5188 Node* m = argument(5);
5189
5190 const Type* a_type = a->Value(&_gvn);
5191 const TypeAryPtr* top_a = a_type->isa_aryptr();
5192 const Type* n_type = a->Value(&_gvn);
5193 const TypeAryPtr* top_n = n_type->isa_aryptr();
5194 const Type* m_type = a->Value(&_gvn);
5195 const TypeAryPtr* top_m = m_type->isa_aryptr();
5196 if (top_a == NULL || top_a->klass() == NULL ||
5197 top_n == NULL || top_n->klass() == NULL ||
5198 top_m == NULL || top_m->klass() == NULL) {
5199 // failed array check
5200 return false;
5201 }
5202
5203 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5204 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5205 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5206 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5207 return false;
5208 }
5209
5210 // Make the call
5211 {
5212 Node* a_start = array_element_address(a, intcon(0), a_elem);
5213 Node* n_start = array_element_address(n, intcon(0), n_elem);
5214 Node* m_start = array_element_address(m, intcon(0), m_elem);
5215
5216 Node* call = make_runtime_call(RC_LEAF,
5217 OptoRuntime::montgomerySquare_Type(),
5218 stubAddr, stubName, TypePtr::BOTTOM,
5219 a_start, n_start, len, inv, top(),
5220 m_start);
5221 set_result(m);
5222 }
5223
5224 return true;
5225 }
5226
5227 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5228 address stubAddr = NULL;
5229 const char* stubName = NULL;
5230
5231 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5232 if (stubAddr == NULL) {
5233 return false; // Intrinsic's stub is not implemented on this platform
5234 }
5235
5236 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5237
5238 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5239
5240 Node* newArr = argument(0);
5241 Node* oldArr = argument(1);
5242 Node* newIdx = argument(2);
5243 Node* shiftCount = argument(3);
5244 Node* numIter = argument(4);
5245
5246 const Type* newArr_type = newArr->Value(&_gvn);
5247 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5248 const Type* oldArr_type = oldArr->Value(&_gvn);
5249 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5250 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5251 || top_oldArr->klass() == NULL) {
5252 return false;
5253 }
5254
5255 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5256 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5257 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5258 return false;
5259 }
5260
5261 // Make the call
5262 {
5263 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5264 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5265
5266 Node* call = make_runtime_call(RC_LEAF,
5267 OptoRuntime::bigIntegerShift_Type(),
5268 stubAddr,
5269 stubName,
5270 TypePtr::BOTTOM,
5271 newArr_start,
5272 oldArr_start,
5273 newIdx,
5274 shiftCount,
5275 numIter);
5276 }
5277
5278 return true;
5279 }
5280
5281 //-------------inline_vectorizedMismatch------------------------------
5282 bool LibraryCallKit::inline_vectorizedMismatch() {
5283 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5284
5285 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5286 Node* obja = argument(0); // Object
5287 Node* aoffset = argument(1); // long
5288 Node* objb = argument(3); // Object
5289 Node* boffset = argument(4); // long
5290 Node* length = argument(6); // int
5291 Node* scale = argument(7); // int
5292
5293 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5294 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5295 if (obja_t == NULL || obja_t->klass() == NULL ||
5296 objb_t == NULL || objb_t->klass() == NULL ||
5297 scale == top()) {
5298 return false; // failed input validation
5299 }
5300
5301 Node* obja_adr = make_unsafe_address(obja, aoffset);
5302 Node* objb_adr = make_unsafe_address(objb, boffset);
5303
5304 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5305 //
5306 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5307 // if (length <= inline_limit) {
5308 // inline_path:
5309 // vmask = VectorMaskGen length
5310 // vload1 = LoadVectorMasked obja, vmask
5311 // vload2 = LoadVectorMasked objb, vmask
5312 // result1 = VectorCmpMasked vload1, vload2, vmask
5313 // } else {
5314 // call_stub_path:
5315 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5316 // }
5317 // exit_block:
5318 // return Phi(result1, result2);
5319 //
5320 enum { inline_path = 1, // input is small enough to process it all at once
5321 stub_path = 2, // input is too large; call into the VM
5322 PATH_LIMIT = 3
5323 };
5324
5325 Node* exit_block = new RegionNode(PATH_LIMIT);
5326 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5327 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5328
5329 Node* call_stub_path = control();
5330
5331 BasicType elem_bt = T_ILLEGAL;
5332
5333 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5334 if (scale_t->is_con()) {
5335 switch (scale_t->get_con()) {
5336 case 0: elem_bt = T_BYTE; break;
5337 case 1: elem_bt = T_SHORT; break;
5338 case 2: elem_bt = T_INT; break;
5339 case 3: elem_bt = T_LONG; break;
5340
5341 default: elem_bt = T_ILLEGAL; break; // not supported
5342 }
5343 }
5344
5345 int inline_limit = 0;
5346 bool do_partial_inline = false;
5347
5348 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5349 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5350 do_partial_inline = inline_limit >= 16;
5351 }
5352
5353 if (do_partial_inline) {
5354 assert(elem_bt != T_ILLEGAL, "sanity");
5355
5356 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5357 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5358 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5359
5360 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5361 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5362 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5363
5364 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5365
5366 if (!stopped()) {
5367 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5368
5369 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5370 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5371 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5372 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5373
5374 Node* vmask = _gvn.transform(new VectorMaskGenNode(ConvI2X(casted_length), TypeVect::VECTMASK, elem_bt));
5375 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5376 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5377 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5378
5379 exit_block->init_req(inline_path, control());
5380 memory_phi->init_req(inline_path, map()->memory());
5381 result_phi->init_req(inline_path, result);
5382
5383 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5384 clear_upper_avx();
5385 }
5386 }
5387 }
5388
5389 if (call_stub_path != NULL) {
5390 set_control(call_stub_path);
5391
5392 Node* call = make_runtime_call(RC_LEAF,
5393 OptoRuntime::vectorizedMismatch_Type(),
5394 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5395 obja_adr, objb_adr, length, scale);
5396
5397 exit_block->init_req(stub_path, control());
5398 memory_phi->init_req(stub_path, map()->memory());
5399 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5400 }
5401
5402 exit_block = _gvn.transform(exit_block);
5403 memory_phi = _gvn.transform(memory_phi);
5404 result_phi = _gvn.transform(result_phi);
5405
5406 set_control(exit_block);
5407 set_all_memory(memory_phi);
5408 set_result(result_phi);
5409
5410 return true;
5411 }
5412
5413 /**
5414 * Calculate CRC32 for byte.
5415 * int java.util.zip.CRC32.update(int crc, int b)
5416 */
5417 bool LibraryCallKit::inline_updateCRC32() {
5418 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5419 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5420 // no receiver since it is static method
5421 Node* crc = argument(0); // type: int
5422 Node* b = argument(1); // type: int
5423
5424 /*
5425 * int c = ~ crc;
5426 * b = timesXtoThe32[(b ^ c) & 0xFF];
5427 * b = b ^ (c >>> 8);
5428 * crc = ~b;
5429 */
5430
5431 Node* M1 = intcon(-1);
5432 crc = _gvn.transform(new XorINode(crc, M1));
5433 Node* result = _gvn.transform(new XorINode(crc, b));
5434 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5435
5436 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5437 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5438 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5439 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5440
5441 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5442 result = _gvn.transform(new XorINode(crc, result));
5443 result = _gvn.transform(new XorINode(result, M1));
5444 set_result(result);
5445 return true;
5446 }
5447
5448 /**
5449 * Calculate CRC32 for byte[] array.
5450 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5451 */
5452 bool LibraryCallKit::inline_updateBytesCRC32() {
5453 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5454 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5455 // no receiver since it is static method
5456 Node* crc = argument(0); // type: int
5457 Node* src = argument(1); // type: oop
5458 Node* offset = argument(2); // type: int
5459 Node* length = argument(3); // type: int
5460
5461 const Type* src_type = src->Value(&_gvn);
5462 const TypeAryPtr* top_src = src_type->isa_aryptr();
5463 if (top_src == NULL || top_src->klass() == NULL) {
5464 // failed array check
5465 return false;
5466 }
5467
5468 // Figure out the size and type of the elements we will be copying.
5469 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5470 if (src_elem != T_BYTE) {
5471 return false;
5472 }
5473
5474 // 'src_start' points to src array + scaled offset
5475 src = must_be_not_null(src, true);
5476 Node* src_start = array_element_address(src, offset, src_elem);
5477
5478 // We assume that range check is done by caller.
5479 // TODO: generate range check (offset+length < src.length) in debug VM.
5480
5481 // Call the stub.
5482 address stubAddr = StubRoutines::updateBytesCRC32();
5483 const char *stubName = "updateBytesCRC32";
5484
5485 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5486 stubAddr, stubName, TypePtr::BOTTOM,
5487 crc, src_start, length);
5488 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5489 set_result(result);
5490 return true;
5491 }
5492
5493 /**
5494 * Calculate CRC32 for ByteBuffer.
5495 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5496 */
5497 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5498 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5499 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5500 // no receiver since it is static method
5501 Node* crc = argument(0); // type: int
5502 Node* src = argument(1); // type: long
5503 Node* offset = argument(3); // type: int
5504 Node* length = argument(4); // type: int
5505
5506 src = ConvL2X(src); // adjust Java long to machine word
5507 Node* base = _gvn.transform(new CastX2PNode(src));
5508 offset = ConvI2X(offset);
5509
5510 // 'src_start' points to src array + scaled offset
5511 Node* src_start = basic_plus_adr(top(), base, offset);
5512
5513 // Call the stub.
5514 address stubAddr = StubRoutines::updateBytesCRC32();
5515 const char *stubName = "updateBytesCRC32";
5516
5517 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5518 stubAddr, stubName, TypePtr::BOTTOM,
5519 crc, src_start, length);
5520 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5521 set_result(result);
5522 return true;
5523 }
5524
5525 //------------------------------get_table_from_crc32c_class-----------------------
5526 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5527 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
5528 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5529
5530 return table;
5531 }
5532
5533 //------------------------------inline_updateBytesCRC32C-----------------------
5534 //
5535 // Calculate CRC32C for byte[] array.
5536 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5537 //
5538 bool LibraryCallKit::inline_updateBytesCRC32C() {
5539 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5540 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5541 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5542 // no receiver since it is a static method
5543 Node* crc = argument(0); // type: int
5544 Node* src = argument(1); // type: oop
5545 Node* offset = argument(2); // type: int
5546 Node* end = argument(3); // type: int
5547
5548 Node* length = _gvn.transform(new SubINode(end, offset));
5549
5550 const Type* src_type = src->Value(&_gvn);
5551 const TypeAryPtr* top_src = src_type->isa_aryptr();
5552 if (top_src == NULL || top_src->klass() == NULL) {
5553 // failed array check
5554 return false;
5555 }
5556
5557 // Figure out the size and type of the elements we will be copying.
5558 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5559 if (src_elem != T_BYTE) {
5560 return false;
5561 }
5562
5563 // 'src_start' points to src array + scaled offset
5564 src = must_be_not_null(src, true);
5565 Node* src_start = array_element_address(src, offset, src_elem);
5566
5567 // static final int[] byteTable in class CRC32C
5568 Node* table = get_table_from_crc32c_class(callee()->holder());
5569 table = must_be_not_null(table, true);
5570 Node* table_start = array_element_address(table, intcon(0), T_INT);
5571
5572 // We assume that range check is done by caller.
5573 // TODO: generate range check (offset+length < src.length) in debug VM.
5574
5575 // Call the stub.
5576 address stubAddr = StubRoutines::updateBytesCRC32C();
5577 const char *stubName = "updateBytesCRC32C";
5578
5579 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5580 stubAddr, stubName, TypePtr::BOTTOM,
5581 crc, src_start, length, table_start);
5582 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5583 set_result(result);
5584 return true;
5585 }
5586
5587 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5588 //
5589 // Calculate CRC32C for DirectByteBuffer.
5590 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5591 //
5592 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5593 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5594 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5595 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5596 // no receiver since it is a static method
5597 Node* crc = argument(0); // type: int
5598 Node* src = argument(1); // type: long
5599 Node* offset = argument(3); // type: int
5600 Node* end = argument(4); // type: int
5601
5602 Node* length = _gvn.transform(new SubINode(end, offset));
5603
5604 src = ConvL2X(src); // adjust Java long to machine word
5605 Node* base = _gvn.transform(new CastX2PNode(src));
5606 offset = ConvI2X(offset);
5607
5608 // 'src_start' points to src array + scaled offset
5609 Node* src_start = basic_plus_adr(top(), base, offset);
5610
5611 // static final int[] byteTable in class CRC32C
5612 Node* table = get_table_from_crc32c_class(callee()->holder());
5613 table = must_be_not_null(table, true);
5614 Node* table_start = array_element_address(table, intcon(0), T_INT);
5615
5616 // Call the stub.
5617 address stubAddr = StubRoutines::updateBytesCRC32C();
5618 const char *stubName = "updateBytesCRC32C";
5619
5620 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5621 stubAddr, stubName, TypePtr::BOTTOM,
5622 crc, src_start, length, table_start);
5623 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5624 set_result(result);
5625 return true;
5626 }
5627
5628 //------------------------------inline_updateBytesAdler32----------------------
5629 //
5630 // Calculate Adler32 checksum for byte[] array.
5631 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5632 //
5633 bool LibraryCallKit::inline_updateBytesAdler32() {
5634 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5635 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5636 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5637 // no receiver since it is static method
5638 Node* crc = argument(0); // type: int
5639 Node* src = argument(1); // type: oop
5640 Node* offset = argument(2); // type: int
5641 Node* length = argument(3); // type: int
5642
5643 const Type* src_type = src->Value(&_gvn);
5644 const TypeAryPtr* top_src = src_type->isa_aryptr();
5645 if (top_src == NULL || top_src->klass() == NULL) {
5646 // failed array check
5647 return false;
5648 }
5649
5650 // Figure out the size and type of the elements we will be copying.
5651 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5652 if (src_elem != T_BYTE) {
5653 return false;
5654 }
5655
5656 // 'src_start' points to src array + scaled offset
5657 Node* src_start = array_element_address(src, offset, src_elem);
5658
5659 // We assume that range check is done by caller.
5660 // TODO: generate range check (offset+length < src.length) in debug VM.
5661
5662 // Call the stub.
5663 address stubAddr = StubRoutines::updateBytesAdler32();
5664 const char *stubName = "updateBytesAdler32";
5665
5666 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5667 stubAddr, stubName, TypePtr::BOTTOM,
5668 crc, src_start, length);
5669 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5670 set_result(result);
5671 return true;
5672 }
5673
5674 //------------------------------inline_updateByteBufferAdler32---------------
5675 //
5676 // Calculate Adler32 checksum for DirectByteBuffer.
5677 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5678 //
5679 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5680 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5681 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5682 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5683 // no receiver since it is static method
5684 Node* crc = argument(0); // type: int
5685 Node* src = argument(1); // type: long
5686 Node* offset = argument(3); // type: int
5687 Node* length = argument(4); // type: int
5688
5689 src = ConvL2X(src); // adjust Java long to machine word
5690 Node* base = _gvn.transform(new CastX2PNode(src));
5691 offset = ConvI2X(offset);
5692
5693 // 'src_start' points to src array + scaled offset
5694 Node* src_start = basic_plus_adr(top(), base, offset);
5695
5696 // Call the stub.
5697 address stubAddr = StubRoutines::updateBytesAdler32();
5698 const char *stubName = "updateBytesAdler32";
5699
5700 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5701 stubAddr, stubName, TypePtr::BOTTOM,
5702 crc, src_start, length);
5703
5704 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5705 set_result(result);
5706 return true;
5707 }
5708
5709 //----------------------------inline_reference_get----------------------------
5710 // public T java.lang.ref.Reference.get();
5711 bool LibraryCallKit::inline_reference_get() {
5712 const int referent_offset = java_lang_ref_Reference::referent_offset();
5713
5714 // Get the argument:
5715 Node* reference_obj = null_check_receiver();
5716 if (stopped()) return true;
5717
5718 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5719 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
5720 decorators, /*is_static*/ false, NULL);
5721 if (result == NULL) return false;
5722
5723 // Add memory barrier to prevent commoning reads from this field
5724 // across safepoint since GC can change its value.
5725 insert_mem_bar(Op_MemBarCPUOrder);
5726
5727 set_result(result);
5728 return true;
5729 }
5730
5731 //----------------------------inline_reference_refersTo0----------------------------
5732 // bool java.lang.ref.Reference.refersTo0();
5733 // bool java.lang.ref.PhantomReference.refersTo0();
5734 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
5735 // Get arguments:
5736 Node* reference_obj = null_check_receiver();
5737 Node* other_obj = argument(1);
5738 if (stopped()) return true;
5739
5740 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
5741 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
5742 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
5743 decorators, /*is_static*/ false, NULL);
5744 if (referent == NULL) return false;
5745
5746 // Add memory barrier to prevent commoning reads from this field
5747 // across safepoint since GC can change its value.
5748 insert_mem_bar(Op_MemBarCPUOrder);
5749
5750 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
5751 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
5752 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
5753
5754 RegionNode* region = new RegionNode(3);
5755 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
5756
5757 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
5758 region->init_req(1, if_true);
5759 phi->init_req(1, intcon(1));
5760
5761 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
5762 region->init_req(2, if_false);
5763 phi->init_req(2, intcon(0));
5764
5765 set_control(_gvn.transform(region));
5766 record_for_igvn(region);
5767 set_result(_gvn.transform(phi));
5768 return true;
5769 }
5770
5771
5772 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
5773 DecoratorSet decorators = IN_HEAP, bool is_static = false,
5774 ciInstanceKlass* fromKls = NULL) {
5775 if (fromKls == NULL) {
5776 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5777 assert(tinst != NULL, "obj is null");
5778 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5779 fromKls = tinst->klass()->as_instance_klass();
5780 } else {
5781 assert(is_static, "only for static field access");
5782 }
5783 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5784 ciSymbol::make(fieldTypeString),
5785 is_static);
5786
5787 assert (field != NULL, "undefined field");
5788 if (field == NULL) return (Node *) NULL;
5789
5790 if (is_static) {
5791 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5792 fromObj = makecon(tip);
5793 }
5794
5795 // Next code copied from Parse::do_get_xxx():
5796
5797 // Compute address and memory type.
5798 int offset = field->offset_in_bytes();
5799 bool is_vol = field->is_volatile();
5800 ciType* field_klass = field->type();
5801 assert(field_klass->is_loaded(), "should be loaded");
5802 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5803 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5804 BasicType bt = field->layout_type();
5805
5806 // Build the resultant type of the load
5807 const Type *type;
5808 if (bt == T_OBJECT) {
5809 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5810 } else {
5811 type = Type::get_const_basic_type(bt);
5812 }
5813
5814 if (is_vol) {
5815 decorators |= MO_SEQ_CST;
5816 }
5817
5818 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5819 }
5820
5821 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5822 bool is_exact = true, bool is_static = false,
5823 ciInstanceKlass * fromKls = NULL) {
5824 if (fromKls == NULL) {
5825 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5826 assert(tinst != NULL, "obj is null");
5827 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5828 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5829 fromKls = tinst->klass()->as_instance_klass();
5830 }
5831 else {
5832 assert(is_static, "only for static field access");
5833 }
5834 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5835 ciSymbol::make(fieldTypeString),
5836 is_static);
5837
5838 assert(field != NULL, "undefined field");
5839 assert(!field->is_volatile(), "not defined for volatile fields");
5840
5841 if (is_static) {
5842 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5843 fromObj = makecon(tip);
5844 }
5845
5846 // Next code copied from Parse::do_get_xxx():
5847
5848 // Compute address and memory type.
5849 int offset = field->offset_in_bytes();
5850 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5851
5852 return adr;
5853 }
5854
5855 //------------------------------inline_aescrypt_Block-----------------------
5856 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5857 address stubAddr = NULL;
5858 const char *stubName;
5859 assert(UseAES, "need AES instruction support");
5860
5861 switch(id) {
5862 case vmIntrinsics::_aescrypt_encryptBlock:
5863 stubAddr = StubRoutines::aescrypt_encryptBlock();
5864 stubName = "aescrypt_encryptBlock";
5865 break;
5866 case vmIntrinsics::_aescrypt_decryptBlock:
5867 stubAddr = StubRoutines::aescrypt_decryptBlock();
5868 stubName = "aescrypt_decryptBlock";
5869 break;
5870 default:
5871 break;
5872 }
5873 if (stubAddr == NULL) return false;
5874
5875 Node* aescrypt_object = argument(0);
5876 Node* src = argument(1);
5877 Node* src_offset = argument(2);
5878 Node* dest = argument(3);
5879 Node* dest_offset = argument(4);
5880
5881 src = must_be_not_null(src, true);
5882 dest = must_be_not_null(dest, true);
5883
5884 // (1) src and dest are arrays.
5885 const Type* src_type = src->Value(&_gvn);
5886 const Type* dest_type = dest->Value(&_gvn);
5887 const TypeAryPtr* top_src = src_type->isa_aryptr();
5888 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5889 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5890
5891 // for the quick and dirty code we will skip all the checks.
5892 // we are just trying to get the call to be generated.
5893 Node* src_start = src;
5894 Node* dest_start = dest;
5895 if (src_offset != NULL || dest_offset != NULL) {
5896 assert(src_offset != NULL && dest_offset != NULL, "");
5897 src_start = array_element_address(src, src_offset, T_BYTE);
5898 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5899 }
5900
5901 // now need to get the start of its expanded key array
5902 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5903 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5904 if (k_start == NULL) return false;
5905
5906 // Call the stub.
5907 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5908 stubAddr, stubName, TypePtr::BOTTOM,
5909 src_start, dest_start, k_start);
5910
5911 return true;
5912 }
5913
5914 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5915 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5916 address stubAddr = NULL;
5917 const char *stubName = NULL;
5918
5919 assert(UseAES, "need AES instruction support");
5920
5921 switch(id) {
5922 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5923 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5924 stubName = "cipherBlockChaining_encryptAESCrypt";
5925 break;
5926 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5927 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5928 stubName = "cipherBlockChaining_decryptAESCrypt";
5929 break;
5930 default:
5931 break;
5932 }
5933 if (stubAddr == NULL) return false;
5934
5935 Node* cipherBlockChaining_object = argument(0);
5936 Node* src = argument(1);
5937 Node* src_offset = argument(2);
5938 Node* len = argument(3);
5939 Node* dest = argument(4);
5940 Node* dest_offset = argument(5);
5941
5942 src = must_be_not_null(src, false);
5943 dest = must_be_not_null(dest, false);
5944
5945 // (1) src and dest are arrays.
5946 const Type* src_type = src->Value(&_gvn);
5947 const Type* dest_type = dest->Value(&_gvn);
5948 const TypeAryPtr* top_src = src_type->isa_aryptr();
5949 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5950 assert (top_src != NULL && top_src->klass() != NULL
5951 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5952
5953 // checks are the responsibility of the caller
5954 Node* src_start = src;
5955 Node* dest_start = dest;
5956 if (src_offset != NULL || dest_offset != NULL) {
5957 assert(src_offset != NULL && dest_offset != NULL, "");
5958 src_start = array_element_address(src, src_offset, T_BYTE);
5959 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5960 }
5961
5962 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5963 // (because of the predicated logic executed earlier).
5964 // so we cast it here safely.
5965 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5966
5967 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
5968 if (embeddedCipherObj == NULL) return false;
5969
5970 // cast it to what we know it will be at runtime
5971 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5972 assert(tinst != NULL, "CBC obj is null");
5973 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5974 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5975 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5976
5977 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5978 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5979 const TypeOopPtr* xtype = aklass->as_instance_type();
5980 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5981 aescrypt_object = _gvn.transform(aescrypt_object);
5982
5983 // we need to get the start of the aescrypt_object's expanded key array
5984 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5985 if (k_start == NULL) return false;
5986
5987 // similarly, get the start address of the r vector
5988 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
5989 if (objRvec == NULL) return false;
5990 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5991
5992 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5993 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5994 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5995 stubAddr, stubName, TypePtr::BOTTOM,
5996 src_start, dest_start, k_start, r_start, len);
5997
5998 // return cipher length (int)
5999 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6000 set_result(retvalue);
6001 return true;
6002 }
6003
6004 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6005 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6006 address stubAddr = NULL;
6007 const char *stubName = NULL;
6008
6009 assert(UseAES, "need AES instruction support");
6010
6011 switch (id) {
6012 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6013 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6014 stubName = "electronicCodeBook_encryptAESCrypt";
6015 break;
6016 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6017 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6018 stubName = "electronicCodeBook_decryptAESCrypt";
6019 break;
6020 default:
6021 break;
6022 }
6023
6024 if (stubAddr == NULL) return false;
6025
6026 Node* electronicCodeBook_object = argument(0);
6027 Node* src = argument(1);
6028 Node* src_offset = argument(2);
6029 Node* len = argument(3);
6030 Node* dest = argument(4);
6031 Node* dest_offset = argument(5);
6032
6033 // (1) src and dest are arrays.
6034 const Type* src_type = src->Value(&_gvn);
6035 const Type* dest_type = dest->Value(&_gvn);
6036 const TypeAryPtr* top_src = src_type->isa_aryptr();
6037 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6038 assert(top_src != NULL && top_src->klass() != NULL
6039 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6040
6041 // checks are the responsibility of the caller
6042 Node* src_start = src;
6043 Node* dest_start = dest;
6044 if (src_offset != NULL || dest_offset != NULL) {
6045 assert(src_offset != NULL && dest_offset != NULL, "");
6046 src_start = array_element_address(src, src_offset, T_BYTE);
6047 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6048 }
6049
6050 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6051 // (because of the predicated logic executed earlier).
6052 // so we cast it here safely.
6053 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6054
6055 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6056 if (embeddedCipherObj == NULL) return false;
6057
6058 // cast it to what we know it will be at runtime
6059 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6060 assert(tinst != NULL, "ECB obj is null");
6061 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6062 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6063 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6064
6065 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6066 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6067 const TypeOopPtr* xtype = aklass->as_instance_type();
6068 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6069 aescrypt_object = _gvn.transform(aescrypt_object);
6070
6071 // we need to get the start of the aescrypt_object's expanded key array
6072 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6073 if (k_start == NULL) return false;
6074
6075 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6076 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6077 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6078 stubAddr, stubName, TypePtr::BOTTOM,
6079 src_start, dest_start, k_start, len);
6080
6081 // return cipher length (int)
6082 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6083 set_result(retvalue);
6084 return true;
6085 }
6086
6087 //------------------------------inline_counterMode_AESCrypt-----------------------
6088 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6089 assert(UseAES, "need AES instruction support");
6090 if (!UseAESCTRIntrinsics) return false;
6091
6092 address stubAddr = NULL;
6093 const char *stubName = NULL;
6094 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6095 stubAddr = StubRoutines::counterMode_AESCrypt();
6096 stubName = "counterMode_AESCrypt";
6097 }
6098 if (stubAddr == NULL) return false;
6099
6100 Node* counterMode_object = argument(0);
6101 Node* src = argument(1);
6102 Node* src_offset = argument(2);
6103 Node* len = argument(3);
6104 Node* dest = argument(4);
6105 Node* dest_offset = argument(5);
6106
6107 // (1) src and dest are arrays.
6108 const Type* src_type = src->Value(&_gvn);
6109 const Type* dest_type = dest->Value(&_gvn);
6110 const TypeAryPtr* top_src = src_type->isa_aryptr();
6111 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6112 assert(top_src != NULL && top_src->klass() != NULL &&
6113 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6114
6115 // checks are the responsibility of the caller
6116 Node* src_start = src;
6117 Node* dest_start = dest;
6118 if (src_offset != NULL || dest_offset != NULL) {
6119 assert(src_offset != NULL && dest_offset != NULL, "");
6120 src_start = array_element_address(src, src_offset, T_BYTE);
6121 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6122 }
6123
6124 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6125 // (because of the predicated logic executed earlier).
6126 // so we cast it here safely.
6127 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6128 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6129 if (embeddedCipherObj == NULL) return false;
6130 // cast it to what we know it will be at runtime
6131 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6132 assert(tinst != NULL, "CTR obj is null");
6133 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6134 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6135 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6136 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6137 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6138 const TypeOopPtr* xtype = aklass->as_instance_type();
6139 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6140 aescrypt_object = _gvn.transform(aescrypt_object);
6141 // we need to get the start of the aescrypt_object's expanded key array
6142 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6143 if (k_start == NULL) return false;
6144 // similarly, get the start address of the r vector
6145 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6146 if (obj_counter == NULL) return false;
6147 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6148
6149 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6150 if (saved_encCounter == NULL) return false;
6151 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6152 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6153
6154 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6155 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6156 OptoRuntime::counterMode_aescrypt_Type(),
6157 stubAddr, stubName, TypePtr::BOTTOM,
6158 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6159
6160 // return cipher length (int)
6161 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6162 set_result(retvalue);
6163 return true;
6164 }
6165
6166 //------------------------------get_key_start_from_aescrypt_object-----------------------
6167 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6168 #if defined(PPC64) || defined(S390)
6169 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6170 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6171 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6172 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6173 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6174 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6175 if (objSessionK == NULL) {
6176 return (Node *) NULL;
6177 }
6178 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6179 #else
6180 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6181 #endif // PPC64
6182 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6183 if (objAESCryptKey == NULL) return (Node *) NULL;
6184
6185 // now have the array, need to get the start address of the K array
6186 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6187 return k_start;
6188 }
6189
6190 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6191 // Return node representing slow path of predicate check.
6192 // the pseudo code we want to emulate with this predicate is:
6193 // for encryption:
6194 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6195 // for decryption:
6196 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6197 // note cipher==plain is more conservative than the original java code but that's OK
6198 //
6199 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6200 // The receiver was checked for NULL already.
6201 Node* objCBC = argument(0);
6202
6203 Node* src = argument(1);
6204 Node* dest = argument(4);
6205
6206 // Load embeddedCipher field of CipherBlockChaining object.
6207 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6208
6209 // get AESCrypt klass for instanceOf check
6210 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6211 // will have same classloader as CipherBlockChaining object
6212 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6213 assert(tinst != NULL, "CBCobj is null");
6214 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6215
6216 // we want to do an instanceof comparison against the AESCrypt class
6217 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6218 if (!klass_AESCrypt->is_loaded()) {
6219 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6220 Node* ctrl = control();
6221 set_control(top()); // no regular fast path
6222 return ctrl;
6223 }
6224
6225 src = must_be_not_null(src, true);
6226 dest = must_be_not_null(dest, true);
6227
6228 // Resolve oops to stable for CmpP below.
6229 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6230
6231 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6232 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6233 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6234
6235 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6236
6237 // for encryption, we are done
6238 if (!decrypting)
6239 return instof_false; // even if it is NULL
6240
6241 // for decryption, we need to add a further check to avoid
6242 // taking the intrinsic path when cipher and plain are the same
6243 // see the original java code for why.
6244 RegionNode* region = new RegionNode(3);
6245 region->init_req(1, instof_false);
6246
6247 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6248 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6249 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6250 region->init_req(2, src_dest_conjoint);
6251
6252 record_for_igvn(region);
6253 return _gvn.transform(region);
6254 }
6255
6256 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6257 // Return node representing slow path of predicate check.
6258 // the pseudo code we want to emulate with this predicate is:
6259 // for encryption:
6260 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6261 // for decryption:
6262 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6263 // note cipher==plain is more conservative than the original java code but that's OK
6264 //
6265 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6266 // The receiver was checked for NULL already.
6267 Node* objECB = argument(0);
6268
6269 // Load embeddedCipher field of ElectronicCodeBook object.
6270 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6271
6272 // get AESCrypt klass for instanceOf check
6273 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6274 // will have same classloader as ElectronicCodeBook object
6275 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6276 assert(tinst != NULL, "ECBobj is null");
6277 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6278
6279 // we want to do an instanceof comparison against the AESCrypt class
6280 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6281 if (!klass_AESCrypt->is_loaded()) {
6282 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6283 Node* ctrl = control();
6284 set_control(top()); // no regular fast path
6285 return ctrl;
6286 }
6287 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6288
6289 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6290 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6291 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6292
6293 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6294
6295 // for encryption, we are done
6296 if (!decrypting)
6297 return instof_false; // even if it is NULL
6298
6299 // for decryption, we need to add a further check to avoid
6300 // taking the intrinsic path when cipher and plain are the same
6301 // see the original java code for why.
6302 RegionNode* region = new RegionNode(3);
6303 region->init_req(1, instof_false);
6304 Node* src = argument(1);
6305 Node* dest = argument(4);
6306 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6307 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6308 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6309 region->init_req(2, src_dest_conjoint);
6310
6311 record_for_igvn(region);
6312 return _gvn.transform(region);
6313 }
6314
6315 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6316 // Return node representing slow path of predicate check.
6317 // the pseudo code we want to emulate with this predicate is:
6318 // for encryption:
6319 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6320 // for decryption:
6321 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6322 // note cipher==plain is more conservative than the original java code but that's OK
6323 //
6324
6325 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6326 // The receiver was checked for NULL already.
6327 Node* objCTR = argument(0);
6328
6329 // Load embeddedCipher field of CipherBlockChaining object.
6330 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6331
6332 // get AESCrypt klass for instanceOf check
6333 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6334 // will have same classloader as CipherBlockChaining object
6335 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6336 assert(tinst != NULL, "CTRobj is null");
6337 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6338
6339 // we want to do an instanceof comparison against the AESCrypt class
6340 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6341 if (!klass_AESCrypt->is_loaded()) {
6342 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6343 Node* ctrl = control();
6344 set_control(top()); // no regular fast path
6345 return ctrl;
6346 }
6347
6348 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6349 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6350 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6351 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6352 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6353
6354 return instof_false; // even if it is NULL
6355 }
6356
6357 //------------------------------inline_ghash_processBlocks
6358 bool LibraryCallKit::inline_ghash_processBlocks() {
6359 address stubAddr;
6360 const char *stubName;
6361 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6362
6363 stubAddr = StubRoutines::ghash_processBlocks();
6364 stubName = "ghash_processBlocks";
6365
6366 Node* data = argument(0);
6367 Node* offset = argument(1);
6368 Node* len = argument(2);
6369 Node* state = argument(3);
6370 Node* subkeyH = argument(4);
6371
6372 state = must_be_not_null(state, true);
6373 subkeyH = must_be_not_null(subkeyH, true);
6374 data = must_be_not_null(data, true);
6375
6376 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6377 assert(state_start, "state is NULL");
6378 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6379 assert(subkeyH_start, "subkeyH is NULL");
6380 Node* data_start = array_element_address(data, offset, T_BYTE);
6381 assert(data_start, "data is NULL");
6382
6383 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6384 OptoRuntime::ghash_processBlocks_Type(),
6385 stubAddr, stubName, TypePtr::BOTTOM,
6386 state_start, subkeyH_start, data_start, len);
6387 return true;
6388 }
6389
6390 bool LibraryCallKit::inline_base64_encodeBlock() {
6391 address stubAddr;
6392 const char *stubName;
6393 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6394 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6395 stubAddr = StubRoutines::base64_encodeBlock();
6396 stubName = "encodeBlock";
6397
6398 if (!stubAddr) return false;
6399 Node* base64obj = argument(0);
6400 Node* src = argument(1);
6401 Node* offset = argument(2);
6402 Node* len = argument(3);
6403 Node* dest = argument(4);
6404 Node* dp = argument(5);
6405 Node* isURL = argument(6);
6406
6407 src = must_be_not_null(src, true);
6408 dest = must_be_not_null(dest, true);
6409
6410 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6411 assert(src_start, "source array is NULL");
6412 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6413 assert(dest_start, "destination array is NULL");
6414
6415 Node* base64 = make_runtime_call(RC_LEAF,
6416 OptoRuntime::base64_encodeBlock_Type(),
6417 stubAddr, stubName, TypePtr::BOTTOM,
6418 src_start, offset, len, dest_start, dp, isURL);
6419 return true;
6420 }
6421
6422 bool LibraryCallKit::inline_base64_decodeBlock() {
6423 address stubAddr;
6424 const char *stubName;
6425 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6426 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6427 stubAddr = StubRoutines::base64_decodeBlock();
6428 stubName = "decodeBlock";
6429
6430 if (!stubAddr) return false;
6431 Node* base64obj = argument(0);
6432 Node* src = argument(1);
6433 Node* src_offset = argument(2);
6434 Node* len = argument(3);
6435 Node* dest = argument(4);
6436 Node* dest_offset = argument(5);
6437 Node* isURL = argument(6);
6438 Node* isMIME = argument(7);
6439
6440 src = must_be_not_null(src, true);
6441 dest = must_be_not_null(dest, true);
6442
6443 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6444 assert(src_start, "source array is NULL");
6445 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6446 assert(dest_start, "destination array is NULL");
6447
6448 Node* call = make_runtime_call(RC_LEAF,
6449 OptoRuntime::base64_decodeBlock_Type(),
6450 stubAddr, stubName, TypePtr::BOTTOM,
6451 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6452 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6453 set_result(result);
6454 return true;
6455 }
6456
6457 //------------------------------inline_digestBase_implCompress-----------------------
6458 //
6459 // Calculate MD5 for single-block byte[] array.
6460 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6461 //
6462 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6463 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6464 //
6465 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6466 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6467 //
6468 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6469 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6470 //
6471 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6472 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6473 //
6474 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6475 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6476
6477 Node* digestBase_obj = argument(0);
6478 Node* src = argument(1); // type oop
6479 Node* ofs = argument(2); // type int
6480
6481 const Type* src_type = src->Value(&_gvn);
6482 const TypeAryPtr* top_src = src_type->isa_aryptr();
6483 if (top_src == NULL || top_src->klass() == NULL) {
6484 // failed array check
6485 return false;
6486 }
6487 // Figure out the size and type of the elements we will be copying.
6488 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6489 if (src_elem != T_BYTE) {
6490 return false;
6491 }
6492 // 'src_start' points to src array + offset
6493 src = must_be_not_null(src, true);
6494 Node* src_start = array_element_address(src, ofs, src_elem);
6495 Node* state = NULL;
6496 Node* digest_length = NULL;
6497 address stubAddr;
6498 const char *stubName;
6499
6500 switch(id) {
6501 case vmIntrinsics::_md5_implCompress:
6502 assert(UseMD5Intrinsics, "need MD5 instruction support");
6503 state = get_state_from_digest_object(digestBase_obj, T_INT);
6504 stubAddr = StubRoutines::md5_implCompress();
6505 stubName = "md5_implCompress";
6506 break;
6507 case vmIntrinsics::_sha_implCompress:
6508 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6509 state = get_state_from_digest_object(digestBase_obj, T_INT);
6510 stubAddr = StubRoutines::sha1_implCompress();
6511 stubName = "sha1_implCompress";
6512 break;
6513 case vmIntrinsics::_sha2_implCompress:
6514 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6515 state = get_state_from_digest_object(digestBase_obj, T_INT);
6516 stubAddr = StubRoutines::sha256_implCompress();
6517 stubName = "sha256_implCompress";
6518 break;
6519 case vmIntrinsics::_sha5_implCompress:
6520 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6521 state = get_state_from_digest_object(digestBase_obj, T_LONG);
6522 stubAddr = StubRoutines::sha512_implCompress();
6523 stubName = "sha512_implCompress";
6524 break;
6525 case vmIntrinsics::_sha3_implCompress:
6526 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
6527 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
6528 stubAddr = StubRoutines::sha3_implCompress();
6529 stubName = "sha3_implCompress";
6530 digest_length = get_digest_length_from_digest_object(digestBase_obj);
6531 if (digest_length == NULL) return false;
6532 break;
6533 default:
6534 fatal_unexpected_iid(id);
6535 return false;
6536 }
6537 if (state == NULL) return false;
6538
6539 assert(stubAddr != NULL, "Stub is generated");
6540 if (stubAddr == NULL) return false;
6541
6542 // Call the stub.
6543 Node* call;
6544 if (digest_length == NULL) {
6545 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
6546 stubAddr, stubName, TypePtr::BOTTOM,
6547 src_start, state);
6548 } else {
6549 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
6550 stubAddr, stubName, TypePtr::BOTTOM,
6551 src_start, state, digest_length);
6552 }
6553
6554 return true;
6555 }
6556
6557 //------------------------------inline_digestBase_implCompressMB-----------------------
6558 //
6559 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
6560 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6561 //
6562 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6563 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6564 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6565 assert((uint)predicate < 5, "sanity");
6566 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6567
6568 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6569 Node* src = argument(1); // byte[] array
6570 Node* ofs = argument(2); // type int
6571 Node* limit = argument(3); // type int
6572
6573 const Type* src_type = src->Value(&_gvn);
6574 const TypeAryPtr* top_src = src_type->isa_aryptr();
6575 if (top_src == NULL || top_src->klass() == NULL) {
6576 // failed array check
6577 return false;
6578 }
6579 // Figure out the size and type of the elements we will be copying.
6580 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6581 if (src_elem != T_BYTE) {
6582 return false;
6583 }
6584 // 'src_start' points to src array + offset
6585 src = must_be_not_null(src, false);
6586 Node* src_start = array_element_address(src, ofs, src_elem);
6587
6588 const char* klass_digestBase_name = NULL;
6589 const char* stub_name = NULL;
6590 address stub_addr = NULL;
6591 BasicType elem_type = T_INT;
6592
6593 switch (predicate) {
6594 case 0:
6595 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
6596 klass_digestBase_name = "sun/security/provider/MD5";
6597 stub_name = "md5_implCompressMB";
6598 stub_addr = StubRoutines::md5_implCompressMB();
6599 }
6600 break;
6601 case 1:
6602 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
6603 klass_digestBase_name = "sun/security/provider/SHA";
6604 stub_name = "sha1_implCompressMB";
6605 stub_addr = StubRoutines::sha1_implCompressMB();
6606 }
6607 break;
6608 case 2:
6609 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
6610 klass_digestBase_name = "sun/security/provider/SHA2";
6611 stub_name = "sha256_implCompressMB";
6612 stub_addr = StubRoutines::sha256_implCompressMB();
6613 }
6614 break;
6615 case 3:
6616 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
6617 klass_digestBase_name = "sun/security/provider/SHA5";
6618 stub_name = "sha512_implCompressMB";
6619 stub_addr = StubRoutines::sha512_implCompressMB();
6620 elem_type = T_LONG;
6621 }
6622 break;
6623 case 4:
6624 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
6625 klass_digestBase_name = "sun/security/provider/SHA3";
6626 stub_name = "sha3_implCompressMB";
6627 stub_addr = StubRoutines::sha3_implCompressMB();
6628 elem_type = T_BYTE;
6629 }
6630 break;
6631 default:
6632 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
6633 }
6634 if (klass_digestBase_name != NULL) {
6635 assert(stub_addr != NULL, "Stub is generated");
6636 if (stub_addr == NULL) return false;
6637
6638 // get DigestBase klass to lookup for SHA klass
6639 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6640 assert(tinst != NULL, "digestBase_obj is not instance???");
6641 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6642
6643 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
6644 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
6645 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
6646 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
6647 }
6648 return false;
6649 }
6650
6651 //------------------------------inline_digestBase_implCompressMB-----------------------
6652 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
6653 BasicType elem_type, address stubAddr, const char *stubName,
6654 Node* src_start, Node* ofs, Node* limit) {
6655 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
6656 const TypeOopPtr* xtype = aklass->as_instance_type();
6657 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6658 digest_obj = _gvn.transform(digest_obj);
6659
6660 Node* state = get_state_from_digest_object(digest_obj, elem_type);
6661 if (state == NULL) return false;
6662
6663 Node* digest_length = NULL;
6664 if (strcmp("sha3_implCompressMB", stubName) == 0) {
6665 digest_length = get_digest_length_from_digest_object(digest_obj);
6666 if (digest_length == NULL) return false;
6667 }
6668
6669 // Call the stub.
6670 Node* call;
6671 if (digest_length == NULL) {
6672 call = make_runtime_call(RC_LEAF|RC_NO_FP,
6673 OptoRuntime::digestBase_implCompressMB_Type(false),
6674 stubAddr, stubName, TypePtr::BOTTOM,
6675 src_start, state, ofs, limit);
6676 } else {
6677 call = make_runtime_call(RC_LEAF|RC_NO_FP,
6678 OptoRuntime::digestBase_implCompressMB_Type(true),
6679 stubAddr, stubName, TypePtr::BOTTOM,
6680 src_start, state, digest_length, ofs, limit);
6681 }
6682
6683 // return ofs (int)
6684 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6685 set_result(result);
6686
6687 return true;
6688 }
6689
6690 //------------------------------get_state_from_digest_object-----------------------
6691 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
6692 const char* state_type;
6693 switch (elem_type) {
6694 case T_BYTE: state_type = "[B"; break;
6695 case T_INT: state_type = "[I"; break;
6696 case T_LONG: state_type = "[J"; break;
6697 default: ShouldNotReachHere();
6698 }
6699 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
6700 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
6701 if (digest_state == NULL) return (Node *) NULL;
6702
6703 // now have the array, need to get the start address of the state array
6704 Node* state = array_element_address(digest_state, intcon(0), elem_type);
6705 return state;
6706 }
6707
6708 //------------------------------get_digest_length_from_sha3_object----------------------------------
6709 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
6710 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
6711 assert (digest_length != NULL, "sanity");
6712 return digest_length;
6713 }
6714
6715 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6716 // Return node representing slow path of predicate check.
6717 // the pseudo code we want to emulate with this predicate is:
6718 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
6719 //
6720 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6721 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6722 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6723 assert((uint)predicate < 5, "sanity");
6724
6725 // The receiver was checked for NULL already.
6726 Node* digestBaseObj = argument(0);
6727
6728 // get DigestBase klass for instanceOf check
6729 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6730 assert(tinst != NULL, "digestBaseObj is null");
6731 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6732
6733 const char* klass_name = NULL;
6734 switch (predicate) {
6735 case 0:
6736 if (UseMD5Intrinsics) {
6737 // we want to do an instanceof comparison against the MD5 class
6738 klass_name = "sun/security/provider/MD5";
6739 }
6740 break;
6741 case 1:
6742 if (UseSHA1Intrinsics) {
6743 // we want to do an instanceof comparison against the SHA class
6744 klass_name = "sun/security/provider/SHA";
6745 }
6746 break;
6747 case 2:
6748 if (UseSHA256Intrinsics) {
6749 // we want to do an instanceof comparison against the SHA2 class
6750 klass_name = "sun/security/provider/SHA2";
6751 }
6752 break;
6753 case 3:
6754 if (UseSHA512Intrinsics) {
6755 // we want to do an instanceof comparison against the SHA5 class
6756 klass_name = "sun/security/provider/SHA5";
6757 }
6758 break;
6759 case 4:
6760 if (UseSHA3Intrinsics) {
6761 // we want to do an instanceof comparison against the SHA3 class
6762 klass_name = "sun/security/provider/SHA3";
6763 }
6764 break;
6765 default:
6766 fatal("unknown SHA intrinsic predicate: %d", predicate);
6767 }
6768
6769 ciKlass* klass = NULL;
6770 if (klass_name != NULL) {
6771 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
6772 }
6773 if ((klass == NULL) || !klass->is_loaded()) {
6774 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6775 Node* ctrl = control();
6776 set_control(top()); // no intrinsic path
6777 return ctrl;
6778 }
6779 ciInstanceKlass* instklass = klass->as_instance_klass();
6780
6781 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
6782 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6783 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6784 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6785
6786 return instof_false; // even if it is NULL
6787 }
6788
6789 //-------------inline_fma-----------------------------------
6790 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6791 Node *a = NULL;
6792 Node *b = NULL;
6793 Node *c = NULL;
6794 Node* result = NULL;
6795 switch (id) {
6796 case vmIntrinsics::_fmaD:
6797 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6798 // no receiver since it is static method
6799 a = round_double_node(argument(0));
6800 b = round_double_node(argument(2));
6801 c = round_double_node(argument(4));
6802 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6803 break;
6804 case vmIntrinsics::_fmaF:
6805 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6806 a = argument(0);
6807 b = argument(1);
6808 c = argument(2);
6809 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6810 break;
6811 default:
6812 fatal_unexpected_iid(id); break;
6813 }
6814 set_result(result);
6815 return true;
6816 }
6817
6818 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6819 // argument(0) is receiver
6820 Node* codePoint = argument(1);
6821 Node* n = NULL;
6822
6823 switch (id) {
6824 case vmIntrinsics::_isDigit :
6825 n = new DigitNode(control(), codePoint);
6826 break;
6827 case vmIntrinsics::_isLowerCase :
6828 n = new LowerCaseNode(control(), codePoint);
6829 break;
6830 case vmIntrinsics::_isUpperCase :
6831 n = new UpperCaseNode(control(), codePoint);
6832 break;
6833 case vmIntrinsics::_isWhitespace :
6834 n = new WhitespaceNode(control(), codePoint);
6835 break;
6836 default:
6837 fatal_unexpected_iid(id);
6838 }
6839
6840 set_result(_gvn.transform(n));
6841 return true;
6842 }
6843
6844 //------------------------------inline_fp_min_max------------------------------
6845 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6846 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6847
6848 // The intrinsic should be used only when the API branches aren't predictable,
6849 // the last one performing the most important comparison. The following heuristic
6850 // uses the branch statistics to eventually bail out if necessary.
6851
6852 ciMethodData *md = callee()->method_data();
6853
6854 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6855 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6856
6857 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6858 // Bail out if the call-site didn't contribute enough to the statistics.
6859 return false;
6860 }
6861
6862 uint taken = 0, not_taken = 0;
6863
6864 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6865 if (p->is_BranchData()) {
6866 taken = ((ciBranchData*)p)->taken();
6867 not_taken = ((ciBranchData*)p)->not_taken();
6868 }
6869 }
6870
6871 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6872 balance = balance < 0 ? -balance : balance;
6873 if ( balance > 0.2 ) {
6874 // Bail out if the most important branch is predictable enough.
6875 return false;
6876 }
6877 }
6878 */
6879
6880 Node *a = NULL;
6881 Node *b = NULL;
6882 Node *n = NULL;
6883 switch (id) {
6884 case vmIntrinsics::_maxF:
6885 case vmIntrinsics::_minF:
6886 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6887 a = argument(0);
6888 b = argument(1);
6889 break;
6890 case vmIntrinsics::_maxD:
6891 case vmIntrinsics::_minD:
6892 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6893 a = round_double_node(argument(0));
6894 b = round_double_node(argument(2));
6895 break;
6896 default:
6897 fatal_unexpected_iid(id);
6898 break;
6899 }
6900 switch (id) {
6901 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6902 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6903 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6904 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6905 default: fatal_unexpected_iid(id); break;
6906 }
6907 set_result(_gvn.transform(n));
6908 return true;
6909 }
6910
6911 bool LibraryCallKit::inline_profileBoolean() {
6912 Node* counts = argument(1);
6913 const TypeAryPtr* ary = NULL;
6914 ciArray* aobj = NULL;
6915 if (counts->is_Con()
6916 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6917 && (aobj = ary->const_oop()->as_array()) != NULL
6918 && (aobj->length() == 2)) {
6919 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6920 jint false_cnt = aobj->element_value(0).as_int();
6921 jint true_cnt = aobj->element_value(1).as_int();
6922
6923 if (C->log() != NULL) {
6924 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6925 false_cnt, true_cnt);
6926 }
6927
6928 if (false_cnt + true_cnt == 0) {
6929 // According to profile, never executed.
6930 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6931 Deoptimization::Action_reinterpret);
6932 return true;
6933 }
6934
6935 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6936 // is a number of each value occurrences.
6937 Node* result = argument(0);
6938 if (false_cnt == 0 || true_cnt == 0) {
6939 // According to profile, one value has been never seen.
6940 int expected_val = (false_cnt == 0) ? 1 : 0;
6941
6942 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6943 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6944
6945 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6946 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6947 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6948
6949 { // Slow path: uncommon trap for never seen value and then reexecute
6950 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6951 // the value has been seen at least once.
6952 PreserveJVMState pjvms(this);
6953 PreserveReexecuteState preexecs(this);
6954 jvms()->set_should_reexecute(true);
6955
6956 set_control(slow_path);
6957 set_i_o(i_o());
6958
6959 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6960 Deoptimization::Action_reinterpret);
6961 }
6962 // The guard for never seen value enables sharpening of the result and
6963 // returning a constant. It allows to eliminate branches on the same value
6964 // later on.
6965 set_control(fast_path);
6966 result = intcon(expected_val);
6967 }
6968 // Stop profiling.
6969 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6970 // By replacing method body with profile data (represented as ProfileBooleanNode
6971 // on IR level) we effectively disable profiling.
6972 // It enables full speed execution once optimized code is generated.
6973 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6974 C->record_for_igvn(profile);
6975 set_result(profile);
6976 return true;
6977 } else {
6978 // Continue profiling.
6979 // Profile data isn't available at the moment. So, execute method's bytecode version.
6980 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6981 // is compiled and counters aren't available since corresponding MethodHandle
6982 // isn't a compile-time constant.
6983 return false;
6984 }
6985 }
6986
6987 bool LibraryCallKit::inline_isCompileConstant() {
6988 Node* n = argument(0);
6989 set_result(n->is_Con() ? intcon(1) : intcon(0));
6990 return true;
6991 }
6992
6993 //------------------------------- inline_getObjectSize --------------------------------------
6994 //
6995 // Calculate the runtime size of the object/array.
6996 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
6997 //
6998 bool LibraryCallKit::inline_getObjectSize() {
6999 Node* obj = argument(3);
7000 Node* klass_node = load_object_klass(obj);
7001
7002 jint layout_con = Klass::_lh_neutral_value;
7003 Node* layout_val = get_layout_helper(klass_node, layout_con);
7004 int layout_is_con = (layout_val == NULL);
7005
7006 if (layout_is_con) {
7007 // Layout helper is constant, can figure out things at compile time.
7008
7009 if (Klass::layout_helper_is_instance(layout_con)) {
7010 // Instance case: layout_con contains the size itself.
7011 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7012 set_result(size);
7013 } else {
7014 // Array case: size is round(header + element_size*arraylength).
7015 // Since arraylength is different for every array instance, we have to
7016 // compute the whole thing at runtime.
7017
7018 Node* arr_length = load_array_length(obj);
7019
7020 int round_mask = MinObjAlignmentInBytes - 1;
7021 int hsize = Klass::layout_helper_header_size(layout_con);
7022 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7023
7024 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7025 round_mask = 0; // strength-reduce it if it goes away completely
7026 }
7027 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7028 Node* header_size = intcon(hsize + round_mask);
7029
7030 Node* lengthx = ConvI2X(arr_length);
7031 Node* headerx = ConvI2X(header_size);
7032
7033 Node* abody = lengthx;
7034 if (eshift != 0) {
7035 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7036 }
7037 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7038 if (round_mask != 0) {
7039 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7040 }
7041 size = ConvX2L(size);
7042 set_result(size);
7043 }
7044 } else {
7045 // Layout helper is not constant, need to test for array-ness at runtime.
7046
7047 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7048 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7049 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7050 record_for_igvn(result_reg);
7051
7052 Node* array_ctl = generate_array_guard(klass_node, NULL);
7053 if (array_ctl != NULL) {
7054 // Array case: size is round(header + element_size*arraylength).
7055 // Since arraylength is different for every array instance, we have to
7056 // compute the whole thing at runtime.
7057
7058 PreserveJVMState pjvms(this);
7059 set_control(array_ctl);
7060 Node* arr_length = load_array_length(obj);
7061
7062 int round_mask = MinObjAlignmentInBytes - 1;
7063 Node* mask = intcon(round_mask);
7064
7065 Node* hss = intcon(Klass::_lh_header_size_shift);
7066 Node* hsm = intcon(Klass::_lh_header_size_mask);
7067 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7068 header_size = _gvn.transform(new AndINode(header_size, hsm));
7069 header_size = _gvn.transform(new AddINode(header_size, mask));
7070
7071 // There is no need to mask or shift this value.
7072 // The semantics of LShiftINode include an implicit mask to 0x1F.
7073 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7074 Node* elem_shift = layout_val;
7075
7076 Node* lengthx = ConvI2X(arr_length);
7077 Node* headerx = ConvI2X(header_size);
7078
7079 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7080 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7081 if (round_mask != 0) {
7082 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7083 }
7084 size = ConvX2L(size);
7085
7086 result_reg->init_req(_array_path, control());
7087 result_val->init_req(_array_path, size);
7088 }
7089
7090 if (!stopped()) {
7091 // Instance case: the layout helper gives us instance size almost directly,
7092 // but we need to mask out the _lh_instance_slow_path_bit.
7093 Node* size = ConvI2X(layout_val);
7094 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7095 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7096 size = _gvn.transform(new AndXNode(size, mask));
7097 size = ConvX2L(size);
7098
7099 result_reg->init_req(_instance_path, control());
7100 result_val->init_req(_instance_path, size);
7101 }
7102
7103 set_result(result_reg, result_val);
7104 }
7105
7106 return true;
7107 }
7108
7109 //------------------------------- inline_blackhole --------------------------------------
7110 //
7111 // Make sure all arguments to this node are alive.
7112 // This matches methods that were requested to be blackholed through compile commands.
7113 //
7114 bool LibraryCallKit::inline_blackhole() {
7115 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7116 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7117 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7118
7119 // Bind call arguments as blackhole arguments to keep them alive
7120 Node* bh = insert_mem_bar(Op_Blackhole);
7121 uint nargs = callee()->arg_size();
7122 for (uint i = 0; i < nargs; i++) {
7123 bh->add_req(argument(i));
7124 }
7125
7126 return true;
7127 }