1 /*
2 * Copyright (c) 1999, 2021, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/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::_galoisCounterMode_AESCrypt:
551 return inline_galoisCounterMode_AESCrypt();
552
553 case vmIntrinsics::_md5_implCompress:
554 case vmIntrinsics::_sha_implCompress:
555 case vmIntrinsics::_sha2_implCompress:
556 case vmIntrinsics::_sha5_implCompress:
557 case vmIntrinsics::_sha3_implCompress:
558 return inline_digestBase_implCompress(intrinsic_id());
559
560 case vmIntrinsics::_digestBase_implCompressMB:
561 return inline_digestBase_implCompressMB(predicate);
562
563 case vmIntrinsics::_multiplyToLen:
564 return inline_multiplyToLen();
565
566 case vmIntrinsics::_squareToLen:
567 return inline_squareToLen();
568
569 case vmIntrinsics::_mulAdd:
570 return inline_mulAdd();
571
572 case vmIntrinsics::_montgomeryMultiply:
573 return inline_montgomeryMultiply();
574 case vmIntrinsics::_montgomerySquare:
575 return inline_montgomerySquare();
576
577 case vmIntrinsics::_bigIntegerRightShiftWorker:
578 return inline_bigIntegerShift(true);
579 case vmIntrinsics::_bigIntegerLeftShiftWorker:
580 return inline_bigIntegerShift(false);
581
582 case vmIntrinsics::_vectorizedMismatch:
583 return inline_vectorizedMismatch();
584
585 case vmIntrinsics::_ghash_processBlocks:
586 return inline_ghash_processBlocks();
587 case vmIntrinsics::_base64_encodeBlock:
588 return inline_base64_encodeBlock();
589 case vmIntrinsics::_base64_decodeBlock:
590 return inline_base64_decodeBlock();
591
592 case vmIntrinsics::_encodeISOArray:
593 case vmIntrinsics::_encodeByteISOArray:
594 return inline_encodeISOArray(false);
595 case vmIntrinsics::_encodeAsciiArray:
596 return inline_encodeISOArray(true);
597
598 case vmIntrinsics::_updateCRC32:
599 return inline_updateCRC32();
600 case vmIntrinsics::_updateBytesCRC32:
601 return inline_updateBytesCRC32();
602 case vmIntrinsics::_updateByteBufferCRC32:
603 return inline_updateByteBufferCRC32();
604
605 case vmIntrinsics::_updateBytesCRC32C:
606 return inline_updateBytesCRC32C();
607 case vmIntrinsics::_updateDirectByteBufferCRC32C:
608 return inline_updateDirectByteBufferCRC32C();
609
610 case vmIntrinsics::_updateBytesAdler32:
611 return inline_updateBytesAdler32();
612 case vmIntrinsics::_updateByteBufferAdler32:
613 return inline_updateByteBufferAdler32();
614
615 case vmIntrinsics::_profileBoolean:
616 return inline_profileBoolean();
617 case vmIntrinsics::_isCompileConstant:
618 return inline_isCompileConstant();
619
620 case vmIntrinsics::_hasNegatives:
621 return inline_hasNegatives();
622
623 case vmIntrinsics::_fmaD:
624 case vmIntrinsics::_fmaF:
625 return inline_fma(intrinsic_id());
626
627 case vmIntrinsics::_isDigit:
628 case vmIntrinsics::_isLowerCase:
629 case vmIntrinsics::_isUpperCase:
630 case vmIntrinsics::_isWhitespace:
631 return inline_character_compare(intrinsic_id());
632
633 case vmIntrinsics::_maxF:
634 case vmIntrinsics::_minF:
635 case vmIntrinsics::_maxD:
636 case vmIntrinsics::_minD:
637 return inline_fp_min_max(intrinsic_id());
638
639 case vmIntrinsics::_VectorUnaryOp:
640 return inline_vector_nary_operation(1);
641 case vmIntrinsics::_VectorBinaryOp:
642 return inline_vector_nary_operation(2);
643 case vmIntrinsics::_VectorTernaryOp:
644 return inline_vector_nary_operation(3);
645 case vmIntrinsics::_VectorBroadcastCoerced:
646 return inline_vector_broadcast_coerced();
647 case vmIntrinsics::_VectorShuffleIota:
648 return inline_vector_shuffle_iota();
649 case vmIntrinsics::_VectorMaskOp:
650 return inline_vector_mask_operation();
651 case vmIntrinsics::_VectorShuffleToVector:
652 return inline_vector_shuffle_to_vector();
653 case vmIntrinsics::_VectorLoadOp:
654 return inline_vector_mem_operation(/*is_store=*/false);
655 case vmIntrinsics::_VectorStoreOp:
656 return inline_vector_mem_operation(/*is_store=*/true);
657 case vmIntrinsics::_VectorGatherOp:
658 return inline_vector_gather_scatter(/*is_scatter*/ false);
659 case vmIntrinsics::_VectorScatterOp:
660 return inline_vector_gather_scatter(/*is_scatter*/ true);
661 case vmIntrinsics::_VectorReductionCoerced:
662 return inline_vector_reduction();
663 case vmIntrinsics::_VectorTest:
664 return inline_vector_test();
665 case vmIntrinsics::_VectorBlend:
666 return inline_vector_blend();
667 case vmIntrinsics::_VectorRearrange:
668 return inline_vector_rearrange();
669 case vmIntrinsics::_VectorCompare:
670 return inline_vector_compare();
671 case vmIntrinsics::_VectorBroadcastInt:
672 return inline_vector_broadcast_int();
673 case vmIntrinsics::_VectorConvert:
674 return inline_vector_convert();
675 case vmIntrinsics::_VectorInsert:
676 return inline_vector_insert();
677 case vmIntrinsics::_VectorExtract:
678 return inline_vector_extract();
679
680 case vmIntrinsics::_getObjectSize:
681 return inline_getObjectSize();
682
683 case vmIntrinsics::_blackhole:
684 return inline_blackhole();
685
686 default:
687 // If you get here, it may be that someone has added a new intrinsic
688 // to the list in vmIntrinsics.hpp without implementing it here.
689 #ifndef PRODUCT
690 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
691 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
692 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
693 }
694 #endif
695 return false;
696 }
697 }
698
699 Node* LibraryCallKit::try_to_predicate(int predicate) {
700 if (!jvms()->has_method()) {
701 // Root JVMState has a null method.
702 assert(map()->memory()->Opcode() == Op_Parm, "");
703 // Insert the memory aliasing node
704 set_all_memory(reset_memory());
705 }
706 assert(merged_memory(), "");
707
708 switch (intrinsic_id()) {
709 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
710 return inline_cipherBlockChaining_AESCrypt_predicate(false);
711 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
712 return inline_cipherBlockChaining_AESCrypt_predicate(true);
713 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
714 return inline_electronicCodeBook_AESCrypt_predicate(false);
715 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
716 return inline_electronicCodeBook_AESCrypt_predicate(true);
717 case vmIntrinsics::_counterMode_AESCrypt:
718 return inline_counterMode_AESCrypt_predicate();
719 case vmIntrinsics::_digestBase_implCompressMB:
720 return inline_digestBase_implCompressMB_predicate(predicate);
721 case vmIntrinsics::_galoisCounterMode_AESCrypt:
722 return inline_galoisCounterMode_AESCrypt_predicate();
723
724 default:
725 // If you get here, it may be that someone has added a new intrinsic
726 // to the list in vmIntrinsics.hpp without implementing it here.
727 #ifndef PRODUCT
728 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
729 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
730 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
731 }
732 #endif
733 Node* slow_ctl = control();
734 set_control(top()); // No fast path instrinsic
735 return slow_ctl;
736 }
737 }
738
739 //------------------------------set_result-------------------------------
740 // Helper function for finishing intrinsics.
741 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
742 record_for_igvn(region);
743 set_control(_gvn.transform(region));
744 set_result( _gvn.transform(value));
745 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
746 }
747
748 //------------------------------generate_guard---------------------------
749 // Helper function for generating guarded fast-slow graph structures.
750 // The given 'test', if true, guards a slow path. If the test fails
751 // then a fast path can be taken. (We generally hope it fails.)
752 // In all cases, GraphKit::control() is updated to the fast path.
753 // The returned value represents the control for the slow path.
754 // The return value is never 'top'; it is either a valid control
755 // or NULL if it is obvious that the slow path can never be taken.
756 // Also, if region and the slow control are not NULL, the slow edge
757 // is appended to the region.
758 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
759 if (stopped()) {
760 // Already short circuited.
761 return NULL;
762 }
763
764 // Build an if node and its projections.
765 // If test is true we take the slow path, which we assume is uncommon.
766 if (_gvn.type(test) == TypeInt::ZERO) {
767 // The slow branch is never taken. No need to build this guard.
768 return NULL;
769 }
770
771 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
772
773 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
774 if (if_slow == top()) {
775 // The slow branch is never taken. No need to build this guard.
776 return NULL;
777 }
778
779 if (region != NULL)
780 region->add_req(if_slow);
781
782 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
783 set_control(if_fast);
784
785 return if_slow;
786 }
787
788 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
789 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
790 }
791 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
792 return generate_guard(test, region, PROB_FAIR);
793 }
794
795 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
796 Node* *pos_index) {
797 if (stopped())
798 return NULL; // already stopped
799 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
800 return NULL; // index is already adequately typed
801 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
802 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
803 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
804 if (is_neg != NULL && pos_index != NULL) {
805 // Emulate effect of Parse::adjust_map_after_if.
806 Node* ccast = new CastIINode(index, TypeInt::POS);
807 ccast->set_req(0, control());
808 (*pos_index) = _gvn.transform(ccast);
809 }
810 return is_neg;
811 }
812
813 // Make sure that 'position' is a valid limit index, in [0..length].
814 // There are two equivalent plans for checking this:
815 // A. (offset + copyLength) unsigned<= arrayLength
816 // B. offset <= (arrayLength - copyLength)
817 // We require that all of the values above, except for the sum and
818 // difference, are already known to be non-negative.
819 // Plan A is robust in the face of overflow, if offset and copyLength
820 // are both hugely positive.
821 //
822 // Plan B is less direct and intuitive, but it does not overflow at
823 // all, since the difference of two non-negatives is always
824 // representable. Whenever Java methods must perform the equivalent
825 // check they generally use Plan B instead of Plan A.
826 // For the moment we use Plan A.
827 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
828 Node* subseq_length,
829 Node* array_length,
830 RegionNode* region) {
831 if (stopped())
832 return NULL; // already stopped
833 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
834 if (zero_offset && subseq_length->eqv_uncast(array_length))
835 return NULL; // common case of whole-array copy
836 Node* last = subseq_length;
837 if (!zero_offset) // last += offset
838 last = _gvn.transform(new AddINode(last, offset));
839 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
840 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
841 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
842 return is_over;
843 }
844
845 // Emit range checks for the given String.value byte array
846 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
847 if (stopped()) {
848 return; // already stopped
849 }
850 RegionNode* bailout = new RegionNode(1);
851 record_for_igvn(bailout);
852 if (char_count) {
853 // Convert char count to byte count
854 count = _gvn.transform(new LShiftINode(count, intcon(1)));
855 }
856
857 // Offset and count must not be negative
858 generate_negative_guard(offset, bailout);
859 generate_negative_guard(count, bailout);
860 // Offset + count must not exceed length of array
861 generate_limit_guard(offset, count, load_array_length(array), bailout);
862
863 if (bailout->req() > 1) {
864 PreserveJVMState pjvms(this);
865 set_control(_gvn.transform(bailout));
866 uncommon_trap(Deoptimization::Reason_intrinsic,
867 Deoptimization::Action_maybe_recompile);
868 }
869 }
870
871 //--------------------------generate_current_thread--------------------
872 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
873 ciKlass* thread_klass = env()->Thread_klass();
874 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
875 Node* thread = _gvn.transform(new ThreadLocalNode());
876 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
877 tls_output = thread;
878 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
879 thread_obj_handle = _gvn.transform(thread_obj_handle);
880 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
881 }
882
883
884 //------------------------------make_string_method_node------------------------
885 // Helper method for String intrinsic functions. This version is called with
886 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
887 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
888 // containing the lengths of str1 and str2.
889 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
890 Node* result = NULL;
891 switch (opcode) {
892 case Op_StrIndexOf:
893 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
894 str1_start, cnt1, str2_start, cnt2, ae);
895 break;
896 case Op_StrComp:
897 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
898 str1_start, cnt1, str2_start, cnt2, ae);
899 break;
900 case Op_StrEquals:
901 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
902 // Use the constant length if there is one because optimized match rule may exist.
903 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
904 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
905 break;
906 default:
907 ShouldNotReachHere();
908 return NULL;
909 }
910
911 // All these intrinsics have checks.
912 C->set_has_split_ifs(true); // Has chance for split-if optimization
913 clear_upper_avx();
914
915 return _gvn.transform(result);
916 }
917
918 //------------------------------inline_string_compareTo------------------------
919 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
920 Node* arg1 = argument(0);
921 Node* arg2 = argument(1);
922
923 arg1 = must_be_not_null(arg1, true);
924 arg2 = must_be_not_null(arg2, true);
925
926 // Get start addr and length of first argument
927 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
928 Node* arg1_cnt = load_array_length(arg1);
929
930 // Get start addr and length of second argument
931 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
932 Node* arg2_cnt = load_array_length(arg2);
933
934 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
935 set_result(result);
936 return true;
937 }
938
939 //------------------------------inline_string_equals------------------------
940 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
941 Node* arg1 = argument(0);
942 Node* arg2 = argument(1);
943
944 // paths (plus control) merge
945 RegionNode* region = new RegionNode(3);
946 Node* phi = new PhiNode(region, TypeInt::BOOL);
947
948 if (!stopped()) {
949
950 arg1 = must_be_not_null(arg1, true);
951 arg2 = must_be_not_null(arg2, true);
952
953 // Get start addr and length of first argument
954 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
955 Node* arg1_cnt = load_array_length(arg1);
956
957 // Get start addr and length of second argument
958 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
959 Node* arg2_cnt = load_array_length(arg2);
960
961 // Check for arg1_cnt != arg2_cnt
962 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
963 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
964 Node* if_ne = generate_slow_guard(bol, NULL);
965 if (if_ne != NULL) {
966 phi->init_req(2, intcon(0));
967 region->init_req(2, if_ne);
968 }
969
970 // Check for count == 0 is done by assembler code for StrEquals.
971
972 if (!stopped()) {
973 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
974 phi->init_req(1, equals);
975 region->init_req(1, control());
976 }
977 }
978
979 // post merge
980 set_control(_gvn.transform(region));
981 record_for_igvn(region);
982
983 set_result(_gvn.transform(phi));
984 return true;
985 }
986
987 //------------------------------inline_array_equals----------------------------
988 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
989 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
990 Node* arg1 = argument(0);
991 Node* arg2 = argument(1);
992
993 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
994 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
995 clear_upper_avx();
996
997 return true;
998 }
999
1000 //------------------------------inline_hasNegatives------------------------------
1001 bool LibraryCallKit::inline_hasNegatives() {
1002 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1003 return false;
1004 }
1005
1006 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1007 // no receiver since it is static method
1008 Node* ba = argument(0);
1009 Node* offset = argument(1);
1010 Node* len = argument(2);
1011
1012 ba = must_be_not_null(ba, true);
1013
1014 // Range checks
1015 generate_string_range_check(ba, offset, len, false);
1016 if (stopped()) {
1017 return true;
1018 }
1019 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1020 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1021 set_result(_gvn.transform(result));
1022 return true;
1023 }
1024
1025 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1026 Node* index = argument(0);
1027 Node* length = bt == T_INT ? argument(1) : argument(2);
1028 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1029 return false;
1030 }
1031
1032 // check that length is positive
1033 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1034 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1035
1036 {
1037 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1038 uncommon_trap(Deoptimization::Reason_intrinsic,
1039 Deoptimization::Action_make_not_entrant);
1040 }
1041
1042 if (stopped()) {
1043 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1044 return true;
1045 }
1046
1047 // length is now known postive, add a cast node to make this explicit
1048 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1049 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1050 casted_length = _gvn.transform(casted_length);
1051 replace_in_map(length, casted_length);
1052 length = casted_length;
1053
1054 // Use an unsigned comparison for the range check itself
1055 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1056 BoolTest::mask btest = BoolTest::lt;
1057 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1058 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1059 _gvn.set_type(rc, rc->Value(&_gvn));
1060 if (!rc_bool->is_Con()) {
1061 record_for_igvn(rc);
1062 }
1063 set_control(_gvn.transform(new IfTrueNode(rc)));
1064 {
1065 PreserveJVMState pjvms(this);
1066 set_control(_gvn.transform(new IfFalseNode(rc)));
1067 uncommon_trap(Deoptimization::Reason_range_check,
1068 Deoptimization::Action_make_not_entrant);
1069 }
1070
1071 if (stopped()) {
1072 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1073 return true;
1074 }
1075
1076 // index is now known to be >= 0 and < length, cast it
1077 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1078 result = _gvn.transform(result);
1079 set_result(result);
1080 replace_in_map(index, result);
1081 clear_upper_avx();
1082 return true;
1083 }
1084
1085 //------------------------------inline_string_indexOf------------------------
1086 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1087 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1088 return false;
1089 }
1090 Node* src = argument(0);
1091 Node* tgt = argument(1);
1092
1093 // Make the merge point
1094 RegionNode* result_rgn = new RegionNode(4);
1095 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1096
1097 src = must_be_not_null(src, true);
1098 tgt = must_be_not_null(tgt, true);
1099
1100 // Get start addr and length of source string
1101 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1102 Node* src_count = load_array_length(src);
1103
1104 // Get start addr and length of substring
1105 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1106 Node* tgt_count = load_array_length(tgt);
1107
1108 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1109 // Divide src size by 2 if String is UTF16 encoded
1110 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1111 }
1112 if (ae == StrIntrinsicNode::UU) {
1113 // Divide substring size by 2 if String is UTF16 encoded
1114 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1115 }
1116
1117 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1118 if (result != NULL) {
1119 result_phi->init_req(3, result);
1120 result_rgn->init_req(3, control());
1121 }
1122 set_control(_gvn.transform(result_rgn));
1123 record_for_igvn(result_rgn);
1124 set_result(_gvn.transform(result_phi));
1125
1126 return true;
1127 }
1128
1129 //-----------------------------inline_string_indexOf-----------------------
1130 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1131 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1132 return false;
1133 }
1134 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1135 return false;
1136 }
1137 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1138 Node* src = argument(0); // byte[]
1139 Node* src_count = argument(1); // char count
1140 Node* tgt = argument(2); // byte[]
1141 Node* tgt_count = argument(3); // char count
1142 Node* from_index = argument(4); // char index
1143
1144 src = must_be_not_null(src, true);
1145 tgt = must_be_not_null(tgt, true);
1146
1147 // Multiply byte array index by 2 if String is UTF16 encoded
1148 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1149 src_count = _gvn.transform(new SubINode(src_count, from_index));
1150 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1151 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1152
1153 // Range checks
1154 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1155 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1156 if (stopped()) {
1157 return true;
1158 }
1159
1160 RegionNode* region = new RegionNode(5);
1161 Node* phi = new PhiNode(region, TypeInt::INT);
1162
1163 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1164 if (result != NULL) {
1165 // The result is index relative to from_index if substring was found, -1 otherwise.
1166 // Generate code which will fold into cmove.
1167 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1168 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1169
1170 Node* if_lt = generate_slow_guard(bol, NULL);
1171 if (if_lt != NULL) {
1172 // result == -1
1173 phi->init_req(3, result);
1174 region->init_req(3, if_lt);
1175 }
1176 if (!stopped()) {
1177 result = _gvn.transform(new AddINode(result, from_index));
1178 phi->init_req(4, result);
1179 region->init_req(4, control());
1180 }
1181 }
1182
1183 set_control(_gvn.transform(region));
1184 record_for_igvn(region);
1185 set_result(_gvn.transform(phi));
1186 clear_upper_avx();
1187
1188 return true;
1189 }
1190
1191 // Create StrIndexOfNode with fast path checks
1192 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1193 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1194 // Check for substr count > string count
1195 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1196 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1197 Node* if_gt = generate_slow_guard(bol, NULL);
1198 if (if_gt != NULL) {
1199 phi->init_req(1, intcon(-1));
1200 region->init_req(1, if_gt);
1201 }
1202 if (!stopped()) {
1203 // Check for substr count == 0
1204 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1205 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1206 Node* if_zero = generate_slow_guard(bol, NULL);
1207 if (if_zero != NULL) {
1208 phi->init_req(2, intcon(0));
1209 region->init_req(2, if_zero);
1210 }
1211 }
1212 if (!stopped()) {
1213 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1214 }
1215 return NULL;
1216 }
1217
1218 //-----------------------------inline_string_indexOfChar-----------------------
1219 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1220 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1221 return false;
1222 }
1223 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1224 return false;
1225 }
1226 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1227 Node* src = argument(0); // byte[]
1228 Node* tgt = argument(1); // tgt is int ch
1229 Node* from_index = argument(2);
1230 Node* max = argument(3);
1231
1232 src = must_be_not_null(src, true);
1233
1234 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1235 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1236 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1237
1238 // Range checks
1239 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1240 if (stopped()) {
1241 return true;
1242 }
1243
1244 RegionNode* region = new RegionNode(3);
1245 Node* phi = new PhiNode(region, TypeInt::INT);
1246
1247 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1248 C->set_has_split_ifs(true); // Has chance for split-if optimization
1249 _gvn.transform(result);
1250
1251 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1252 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1253
1254 Node* if_lt = generate_slow_guard(bol, NULL);
1255 if (if_lt != NULL) {
1256 // result == -1
1257 phi->init_req(2, result);
1258 region->init_req(2, if_lt);
1259 }
1260 if (!stopped()) {
1261 result = _gvn.transform(new AddINode(result, from_index));
1262 phi->init_req(1, result);
1263 region->init_req(1, control());
1264 }
1265 set_control(_gvn.transform(region));
1266 record_for_igvn(region);
1267 set_result(_gvn.transform(phi));
1268
1269 return true;
1270 }
1271 //---------------------------inline_string_copy---------------------
1272 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1273 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1274 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1275 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1276 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1277 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1278 bool LibraryCallKit::inline_string_copy(bool compress) {
1279 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1280 return false;
1281 }
1282 int nargs = 5; // 2 oops, 3 ints
1283 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1284
1285 Node* src = argument(0);
1286 Node* src_offset = argument(1);
1287 Node* dst = argument(2);
1288 Node* dst_offset = argument(3);
1289 Node* length = argument(4);
1290
1291 // Check for allocation before we add nodes that would confuse
1292 // tightly_coupled_allocation()
1293 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1294
1295 // Figure out the size and type of the elements we will be copying.
1296 const Type* src_type = src->Value(&_gvn);
1297 const Type* dst_type = dst->Value(&_gvn);
1298 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1299 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1300 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1301 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1302 "Unsupported array types for inline_string_copy");
1303
1304 src = must_be_not_null(src, true);
1305 dst = must_be_not_null(dst, true);
1306
1307 // Convert char[] offsets to byte[] offsets
1308 bool convert_src = (compress && src_elem == T_BYTE);
1309 bool convert_dst = (!compress && dst_elem == T_BYTE);
1310 if (convert_src) {
1311 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1312 } else if (convert_dst) {
1313 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1314 }
1315
1316 // Range checks
1317 generate_string_range_check(src, src_offset, length, convert_src);
1318 generate_string_range_check(dst, dst_offset, length, convert_dst);
1319 if (stopped()) {
1320 return true;
1321 }
1322
1323 Node* src_start = array_element_address(src, src_offset, src_elem);
1324 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1325 // 'src_start' points to src array + scaled offset
1326 // 'dst_start' points to dst array + scaled offset
1327 Node* count = NULL;
1328 if (compress) {
1329 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1330 } else {
1331 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1332 }
1333
1334 if (alloc != NULL) {
1335 if (alloc->maybe_set_complete(&_gvn)) {
1336 // "You break it, you buy it."
1337 InitializeNode* init = alloc->initialization();
1338 assert(init->is_complete(), "we just did this");
1339 init->set_complete_with_arraycopy();
1340 assert(dst->is_CheckCastPP(), "sanity");
1341 assert(dst->in(0)->in(0) == init, "dest pinned");
1342 }
1343 // Do not let stores that initialize this object be reordered with
1344 // a subsequent store that would make this object accessible by
1345 // other threads.
1346 // Record what AllocateNode this StoreStore protects so that
1347 // escape analysis can go from the MemBarStoreStoreNode to the
1348 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1349 // based on the escape status of the AllocateNode.
1350 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1351 }
1352 if (compress) {
1353 set_result(_gvn.transform(count));
1354 }
1355 clear_upper_avx();
1356
1357 return true;
1358 }
1359
1360 #ifdef _LP64
1361 #define XTOP ,top() /*additional argument*/
1362 #else //_LP64
1363 #define XTOP /*no additional argument*/
1364 #endif //_LP64
1365
1366 //------------------------inline_string_toBytesU--------------------------
1367 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1368 bool LibraryCallKit::inline_string_toBytesU() {
1369 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1370 return false;
1371 }
1372 // Get the arguments.
1373 Node* value = argument(0);
1374 Node* offset = argument(1);
1375 Node* length = argument(2);
1376
1377 Node* newcopy = NULL;
1378
1379 // Set the original stack and the reexecute bit for the interpreter to reexecute
1380 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1381 { PreserveReexecuteState preexecs(this);
1382 jvms()->set_should_reexecute(true);
1383
1384 // Check if a null path was taken unconditionally.
1385 value = null_check(value);
1386
1387 RegionNode* bailout = new RegionNode(1);
1388 record_for_igvn(bailout);
1389
1390 // Range checks
1391 generate_negative_guard(offset, bailout);
1392 generate_negative_guard(length, bailout);
1393 generate_limit_guard(offset, length, load_array_length(value), bailout);
1394 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1395 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1396
1397 if (bailout->req() > 1) {
1398 PreserveJVMState pjvms(this);
1399 set_control(_gvn.transform(bailout));
1400 uncommon_trap(Deoptimization::Reason_intrinsic,
1401 Deoptimization::Action_maybe_recompile);
1402 }
1403 if (stopped()) {
1404 return true;
1405 }
1406
1407 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1408 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1409 newcopy = new_array(klass_node, size, 0); // no arguments to push
1410 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1411 guarantee(alloc != NULL, "created above");
1412
1413 // Calculate starting addresses.
1414 Node* src_start = array_element_address(value, offset, T_CHAR);
1415 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1416
1417 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1418 const TypeInt* toffset = gvn().type(offset)->is_int();
1419 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1420
1421 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1422 const char* copyfunc_name = "arraycopy";
1423 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1424 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1425 OptoRuntime::fast_arraycopy_Type(),
1426 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1427 src_start, dst_start, ConvI2X(length) XTOP);
1428 // Do not let reads from the cloned object float above the arraycopy.
1429 if (alloc->maybe_set_complete(&_gvn)) {
1430 // "You break it, you buy it."
1431 InitializeNode* init = alloc->initialization();
1432 assert(init->is_complete(), "we just did this");
1433 init->set_complete_with_arraycopy();
1434 assert(newcopy->is_CheckCastPP(), "sanity");
1435 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1436 }
1437 // Do not let stores that initialize this object be reordered with
1438 // a subsequent store that would make this object accessible by
1439 // other threads.
1440 // Record what AllocateNode this StoreStore protects so that
1441 // escape analysis can go from the MemBarStoreStoreNode to the
1442 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1443 // based on the escape status of the AllocateNode.
1444 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1445 } // original reexecute is set back here
1446
1447 C->set_has_split_ifs(true); // Has chance for split-if optimization
1448 if (!stopped()) {
1449 set_result(newcopy);
1450 }
1451 clear_upper_avx();
1452
1453 return true;
1454 }
1455
1456 //------------------------inline_string_getCharsU--------------------------
1457 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1458 bool LibraryCallKit::inline_string_getCharsU() {
1459 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1460 return false;
1461 }
1462
1463 // Get the arguments.
1464 Node* src = argument(0);
1465 Node* src_begin = argument(1);
1466 Node* src_end = argument(2); // exclusive offset (i < src_end)
1467 Node* dst = argument(3);
1468 Node* dst_begin = argument(4);
1469
1470 // Check for allocation before we add nodes that would confuse
1471 // tightly_coupled_allocation()
1472 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1473
1474 // Check if a null path was taken unconditionally.
1475 src = null_check(src);
1476 dst = null_check(dst);
1477 if (stopped()) {
1478 return true;
1479 }
1480
1481 // Get length and convert char[] offset to byte[] offset
1482 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1483 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1484
1485 // Range checks
1486 generate_string_range_check(src, src_begin, length, true);
1487 generate_string_range_check(dst, dst_begin, length, false);
1488 if (stopped()) {
1489 return true;
1490 }
1491
1492 if (!stopped()) {
1493 // Calculate starting addresses.
1494 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1495 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1496
1497 // Check if array addresses are aligned to HeapWordSize
1498 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1499 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1500 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1501 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1502
1503 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1504 const char* copyfunc_name = "arraycopy";
1505 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1506 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1507 OptoRuntime::fast_arraycopy_Type(),
1508 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1509 src_start, dst_start, ConvI2X(length) XTOP);
1510 // Do not let reads from the cloned object float above the arraycopy.
1511 if (alloc != NULL) {
1512 if (alloc->maybe_set_complete(&_gvn)) {
1513 // "You break it, you buy it."
1514 InitializeNode* init = alloc->initialization();
1515 assert(init->is_complete(), "we just did this");
1516 init->set_complete_with_arraycopy();
1517 assert(dst->is_CheckCastPP(), "sanity");
1518 assert(dst->in(0)->in(0) == init, "dest pinned");
1519 }
1520 // Do not let stores that initialize this object be reordered with
1521 // a subsequent store that would make this object accessible by
1522 // other threads.
1523 // Record what AllocateNode this StoreStore protects so that
1524 // escape analysis can go from the MemBarStoreStoreNode to the
1525 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1526 // based on the escape status of the AllocateNode.
1527 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1528 } else {
1529 insert_mem_bar(Op_MemBarCPUOrder);
1530 }
1531 }
1532
1533 C->set_has_split_ifs(true); // Has chance for split-if optimization
1534 return true;
1535 }
1536
1537 //----------------------inline_string_char_access----------------------------
1538 // Store/Load char to/from byte[] array.
1539 // static void StringUTF16.putChar(byte[] val, int index, int c)
1540 // static char StringUTF16.getChar(byte[] val, int index)
1541 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1542 Node* value = argument(0);
1543 Node* index = argument(1);
1544 Node* ch = is_store ? argument(2) : NULL;
1545
1546 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1547 // correctly requires matched array shapes.
1548 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1549 "sanity: byte[] and char[] bases agree");
1550 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1551 "sanity: byte[] and char[] scales agree");
1552
1553 // Bail when getChar over constants is requested: constant folding would
1554 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1555 // Java method would constant fold nicely instead.
1556 if (!is_store && value->is_Con() && index->is_Con()) {
1557 return false;
1558 }
1559
1560 value = must_be_not_null(value, true);
1561
1562 Node* adr = array_element_address(value, index, T_CHAR);
1563 if (adr->is_top()) {
1564 return false;
1565 }
1566 if (is_store) {
1567 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1568 } else {
1569 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1570 set_result(ch);
1571 }
1572 return true;
1573 }
1574
1575 //--------------------------round_double_node--------------------------------
1576 // Round a double node if necessary.
1577 Node* LibraryCallKit::round_double_node(Node* n) {
1578 if (Matcher::strict_fp_requires_explicit_rounding) {
1579 #ifdef IA32
1580 if (UseSSE < 2) {
1581 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1582 }
1583 #else
1584 Unimplemented();
1585 #endif // IA32
1586 }
1587 return n;
1588 }
1589
1590 //------------------------------inline_math-----------------------------------
1591 // public static double Math.abs(double)
1592 // public static double Math.sqrt(double)
1593 // public static double Math.log(double)
1594 // public static double Math.log10(double)
1595 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1596 Node* arg = round_double_node(argument(0));
1597 Node* n = NULL;
1598 switch (id) {
1599 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1600 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1601 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1602 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1603 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1604 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1605 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1606 default: fatal_unexpected_iid(id); break;
1607 }
1608 set_result(_gvn.transform(n));
1609 return true;
1610 }
1611
1612 //------------------------------inline_math-----------------------------------
1613 // public static float Math.abs(float)
1614 // public static int Math.abs(int)
1615 // public static long Math.abs(long)
1616 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1617 Node* arg = argument(0);
1618 Node* n = NULL;
1619 switch (id) {
1620 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1621 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1622 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1623 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1624 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1625 default: fatal_unexpected_iid(id); break;
1626 }
1627 set_result(_gvn.transform(n));
1628 return true;
1629 }
1630
1631 //------------------------------runtime_math-----------------------------
1632 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1633 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1634 "must be (DD)D or (D)D type");
1635
1636 // Inputs
1637 Node* a = round_double_node(argument(0));
1638 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1639
1640 const TypePtr* no_memory_effects = NULL;
1641 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1642 no_memory_effects,
1643 a, top(), b, b ? top() : NULL);
1644 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1645 #ifdef ASSERT
1646 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1647 assert(value_top == top(), "second value must be top");
1648 #endif
1649
1650 set_result(value);
1651 return true;
1652 }
1653
1654 //------------------------------inline_math_pow-----------------------------
1655 bool LibraryCallKit::inline_math_pow() {
1656 Node* exp = round_double_node(argument(2));
1657 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1658 if (d != NULL) {
1659 if (d->getd() == 2.0) {
1660 // Special case: pow(x, 2.0) => x * x
1661 Node* base = round_double_node(argument(0));
1662 set_result(_gvn.transform(new MulDNode(base, base)));
1663 return true;
1664 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1665 // Special case: pow(x, 0.5) => sqrt(x)
1666 Node* base = round_double_node(argument(0));
1667 Node* zero = _gvn.zerocon(T_DOUBLE);
1668
1669 RegionNode* region = new RegionNode(3);
1670 Node* phi = new PhiNode(region, Type::DOUBLE);
1671
1672 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1673 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1674 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1675 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1676 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1677
1678 Node* if_pow = generate_slow_guard(test, NULL);
1679 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1680 phi->init_req(1, value_sqrt);
1681 region->init_req(1, control());
1682
1683 if (if_pow != NULL) {
1684 set_control(if_pow);
1685 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1686 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1687 const TypePtr* no_memory_effects = NULL;
1688 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1689 no_memory_effects, base, top(), exp, top());
1690 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1691 #ifdef ASSERT
1692 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1693 assert(value_top == top(), "second value must be top");
1694 #endif
1695 phi->init_req(2, value_pow);
1696 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1697 }
1698
1699 C->set_has_split_ifs(true); // Has chance for split-if optimization
1700 set_control(_gvn.transform(region));
1701 record_for_igvn(region);
1702 set_result(_gvn.transform(phi));
1703
1704 return true;
1705 }
1706 }
1707
1708 return StubRoutines::dpow() != NULL ?
1709 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1710 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1711 }
1712
1713 //------------------------------inline_math_native-----------------------------
1714 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1715 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1716 switch (id) {
1717 // These intrinsics are not properly supported on all hardware
1718 case vmIntrinsics::_dsin:
1719 return StubRoutines::dsin() != NULL ?
1720 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1721 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1722 case vmIntrinsics::_dcos:
1723 return StubRoutines::dcos() != NULL ?
1724 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1725 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1726 case vmIntrinsics::_dtan:
1727 return StubRoutines::dtan() != NULL ?
1728 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1729 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1730 case vmIntrinsics::_dlog:
1731 return StubRoutines::dlog() != NULL ?
1732 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1733 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1734 case vmIntrinsics::_dlog10:
1735 return StubRoutines::dlog10() != NULL ?
1736 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1737 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1738
1739 // These intrinsics are supported on all hardware
1740 case vmIntrinsics::_ceil:
1741 case vmIntrinsics::_floor:
1742 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1743 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1744 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1745 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1746 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1747 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1748
1749 case vmIntrinsics::_dexp:
1750 return StubRoutines::dexp() != NULL ?
1751 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1752 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1753 #undef FN_PTR
1754
1755 case vmIntrinsics::_dpow: return inline_math_pow();
1756 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1757 case vmIntrinsics::_fcopySign: return inline_math(id);
1758 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1759 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1760
1761 // These intrinsics are not yet correctly implemented
1762 case vmIntrinsics::_datan2:
1763 return false;
1764
1765 default:
1766 fatal_unexpected_iid(id);
1767 return false;
1768 }
1769 }
1770
1771 static bool is_simple_name(Node* n) {
1772 return (n->req() == 1 // constant
1773 || (n->is_Type() && n->as_Type()->type()->singleton())
1774 || n->is_Proj() // parameter or return value
1775 || n->is_Phi() // local of some sort
1776 );
1777 }
1778
1779 //----------------------------inline_notify-----------------------------------*
1780 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1781 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1782 address func;
1783 if (id == vmIntrinsics::_notify) {
1784 func = OptoRuntime::monitor_notify_Java();
1785 } else {
1786 func = OptoRuntime::monitor_notifyAll_Java();
1787 }
1788 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1789 make_slow_call_ex(call, env()->Throwable_klass(), false);
1790 return true;
1791 }
1792
1793
1794 //----------------------------inline_min_max-----------------------------------
1795 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1796 set_result(generate_min_max(id, argument(0), argument(1)));
1797 return true;
1798 }
1799
1800 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1801 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1802 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1803 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1804 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1805
1806 {
1807 PreserveJVMState pjvms(this);
1808 PreserveReexecuteState preexecs(this);
1809 jvms()->set_should_reexecute(true);
1810
1811 set_control(slow_path);
1812 set_i_o(i_o());
1813
1814 uncommon_trap(Deoptimization::Reason_intrinsic,
1815 Deoptimization::Action_none);
1816 }
1817
1818 set_control(fast_path);
1819 set_result(math);
1820 }
1821
1822 template <typename OverflowOp>
1823 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1824 typedef typename OverflowOp::MathOp MathOp;
1825
1826 MathOp* mathOp = new MathOp(arg1, arg2);
1827 Node* operation = _gvn.transform( mathOp );
1828 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1829 inline_math_mathExact(operation, ofcheck);
1830 return true;
1831 }
1832
1833 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1834 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1835 }
1836
1837 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1838 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1839 }
1840
1841 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1842 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1843 }
1844
1845 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1846 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1847 }
1848
1849 bool LibraryCallKit::inline_math_negateExactI() {
1850 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1851 }
1852
1853 bool LibraryCallKit::inline_math_negateExactL() {
1854 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1855 }
1856
1857 bool LibraryCallKit::inline_math_multiplyExactI() {
1858 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1859 }
1860
1861 bool LibraryCallKit::inline_math_multiplyExactL() {
1862 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1863 }
1864
1865 bool LibraryCallKit::inline_math_multiplyHigh() {
1866 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1867 return true;
1868 }
1869
1870 Node*
1871 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1872 // These are the candidate return value:
1873 Node* xvalue = x0;
1874 Node* yvalue = y0;
1875
1876 if (xvalue == yvalue) {
1877 return xvalue;
1878 }
1879
1880 bool want_max = (id == vmIntrinsics::_max);
1881
1882 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1883 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1884 if (txvalue == NULL || tyvalue == NULL) return top();
1885 // This is not really necessary, but it is consistent with a
1886 // hypothetical MaxINode::Value method:
1887 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1888
1889 // %%% This folding logic should (ideally) be in a different place.
1890 // Some should be inside IfNode, and there to be a more reliable
1891 // transformation of ?: style patterns into cmoves. We also want
1892 // more powerful optimizations around cmove and min/max.
1893
1894 // Try to find a dominating comparison of these guys.
1895 // It can simplify the index computation for Arrays.copyOf
1896 // and similar uses of System.arraycopy.
1897 // First, compute the normalized version of CmpI(x, y).
1898 int cmp_op = Op_CmpI;
1899 Node* xkey = xvalue;
1900 Node* ykey = yvalue;
1901 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1902 if (ideal_cmpxy->is_Cmp()) {
1903 // E.g., if we have CmpI(length - offset, count),
1904 // it might idealize to CmpI(length, count + offset)
1905 cmp_op = ideal_cmpxy->Opcode();
1906 xkey = ideal_cmpxy->in(1);
1907 ykey = ideal_cmpxy->in(2);
1908 }
1909
1910 // Start by locating any relevant comparisons.
1911 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1912 Node* cmpxy = NULL;
1913 Node* cmpyx = NULL;
1914 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1915 Node* cmp = start_from->fast_out(k);
1916 if (cmp->outcnt() > 0 && // must have prior uses
1917 cmp->in(0) == NULL && // must be context-independent
1918 cmp->Opcode() == cmp_op) { // right kind of compare
1919 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1920 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1921 }
1922 }
1923
1924 const int NCMPS = 2;
1925 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1926 int cmpn;
1927 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1928 if (cmps[cmpn] != NULL) break; // find a result
1929 }
1930 if (cmpn < NCMPS) {
1931 // Look for a dominating test that tells us the min and max.
1932 int depth = 0; // Limit search depth for speed
1933 Node* dom = control();
1934 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1935 if (++depth >= 100) break;
1936 Node* ifproj = dom;
1937 if (!ifproj->is_Proj()) continue;
1938 Node* iff = ifproj->in(0);
1939 if (!iff->is_If()) continue;
1940 Node* bol = iff->in(1);
1941 if (!bol->is_Bool()) continue;
1942 Node* cmp = bol->in(1);
1943 if (cmp == NULL) continue;
1944 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1945 if (cmps[cmpn] == cmp) break;
1946 if (cmpn == NCMPS) continue;
1947 BoolTest::mask btest = bol->as_Bool()->_test._test;
1948 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1949 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1950 // At this point, we know that 'x btest y' is true.
1951 switch (btest) {
1952 case BoolTest::eq:
1953 // They are proven equal, so we can collapse the min/max.
1954 // Either value is the answer. Choose the simpler.
1955 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1956 return yvalue;
1957 return xvalue;
1958 case BoolTest::lt: // x < y
1959 case BoolTest::le: // x <= y
1960 return (want_max ? yvalue : xvalue);
1961 case BoolTest::gt: // x > y
1962 case BoolTest::ge: // x >= y
1963 return (want_max ? xvalue : yvalue);
1964 default:
1965 break;
1966 }
1967 }
1968 }
1969
1970 // We failed to find a dominating test.
1971 // Let's pick a test that might GVN with prior tests.
1972 Node* best_bol = NULL;
1973 BoolTest::mask best_btest = BoolTest::illegal;
1974 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1975 Node* cmp = cmps[cmpn];
1976 if (cmp == NULL) continue;
1977 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
1978 Node* bol = cmp->fast_out(j);
1979 if (!bol->is_Bool()) continue;
1980 BoolTest::mask btest = bol->as_Bool()->_test._test;
1981 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
1982 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1983 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
1984 best_bol = bol->as_Bool();
1985 best_btest = btest;
1986 }
1987 }
1988 }
1989
1990 Node* answer_if_true = NULL;
1991 Node* answer_if_false = NULL;
1992 switch (best_btest) {
1993 default:
1994 if (cmpxy == NULL)
1995 cmpxy = ideal_cmpxy;
1996 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
1997 // and fall through:
1998 case BoolTest::lt: // x < y
1999 case BoolTest::le: // x <= y
2000 answer_if_true = (want_max ? yvalue : xvalue);
2001 answer_if_false = (want_max ? xvalue : yvalue);
2002 break;
2003 case BoolTest::gt: // x > y
2004 case BoolTest::ge: // x >= y
2005 answer_if_true = (want_max ? xvalue : yvalue);
2006 answer_if_false = (want_max ? yvalue : xvalue);
2007 break;
2008 }
2009
2010 jint hi, lo;
2011 if (want_max) {
2012 // We can sharpen the minimum.
2013 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2014 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2015 } else {
2016 // We can sharpen the maximum.
2017 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2018 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2019 }
2020
2021 // Use a flow-free graph structure, to avoid creating excess control edges
2022 // which could hinder other optimizations.
2023 // Since Math.min/max is often used with arraycopy, we want
2024 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2025 Node* cmov = CMoveNode::make(NULL, best_bol,
2026 answer_if_false, answer_if_true,
2027 TypeInt::make(lo, hi, widen));
2028
2029 return _gvn.transform(cmov);
2030
2031 /*
2032 // This is not as desirable as it may seem, since Min and Max
2033 // nodes do not have a full set of optimizations.
2034 // And they would interfere, anyway, with 'if' optimizations
2035 // and with CMoveI canonical forms.
2036 switch (id) {
2037 case vmIntrinsics::_min:
2038 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2039 case vmIntrinsics::_max:
2040 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2041 default:
2042 ShouldNotReachHere();
2043 }
2044 */
2045 }
2046
2047 inline int
2048 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2049 const TypePtr* base_type = TypePtr::NULL_PTR;
2050 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2051 if (base_type == NULL) {
2052 // Unknown type.
2053 return Type::AnyPtr;
2054 } else if (base_type == TypePtr::NULL_PTR) {
2055 // Since this is a NULL+long form, we have to switch to a rawptr.
2056 base = _gvn.transform(new CastX2PNode(offset));
2057 offset = MakeConX(0);
2058 return Type::RawPtr;
2059 } else if (base_type->base() == Type::RawPtr) {
2060 return Type::RawPtr;
2061 } else if (base_type->isa_oopptr()) {
2062 // Base is never null => always a heap address.
2063 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2064 return Type::OopPtr;
2065 }
2066 // Offset is small => always a heap address.
2067 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2068 if (offset_type != NULL &&
2069 base_type->offset() == 0 && // (should always be?)
2070 offset_type->_lo >= 0 &&
2071 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2072 return Type::OopPtr;
2073 } else if (type == T_OBJECT) {
2074 // off heap access to an oop doesn't make any sense. Has to be on
2075 // heap.
2076 return Type::OopPtr;
2077 }
2078 // Otherwise, it might either be oop+off or NULL+addr.
2079 return Type::AnyPtr;
2080 } else {
2081 // No information:
2082 return Type::AnyPtr;
2083 }
2084 }
2085
2086 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2087 Node* uncasted_base = base;
2088 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2089 if (kind == Type::RawPtr) {
2090 return basic_plus_adr(top(), uncasted_base, offset);
2091 } else if (kind == Type::AnyPtr) {
2092 assert(base == uncasted_base, "unexpected base change");
2093 if (can_cast) {
2094 if (!_gvn.type(base)->speculative_maybe_null() &&
2095 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2096 // According to profiling, this access is always on
2097 // heap. Casting the base to not null and thus avoiding membars
2098 // around the access should allow better optimizations
2099 Node* null_ctl = top();
2100 base = null_check_oop(base, &null_ctl, true, true, true);
2101 assert(null_ctl->is_top(), "no null control here");
2102 return basic_plus_adr(base, offset);
2103 } else if (_gvn.type(base)->speculative_always_null() &&
2104 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2105 // According to profiling, this access is always off
2106 // heap.
2107 base = null_assert(base);
2108 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2109 offset = MakeConX(0);
2110 return basic_plus_adr(top(), raw_base, offset);
2111 }
2112 }
2113 // We don't know if it's an on heap or off heap access. Fall back
2114 // to raw memory access.
2115 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2116 return basic_plus_adr(top(), raw, offset);
2117 } else {
2118 assert(base == uncasted_base, "unexpected base change");
2119 // We know it's an on heap access so base can't be null
2120 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2121 base = must_be_not_null(base, true);
2122 }
2123 return basic_plus_adr(base, offset);
2124 }
2125 }
2126
2127 //--------------------------inline_number_methods-----------------------------
2128 // inline int Integer.numberOfLeadingZeros(int)
2129 // inline int Long.numberOfLeadingZeros(long)
2130 //
2131 // inline int Integer.numberOfTrailingZeros(int)
2132 // inline int Long.numberOfTrailingZeros(long)
2133 //
2134 // inline int Integer.bitCount(int)
2135 // inline int Long.bitCount(long)
2136 //
2137 // inline char Character.reverseBytes(char)
2138 // inline short Short.reverseBytes(short)
2139 // inline int Integer.reverseBytes(int)
2140 // inline long Long.reverseBytes(long)
2141 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2142 Node* arg = argument(0);
2143 Node* n = NULL;
2144 switch (id) {
2145 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2146 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2147 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2148 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2149 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2150 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2151 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2152 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2153 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2154 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2155 default: fatal_unexpected_iid(id); break;
2156 }
2157 set_result(_gvn.transform(n));
2158 return true;
2159 }
2160
2161 //----------------------------inline_unsafe_access----------------------------
2162
2163 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2164 // Attempt to infer a sharper value type from the offset and base type.
2165 ciKlass* sharpened_klass = NULL;
2166
2167 // See if it is an instance field, with an object type.
2168 if (alias_type->field() != NULL) {
2169 if (alias_type->field()->type()->is_klass()) {
2170 sharpened_klass = alias_type->field()->type()->as_klass();
2171 }
2172 }
2173
2174 // See if it is a narrow oop array.
2175 if (adr_type->isa_aryptr()) {
2176 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2177 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2178 if (elem_type != NULL) {
2179 sharpened_klass = elem_type->klass();
2180 }
2181 }
2182 }
2183
2184 // The sharpened class might be unloaded if there is no class loader
2185 // contraint in place.
2186 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2187 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2188
2189 #ifndef PRODUCT
2190 if (C->print_intrinsics() || C->print_inlining()) {
2191 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2192 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2193 }
2194 #endif
2195 // Sharpen the value type.
2196 return tjp;
2197 }
2198 return NULL;
2199 }
2200
2201 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2202 switch (kind) {
2203 case Relaxed:
2204 return MO_UNORDERED;
2205 case Opaque:
2206 return MO_RELAXED;
2207 case Acquire:
2208 return MO_ACQUIRE;
2209 case Release:
2210 return MO_RELEASE;
2211 case Volatile:
2212 return MO_SEQ_CST;
2213 default:
2214 ShouldNotReachHere();
2215 return 0;
2216 }
2217 }
2218
2219 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2220 if (callee()->is_static()) return false; // caller must have the capability!
2221 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2222 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2223 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2224 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2225
2226 if (is_reference_type(type)) {
2227 decorators |= ON_UNKNOWN_OOP_REF;
2228 }
2229
2230 if (unaligned) {
2231 decorators |= C2_UNALIGNED;
2232 }
2233
2234 #ifndef PRODUCT
2235 {
2236 ResourceMark rm;
2237 // Check the signatures.
2238 ciSignature* sig = callee()->signature();
2239 #ifdef ASSERT
2240 if (!is_store) {
2241 // Object getReference(Object base, int/long offset), etc.
2242 BasicType rtype = sig->return_type()->basic_type();
2243 assert(rtype == type, "getter must return the expected value");
2244 assert(sig->count() == 2, "oop getter has 2 arguments");
2245 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2246 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2247 } else {
2248 // void putReference(Object base, int/long offset, Object x), etc.
2249 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2250 assert(sig->count() == 3, "oop putter has 3 arguments");
2251 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2252 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2253 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2254 assert(vtype == type, "putter must accept the expected value");
2255 }
2256 #endif // ASSERT
2257 }
2258 #endif //PRODUCT
2259
2260 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2261
2262 Node* receiver = argument(0); // type: oop
2263
2264 // Build address expression.
2265 Node* heap_base_oop = top();
2266
2267 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2268 Node* base = argument(1); // type: oop
2269 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2270 Node* offset = argument(2); // type: long
2271 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2272 // to be plain byte offsets, which are also the same as those accepted
2273 // by oopDesc::field_addr.
2274 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2275 "fieldOffset must be byte-scaled");
2276 // 32-bit machines ignore the high half!
2277 offset = ConvL2X(offset);
2278
2279 // Save state and restore on bailout
2280 uint old_sp = sp();
2281 SafePointNode* old_map = clone_map();
2282
2283 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2284
2285 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2286 if (type != T_OBJECT) {
2287 decorators |= IN_NATIVE; // off-heap primitive access
2288 } else {
2289 set_map(old_map);
2290 set_sp(old_sp);
2291 return false; // off-heap oop accesses are not supported
2292 }
2293 } else {
2294 heap_base_oop = base; // on-heap or mixed access
2295 }
2296
2297 // Can base be NULL? Otherwise, always on-heap access.
2298 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2299
2300 if (!can_access_non_heap) {
2301 decorators |= IN_HEAP;
2302 }
2303
2304 Node* val = is_store ? argument(4) : NULL;
2305
2306 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2307 if (adr_type == TypePtr::NULL_PTR) {
2308 set_map(old_map);
2309 set_sp(old_sp);
2310 return false; // off-heap access with zero address
2311 }
2312
2313 // Try to categorize the address.
2314 Compile::AliasType* alias_type = C->alias_type(adr_type);
2315 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2316
2317 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2318 alias_type->adr_type() == TypeAryPtr::RANGE) {
2319 set_map(old_map);
2320 set_sp(old_sp);
2321 return false; // not supported
2322 }
2323
2324 bool mismatched = false;
2325 BasicType bt = alias_type->basic_type();
2326 if (bt != T_ILLEGAL) {
2327 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2328 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2329 // Alias type doesn't differentiate between byte[] and boolean[]).
2330 // Use address type to get the element type.
2331 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2332 }
2333 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2334 // accessing an array field with getReference is not a mismatch
2335 bt = T_OBJECT;
2336 }
2337 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2338 // Don't intrinsify mismatched object accesses
2339 set_map(old_map);
2340 set_sp(old_sp);
2341 return false;
2342 }
2343 mismatched = (bt != type);
2344 } else if (alias_type->adr_type()->isa_oopptr()) {
2345 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2346 }
2347
2348 old_map->destruct(&_gvn);
2349 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2350
2351 if (mismatched) {
2352 decorators |= C2_MISMATCHED;
2353 }
2354
2355 // First guess at the value type.
2356 const Type *value_type = Type::get_const_basic_type(type);
2357
2358 // Figure out the memory ordering.
2359 decorators |= mo_decorator_for_access_kind(kind);
2360
2361 if (!is_store && type == T_OBJECT) {
2362 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2363 if (tjp != NULL) {
2364 value_type = tjp;
2365 }
2366 }
2367
2368 receiver = null_check(receiver);
2369 if (stopped()) {
2370 return true;
2371 }
2372 // Heap pointers get a null-check from the interpreter,
2373 // as a courtesy. However, this is not guaranteed by Unsafe,
2374 // and it is not possible to fully distinguish unintended nulls
2375 // from intended ones in this API.
2376
2377 if (!is_store) {
2378 Node* p = NULL;
2379 // Try to constant fold a load from a constant field
2380 ciField* field = alias_type->field();
2381 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2382 // final or stable field
2383 p = make_constant_from_field(field, heap_base_oop);
2384 }
2385
2386 if (p == NULL) { // Could not constant fold the load
2387 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2388 // Normalize the value returned by getBoolean in the following cases
2389 if (type == T_BOOLEAN &&
2390 (mismatched ||
2391 heap_base_oop == top() || // - heap_base_oop is NULL or
2392 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2393 // and the unsafe access is made to large offset
2394 // (i.e., larger than the maximum offset necessary for any
2395 // field access)
2396 ) {
2397 IdealKit ideal = IdealKit(this);
2398 #define __ ideal.
2399 IdealVariable normalized_result(ideal);
2400 __ declarations_done();
2401 __ set(normalized_result, p);
2402 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2403 __ set(normalized_result, ideal.ConI(1));
2404 ideal.end_if();
2405 final_sync(ideal);
2406 p = __ value(normalized_result);
2407 #undef __
2408 }
2409 }
2410 if (type == T_ADDRESS) {
2411 p = gvn().transform(new CastP2XNode(NULL, p));
2412 p = ConvX2UL(p);
2413 }
2414 // The load node has the control of the preceding MemBarCPUOrder. All
2415 // following nodes will have the control of the MemBarCPUOrder inserted at
2416 // the end of this method. So, pushing the load onto the stack at a later
2417 // point is fine.
2418 set_result(p);
2419 } else {
2420 if (bt == T_ADDRESS) {
2421 // Repackage the long as a pointer.
2422 val = ConvL2X(val);
2423 val = gvn().transform(new CastX2PNode(val));
2424 }
2425 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2426 }
2427
2428 return true;
2429 }
2430
2431 //----------------------------inline_unsafe_load_store----------------------------
2432 // This method serves a couple of different customers (depending on LoadStoreKind):
2433 //
2434 // LS_cmp_swap:
2435 //
2436 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2437 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2438 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2439 //
2440 // LS_cmp_swap_weak:
2441 //
2442 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2443 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2444 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2445 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2446 //
2447 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2448 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2449 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2450 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2451 //
2452 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2453 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2454 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2455 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2456 //
2457 // LS_cmp_exchange:
2458 //
2459 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2460 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2461 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2462 //
2463 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2464 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2465 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2466 //
2467 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2468 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2469 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2470 //
2471 // LS_get_add:
2472 //
2473 // int getAndAddInt( Object o, long offset, int delta)
2474 // long getAndAddLong(Object o, long offset, long delta)
2475 //
2476 // LS_get_set:
2477 //
2478 // int getAndSet(Object o, long offset, int newValue)
2479 // long getAndSet(Object o, long offset, long newValue)
2480 // Object getAndSet(Object o, long offset, Object newValue)
2481 //
2482 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2483 // This basic scheme here is the same as inline_unsafe_access, but
2484 // differs in enough details that combining them would make the code
2485 // overly confusing. (This is a true fact! I originally combined
2486 // them, but even I was confused by it!) As much code/comments as
2487 // possible are retained from inline_unsafe_access though to make
2488 // the correspondences clearer. - dl
2489
2490 if (callee()->is_static()) return false; // caller must have the capability!
2491
2492 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2493 decorators |= mo_decorator_for_access_kind(access_kind);
2494
2495 #ifndef PRODUCT
2496 BasicType rtype;
2497 {
2498 ResourceMark rm;
2499 // Check the signatures.
2500 ciSignature* sig = callee()->signature();
2501 rtype = sig->return_type()->basic_type();
2502 switch(kind) {
2503 case LS_get_add:
2504 case LS_get_set: {
2505 // Check the signatures.
2506 #ifdef ASSERT
2507 assert(rtype == type, "get and set must return the expected type");
2508 assert(sig->count() == 3, "get and set has 3 arguments");
2509 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2510 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2511 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2512 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2513 #endif // ASSERT
2514 break;
2515 }
2516 case LS_cmp_swap:
2517 case LS_cmp_swap_weak: {
2518 // Check the signatures.
2519 #ifdef ASSERT
2520 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2521 assert(sig->count() == 4, "CAS has 4 arguments");
2522 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2523 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2524 #endif // ASSERT
2525 break;
2526 }
2527 case LS_cmp_exchange: {
2528 // Check the signatures.
2529 #ifdef ASSERT
2530 assert(rtype == type, "CAS must return the expected type");
2531 assert(sig->count() == 4, "CAS has 4 arguments");
2532 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2533 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2534 #endif // ASSERT
2535 break;
2536 }
2537 default:
2538 ShouldNotReachHere();
2539 }
2540 }
2541 #endif //PRODUCT
2542
2543 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2544
2545 // Get arguments:
2546 Node* receiver = NULL;
2547 Node* base = NULL;
2548 Node* offset = NULL;
2549 Node* oldval = NULL;
2550 Node* newval = NULL;
2551 switch(kind) {
2552 case LS_cmp_swap:
2553 case LS_cmp_swap_weak:
2554 case LS_cmp_exchange: {
2555 const bool two_slot_type = type2size[type] == 2;
2556 receiver = argument(0); // type: oop
2557 base = argument(1); // type: oop
2558 offset = argument(2); // type: long
2559 oldval = argument(4); // type: oop, int, or long
2560 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2561 break;
2562 }
2563 case LS_get_add:
2564 case LS_get_set: {
2565 receiver = argument(0); // type: oop
2566 base = argument(1); // type: oop
2567 offset = argument(2); // type: long
2568 oldval = NULL;
2569 newval = argument(4); // type: oop, int, or long
2570 break;
2571 }
2572 default:
2573 ShouldNotReachHere();
2574 }
2575
2576 // Build field offset expression.
2577 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2578 // to be plain byte offsets, which are also the same as those accepted
2579 // by oopDesc::field_addr.
2580 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2581 // 32-bit machines ignore the high half of long offsets
2582 offset = ConvL2X(offset);
2583 // Save state and restore on bailout
2584 uint old_sp = sp();
2585 SafePointNode* old_map = clone_map();
2586 Node* adr = make_unsafe_address(base, offset,type, false);
2587 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2588
2589 Compile::AliasType* alias_type = C->alias_type(adr_type);
2590 BasicType bt = alias_type->basic_type();
2591 if (bt != T_ILLEGAL &&
2592 (is_reference_type(bt) != (type == T_OBJECT))) {
2593 // Don't intrinsify mismatched object accesses.
2594 set_map(old_map);
2595 set_sp(old_sp);
2596 return false;
2597 }
2598
2599 old_map->destruct(&_gvn);
2600
2601 // For CAS, unlike inline_unsafe_access, there seems no point in
2602 // trying to refine types. Just use the coarse types here.
2603 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2604 const Type *value_type = Type::get_const_basic_type(type);
2605
2606 switch (kind) {
2607 case LS_get_set:
2608 case LS_cmp_exchange: {
2609 if (type == T_OBJECT) {
2610 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2611 if (tjp != NULL) {
2612 value_type = tjp;
2613 }
2614 }
2615 break;
2616 }
2617 case LS_cmp_swap:
2618 case LS_cmp_swap_weak:
2619 case LS_get_add:
2620 break;
2621 default:
2622 ShouldNotReachHere();
2623 }
2624
2625 // Null check receiver.
2626 receiver = null_check(receiver);
2627 if (stopped()) {
2628 return true;
2629 }
2630
2631 int alias_idx = C->get_alias_index(adr_type);
2632
2633 if (is_reference_type(type)) {
2634 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2635
2636 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2637 // could be delayed during Parse (for example, in adjust_map_after_if()).
2638 // Execute transformation here to avoid barrier generation in such case.
2639 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2640 newval = _gvn.makecon(TypePtr::NULL_PTR);
2641
2642 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2643 // Refine the value to a null constant, when it is known to be null
2644 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2645 }
2646 }
2647
2648 Node* result = NULL;
2649 switch (kind) {
2650 case LS_cmp_exchange: {
2651 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2652 oldval, newval, value_type, type, decorators);
2653 break;
2654 }
2655 case LS_cmp_swap_weak:
2656 decorators |= C2_WEAK_CMPXCHG;
2657 case LS_cmp_swap: {
2658 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2659 oldval, newval, value_type, type, decorators);
2660 break;
2661 }
2662 case LS_get_set: {
2663 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2664 newval, value_type, type, decorators);
2665 break;
2666 }
2667 case LS_get_add: {
2668 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2669 newval, value_type, type, decorators);
2670 break;
2671 }
2672 default:
2673 ShouldNotReachHere();
2674 }
2675
2676 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2677 set_result(result);
2678 return true;
2679 }
2680
2681 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2682 // Regardless of form, don't allow previous ld/st to move down,
2683 // then issue acquire, release, or volatile mem_bar.
2684 insert_mem_bar(Op_MemBarCPUOrder);
2685 switch(id) {
2686 case vmIntrinsics::_loadFence:
2687 insert_mem_bar(Op_LoadFence);
2688 return true;
2689 case vmIntrinsics::_storeFence:
2690 insert_mem_bar(Op_StoreFence);
2691 return true;
2692 case vmIntrinsics::_fullFence:
2693 insert_mem_bar(Op_MemBarVolatile);
2694 return true;
2695 default:
2696 fatal_unexpected_iid(id);
2697 return false;
2698 }
2699 }
2700
2701 bool LibraryCallKit::inline_onspinwait() {
2702 insert_mem_bar(Op_OnSpinWait);
2703 return true;
2704 }
2705
2706 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2707 if (!kls->is_Con()) {
2708 return true;
2709 }
2710 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2711 if (klsptr == NULL) {
2712 return true;
2713 }
2714 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2715 // don't need a guard for a klass that is already initialized
2716 return !ik->is_initialized();
2717 }
2718
2719 //----------------------------inline_unsafe_writeback0-------------------------
2720 // public native void Unsafe.writeback0(long address)
2721 bool LibraryCallKit::inline_unsafe_writeback0() {
2722 if (!Matcher::has_match_rule(Op_CacheWB)) {
2723 return false;
2724 }
2725 #ifndef PRODUCT
2726 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2727 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2728 ciSignature* sig = callee()->signature();
2729 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2730 #endif
2731 null_check_receiver(); // null-check, then ignore
2732 Node *addr = argument(1);
2733 addr = new CastX2PNode(addr);
2734 addr = _gvn.transform(addr);
2735 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2736 flush = _gvn.transform(flush);
2737 set_memory(flush, TypeRawPtr::BOTTOM);
2738 return true;
2739 }
2740
2741 //----------------------------inline_unsafe_writeback0-------------------------
2742 // public native void Unsafe.writeback0(long address)
2743 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2744 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2745 return false;
2746 }
2747 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2748 return false;
2749 }
2750 #ifndef PRODUCT
2751 assert(Matcher::has_match_rule(Op_CacheWB),
2752 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2753 : "found match rule for CacheWBPostSync but not CacheWB"));
2754
2755 #endif
2756 null_check_receiver(); // null-check, then ignore
2757 Node *sync;
2758 if (is_pre) {
2759 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2760 } else {
2761 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2762 }
2763 sync = _gvn.transform(sync);
2764 set_memory(sync, TypeRawPtr::BOTTOM);
2765 return true;
2766 }
2767
2768 //----------------------------inline_unsafe_allocate---------------------------
2769 // public native Object Unsafe.allocateInstance(Class<?> cls);
2770 bool LibraryCallKit::inline_unsafe_allocate() {
2771 if (callee()->is_static()) return false; // caller must have the capability!
2772
2773 null_check_receiver(); // null-check, then ignore
2774 Node* cls = null_check(argument(1));
2775 if (stopped()) return true;
2776
2777 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2778 kls = null_check(kls);
2779 if (stopped()) return true; // argument was like int.class
2780
2781 Node* test = NULL;
2782 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2783 // Note: The argument might still be an illegal value like
2784 // Serializable.class or Object[].class. The runtime will handle it.
2785 // But we must make an explicit check for initialization.
2786 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2787 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2788 // can generate code to load it as unsigned byte.
2789 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2790 Node* bits = intcon(InstanceKlass::fully_initialized);
2791 test = _gvn.transform(new SubINode(inst, bits));
2792 // The 'test' is non-zero if we need to take a slow path.
2793 }
2794
2795 Node* obj = new_instance(kls, test);
2796 set_result(obj);
2797 return true;
2798 }
2799
2800 //------------------------inline_native_time_funcs--------------
2801 // inline code for System.currentTimeMillis() and System.nanoTime()
2802 // these have the same type and signature
2803 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2804 const TypeFunc* tf = OptoRuntime::void_long_Type();
2805 const TypePtr* no_memory_effects = NULL;
2806 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2807 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2808 #ifdef ASSERT
2809 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2810 assert(value_top == top(), "second value must be top");
2811 #endif
2812 set_result(value);
2813 return true;
2814 }
2815
2816 #ifdef JFR_HAVE_INTRINSICS
2817
2818 /**
2819 * if oop->klass != null
2820 * // normal class
2821 * epoch = _epoch_state ? 2 : 1
2822 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2823 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2824 * }
2825 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2826 * else
2827 * // primitive class
2828 * if oop->array_klass != null
2829 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2830 * else
2831 * id = LAST_TYPE_ID + 1 // void class path
2832 * if (!signaled)
2833 * signaled = true
2834 */
2835 bool LibraryCallKit::inline_native_classID() {
2836 Node* cls = argument(0);
2837
2838 IdealKit ideal(this);
2839 #define __ ideal.
2840 IdealVariable result(ideal); __ declarations_done();
2841 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2842 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2843 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2844
2845
2846 __ if_then(kls, BoolTest::ne, null()); {
2847 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2848 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2849
2850 Node* epoch_address = makecon(TypeRawPtr::make(Jfr::epoch_address()));
2851 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2852 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2853 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2854 mask = _gvn.transform(new OrLNode(mask, epoch));
2855 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2856
2857 float unlikely = PROB_UNLIKELY(0.999);
2858 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
2859 sync_kit(ideal);
2860 make_runtime_call(RC_LEAF,
2861 OptoRuntime::get_class_id_intrinsic_Type(),
2862 CAST_FROM_FN_PTR(address, Jfr::get_class_id_intrinsic),
2863 "get_class_id_intrinsic",
2864 TypePtr::BOTTOM,
2865 kls);
2866 ideal.sync_kit(this);
2867 } __ end_if();
2868
2869 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
2870 } __ else_(); {
2871 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2872 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
2873 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2874 __ if_then(array_kls, BoolTest::ne, null()); {
2875 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2876 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2877 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
2878 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
2879 } __ else_(); {
2880 // void class case
2881 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
2882 } __ end_if();
2883
2884 Node* signaled_flag_address = makecon(TypeRawPtr::make(Jfr::signal_address()));
2885 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
2886 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
2887 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
2888 } __ end_if();
2889 } __ end_if();
2890
2891 final_sync(ideal);
2892 set_result(ideal.value(result));
2893 #undef __
2894 return true;
2895 }
2896
2897 bool LibraryCallKit::inline_native_getEventWriter() {
2898 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2899
2900 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2901 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2902
2903 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2904
2905 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2906 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2907
2908 IfNode* iff_jobj_null =
2909 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2910
2911 enum { _normal_path = 1,
2912 _null_path = 2,
2913 PATH_LIMIT };
2914
2915 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2916 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
2917
2918 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2919 result_rgn->init_req(_null_path, jobj_is_null);
2920 result_val->init_req(_null_path, null());
2921
2922 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2923 set_control(jobj_is_not_null);
2924 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
2925 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
2926 result_rgn->init_req(_normal_path, control());
2927 result_val->init_req(_normal_path, res);
2928
2929 set_result(result_rgn, result_val);
2930
2931 return true;
2932 }
2933
2934 #endif // JFR_HAVE_INTRINSICS
2935
2936 //------------------------inline_native_currentThread------------------
2937 bool LibraryCallKit::inline_native_currentThread() {
2938 Node* junk = NULL;
2939 set_result(generate_current_thread(junk));
2940 return true;
2941 }
2942
2943 //---------------------------load_mirror_from_klass----------------------------
2944 // Given a klass oop, load its java mirror (a java.lang.Class oop).
2945 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
2946 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
2947 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
2948 // mirror = ((OopHandle)mirror)->resolve();
2949 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
2950 }
2951
2952 //-----------------------load_klass_from_mirror_common-------------------------
2953 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
2954 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
2955 // and branch to the given path on the region.
2956 // If never_see_null, take an uncommon trap on null, so we can optimistically
2957 // compile for the non-null case.
2958 // If the region is NULL, force never_see_null = true.
2959 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
2960 bool never_see_null,
2961 RegionNode* region,
2962 int null_path,
2963 int offset) {
2964 if (region == NULL) never_see_null = true;
2965 Node* p = basic_plus_adr(mirror, offset);
2966 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
2967 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
2968 Node* null_ctl = top();
2969 kls = null_check_oop(kls, &null_ctl, never_see_null);
2970 if (region != NULL) {
2971 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
2972 region->init_req(null_path, null_ctl);
2973 } else {
2974 assert(null_ctl == top(), "no loose ends");
2975 }
2976 return kls;
2977 }
2978
2979 //--------------------(inline_native_Class_query helpers)---------------------
2980 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
2981 // Fall through if (mods & mask) == bits, take the guard otherwise.
2982 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
2983 // Branch around if the given klass has the given modifier bit set.
2984 // Like generate_guard, adds a new path onto the region.
2985 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
2986 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
2987 Node* mask = intcon(modifier_mask);
2988 Node* bits = intcon(modifier_bits);
2989 Node* mbit = _gvn.transform(new AndINode(mods, mask));
2990 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
2991 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
2992 return generate_fair_guard(bol, region);
2993 }
2994 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
2995 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
2996 }
2997 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
2998 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
2999 }
3000
3001 //-------------------------inline_native_Class_query-------------------
3002 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3003 const Type* return_type = TypeInt::BOOL;
3004 Node* prim_return_value = top(); // what happens if it's a primitive class?
3005 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3006 bool expect_prim = false; // most of these guys expect to work on refs
3007
3008 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3009
3010 Node* mirror = argument(0);
3011 Node* obj = top();
3012
3013 switch (id) {
3014 case vmIntrinsics::_isInstance:
3015 // nothing is an instance of a primitive type
3016 prim_return_value = intcon(0);
3017 obj = argument(1);
3018 break;
3019 case vmIntrinsics::_getModifiers:
3020 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3021 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3022 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3023 break;
3024 case vmIntrinsics::_isInterface:
3025 prim_return_value = intcon(0);
3026 break;
3027 case vmIntrinsics::_isArray:
3028 prim_return_value = intcon(0);
3029 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3030 break;
3031 case vmIntrinsics::_isPrimitive:
3032 prim_return_value = intcon(1);
3033 expect_prim = true; // obviously
3034 break;
3035 case vmIntrinsics::_isHidden:
3036 prim_return_value = intcon(0);
3037 break;
3038 case vmIntrinsics::_getSuperclass:
3039 prim_return_value = null();
3040 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3041 break;
3042 case vmIntrinsics::_getClassAccessFlags:
3043 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3044 return_type = TypeInt::INT; // not bool! 6297094
3045 break;
3046 default:
3047 fatal_unexpected_iid(id);
3048 break;
3049 }
3050
3051 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3052 if (mirror_con == NULL) return false; // cannot happen?
3053
3054 #ifndef PRODUCT
3055 if (C->print_intrinsics() || C->print_inlining()) {
3056 ciType* k = mirror_con->java_mirror_type();
3057 if (k) {
3058 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3059 k->print_name();
3060 tty->cr();
3061 }
3062 }
3063 #endif
3064
3065 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3066 RegionNode* region = new RegionNode(PATH_LIMIT);
3067 record_for_igvn(region);
3068 PhiNode* phi = new PhiNode(region, return_type);
3069
3070 // The mirror will never be null of Reflection.getClassAccessFlags, however
3071 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3072 // if it is. See bug 4774291.
3073
3074 // For Reflection.getClassAccessFlags(), the null check occurs in
3075 // the wrong place; see inline_unsafe_access(), above, for a similar
3076 // situation.
3077 mirror = null_check(mirror);
3078 // If mirror or obj is dead, only null-path is taken.
3079 if (stopped()) return true;
3080
3081 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3082
3083 // Now load the mirror's klass metaobject, and null-check it.
3084 // Side-effects region with the control path if the klass is null.
3085 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3086 // If kls is null, we have a primitive mirror.
3087 phi->init_req(_prim_path, prim_return_value);
3088 if (stopped()) { set_result(region, phi); return true; }
3089 bool safe_for_replace = (region->in(_prim_path) == top());
3090
3091 Node* p; // handy temp
3092 Node* null_ctl;
3093
3094 // Now that we have the non-null klass, we can perform the real query.
3095 // For constant classes, the query will constant-fold in LoadNode::Value.
3096 Node* query_value = top();
3097 switch (id) {
3098 case vmIntrinsics::_isInstance:
3099 // nothing is an instance of a primitive type
3100 query_value = gen_instanceof(obj, kls, safe_for_replace);
3101 break;
3102
3103 case vmIntrinsics::_getModifiers:
3104 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3105 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3106 break;
3107
3108 case vmIntrinsics::_isInterface:
3109 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3110 if (generate_interface_guard(kls, region) != NULL)
3111 // A guard was added. If the guard is taken, it was an interface.
3112 phi->add_req(intcon(1));
3113 // If we fall through, it's a plain class.
3114 query_value = intcon(0);
3115 break;
3116
3117 case vmIntrinsics::_isArray:
3118 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3119 if (generate_array_guard(kls, region) != NULL)
3120 // A guard was added. If the guard is taken, it was an array.
3121 phi->add_req(intcon(1));
3122 // If we fall through, it's a plain class.
3123 query_value = intcon(0);
3124 break;
3125
3126 case vmIntrinsics::_isPrimitive:
3127 query_value = intcon(0); // "normal" path produces false
3128 break;
3129
3130 case vmIntrinsics::_isHidden:
3131 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3132 if (generate_hidden_class_guard(kls, region) != NULL)
3133 // A guard was added. If the guard is taken, it was an hidden class.
3134 phi->add_req(intcon(1));
3135 // If we fall through, it's a plain class.
3136 query_value = intcon(0);
3137 break;
3138
3139
3140 case vmIntrinsics::_getSuperclass:
3141 // The rules here are somewhat unfortunate, but we can still do better
3142 // with random logic than with a JNI call.
3143 // Interfaces store null or Object as _super, but must report null.
3144 // Arrays store an intermediate super as _super, but must report Object.
3145 // Other types can report the actual _super.
3146 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3147 if (generate_interface_guard(kls, region) != NULL)
3148 // A guard was added. If the guard is taken, it was an interface.
3149 phi->add_req(null());
3150 if (generate_array_guard(kls, region) != NULL)
3151 // A guard was added. If the guard is taken, it was an array.
3152 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3153 // If we fall through, it's a plain class. Get its _super.
3154 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3155 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3156 null_ctl = top();
3157 kls = null_check_oop(kls, &null_ctl);
3158 if (null_ctl != top()) {
3159 // If the guard is taken, Object.superClass is null (both klass and mirror).
3160 region->add_req(null_ctl);
3161 phi ->add_req(null());
3162 }
3163 if (!stopped()) {
3164 query_value = load_mirror_from_klass(kls);
3165 }
3166 break;
3167
3168 case vmIntrinsics::_getClassAccessFlags:
3169 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3170 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3171 break;
3172
3173 default:
3174 fatal_unexpected_iid(id);
3175 break;
3176 }
3177
3178 // Fall-through is the normal case of a query to a real class.
3179 phi->init_req(1, query_value);
3180 region->init_req(1, control());
3181
3182 C->set_has_split_ifs(true); // Has chance for split-if optimization
3183 set_result(region, phi);
3184 return true;
3185 }
3186
3187 //-------------------------inline_Class_cast-------------------
3188 bool LibraryCallKit::inline_Class_cast() {
3189 Node* mirror = argument(0); // Class
3190 Node* obj = argument(1);
3191 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3192 if (mirror_con == NULL) {
3193 return false; // dead path (mirror->is_top()).
3194 }
3195 if (obj == NULL || obj->is_top()) {
3196 return false; // dead path
3197 }
3198 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3199
3200 // First, see if Class.cast() can be folded statically.
3201 // java_mirror_type() returns non-null for compile-time Class constants.
3202 ciType* tm = mirror_con->java_mirror_type();
3203 if (tm != NULL && tm->is_klass() &&
3204 tp != NULL && tp->klass() != NULL) {
3205 if (!tp->klass()->is_loaded()) {
3206 // Don't use intrinsic when class is not loaded.
3207 return false;
3208 } else {
3209 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3210 if (static_res == Compile::SSC_always_true) {
3211 // isInstance() is true - fold the code.
3212 set_result(obj);
3213 return true;
3214 } else if (static_res == Compile::SSC_always_false) {
3215 // Don't use intrinsic, have to throw ClassCastException.
3216 // If the reference is null, the non-intrinsic bytecode will
3217 // be optimized appropriately.
3218 return false;
3219 }
3220 }
3221 }
3222
3223 // Bailout intrinsic and do normal inlining if exception path is frequent.
3224 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3225 return false;
3226 }
3227
3228 // Generate dynamic checks.
3229 // Class.cast() is java implementation of _checkcast bytecode.
3230 // Do checkcast (Parse::do_checkcast()) optimizations here.
3231
3232 mirror = null_check(mirror);
3233 // If mirror is dead, only null-path is taken.
3234 if (stopped()) {
3235 return true;
3236 }
3237
3238 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3239 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3240 RegionNode* region = new RegionNode(PATH_LIMIT);
3241 record_for_igvn(region);
3242
3243 // Now load the mirror's klass metaobject, and null-check it.
3244 // If kls is null, we have a primitive mirror and
3245 // nothing is an instance of a primitive type.
3246 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3247
3248 Node* res = top();
3249 if (!stopped()) {
3250 Node* bad_type_ctrl = top();
3251 // Do checkcast optimizations.
3252 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3253 region->init_req(_bad_type_path, bad_type_ctrl);
3254 }
3255 if (region->in(_prim_path) != top() ||
3256 region->in(_bad_type_path) != top()) {
3257 // Let Interpreter throw ClassCastException.
3258 PreserveJVMState pjvms(this);
3259 set_control(_gvn.transform(region));
3260 uncommon_trap(Deoptimization::Reason_intrinsic,
3261 Deoptimization::Action_maybe_recompile);
3262 }
3263 if (!stopped()) {
3264 set_result(res);
3265 }
3266 return true;
3267 }
3268
3269
3270 //--------------------------inline_native_subtype_check------------------------
3271 // This intrinsic takes the JNI calls out of the heart of
3272 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3273 bool LibraryCallKit::inline_native_subtype_check() {
3274 // Pull both arguments off the stack.
3275 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3276 args[0] = argument(0);
3277 args[1] = argument(1);
3278 Node* klasses[2]; // corresponding Klasses: superk, subk
3279 klasses[0] = klasses[1] = top();
3280
3281 enum {
3282 // A full decision tree on {superc is prim, subc is prim}:
3283 _prim_0_path = 1, // {P,N} => false
3284 // {P,P} & superc!=subc => false
3285 _prim_same_path, // {P,P} & superc==subc => true
3286 _prim_1_path, // {N,P} => false
3287 _ref_subtype_path, // {N,N} & subtype check wins => true
3288 _both_ref_path, // {N,N} & subtype check loses => false
3289 PATH_LIMIT
3290 };
3291
3292 RegionNode* region = new RegionNode(PATH_LIMIT);
3293 Node* phi = new PhiNode(region, TypeInt::BOOL);
3294 record_for_igvn(region);
3295
3296 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3297 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3298 int class_klass_offset = java_lang_Class::klass_offset();
3299
3300 // First null-check both mirrors and load each mirror's klass metaobject.
3301 int which_arg;
3302 for (which_arg = 0; which_arg <= 1; which_arg++) {
3303 Node* arg = args[which_arg];
3304 arg = null_check(arg);
3305 if (stopped()) break;
3306 args[which_arg] = arg;
3307
3308 Node* p = basic_plus_adr(arg, class_klass_offset);
3309 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3310 klasses[which_arg] = _gvn.transform(kls);
3311 }
3312
3313 // Having loaded both klasses, test each for null.
3314 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3315 for (which_arg = 0; which_arg <= 1; which_arg++) {
3316 Node* kls = klasses[which_arg];
3317 Node* null_ctl = top();
3318 kls = null_check_oop(kls, &null_ctl, never_see_null);
3319 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3320 region->init_req(prim_path, null_ctl);
3321 if (stopped()) break;
3322 klasses[which_arg] = kls;
3323 }
3324
3325 if (!stopped()) {
3326 // now we have two reference types, in klasses[0..1]
3327 Node* subk = klasses[1]; // the argument to isAssignableFrom
3328 Node* superk = klasses[0]; // the receiver
3329 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3330 // now we have a successful reference subtype check
3331 region->set_req(_ref_subtype_path, control());
3332 }
3333
3334 // If both operands are primitive (both klasses null), then
3335 // we must return true when they are identical primitives.
3336 // It is convenient to test this after the first null klass check.
3337 set_control(region->in(_prim_0_path)); // go back to first null check
3338 if (!stopped()) {
3339 // Since superc is primitive, make a guard for the superc==subc case.
3340 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3341 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3342 generate_guard(bol_eq, region, PROB_FAIR);
3343 if (region->req() == PATH_LIMIT+1) {
3344 // A guard was added. If the added guard is taken, superc==subc.
3345 region->swap_edges(PATH_LIMIT, _prim_same_path);
3346 region->del_req(PATH_LIMIT);
3347 }
3348 region->set_req(_prim_0_path, control()); // Not equal after all.
3349 }
3350
3351 // these are the only paths that produce 'true':
3352 phi->set_req(_prim_same_path, intcon(1));
3353 phi->set_req(_ref_subtype_path, intcon(1));
3354
3355 // pull together the cases:
3356 assert(region->req() == PATH_LIMIT, "sane region");
3357 for (uint i = 1; i < region->req(); i++) {
3358 Node* ctl = region->in(i);
3359 if (ctl == NULL || ctl == top()) {
3360 region->set_req(i, top());
3361 phi ->set_req(i, top());
3362 } else if (phi->in(i) == NULL) {
3363 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3364 }
3365 }
3366
3367 set_control(_gvn.transform(region));
3368 set_result(_gvn.transform(phi));
3369 return true;
3370 }
3371
3372 //---------------------generate_array_guard_common------------------------
3373 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3374 bool obj_array, bool not_array) {
3375
3376 if (stopped()) {
3377 return NULL;
3378 }
3379
3380 // If obj_array/non_array==false/false:
3381 // Branch around if the given klass is in fact an array (either obj or prim).
3382 // If obj_array/non_array==false/true:
3383 // Branch around if the given klass is not an array klass of any kind.
3384 // If obj_array/non_array==true/true:
3385 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3386 // If obj_array/non_array==true/false:
3387 // Branch around if the kls is an oop array (Object[] or subtype)
3388 //
3389 // Like generate_guard, adds a new path onto the region.
3390 jint layout_con = 0;
3391 Node* layout_val = get_layout_helper(kls, layout_con);
3392 if (layout_val == NULL) {
3393 bool query = (obj_array
3394 ? Klass::layout_helper_is_objArray(layout_con)
3395 : Klass::layout_helper_is_array(layout_con));
3396 if (query == not_array) {
3397 return NULL; // never a branch
3398 } else { // always a branch
3399 Node* always_branch = control();
3400 if (region != NULL)
3401 region->add_req(always_branch);
3402 set_control(top());
3403 return always_branch;
3404 }
3405 }
3406 // Now test the correct condition.
3407 jint nval = (obj_array
3408 ? (jint)(Klass::_lh_array_tag_type_value
3409 << Klass::_lh_array_tag_shift)
3410 : Klass::_lh_neutral_value);
3411 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3412 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3413 // invert the test if we are looking for a non-array
3414 if (not_array) btest = BoolTest(btest).negate();
3415 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3416 return generate_fair_guard(bol, region);
3417 }
3418
3419
3420 //-----------------------inline_native_newArray--------------------------
3421 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3422 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3423 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3424 Node* mirror;
3425 Node* count_val;
3426 if (uninitialized) {
3427 mirror = argument(1);
3428 count_val = argument(2);
3429 } else {
3430 mirror = argument(0);
3431 count_val = argument(1);
3432 }
3433
3434 mirror = null_check(mirror);
3435 // If mirror or obj is dead, only null-path is taken.
3436 if (stopped()) return true;
3437
3438 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3439 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3440 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3441 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3442 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3443
3444 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3445 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3446 result_reg, _slow_path);
3447 Node* normal_ctl = control();
3448 Node* no_array_ctl = result_reg->in(_slow_path);
3449
3450 // Generate code for the slow case. We make a call to newArray().
3451 set_control(no_array_ctl);
3452 if (!stopped()) {
3453 // Either the input type is void.class, or else the
3454 // array klass has not yet been cached. Either the
3455 // ensuing call will throw an exception, or else it
3456 // will cache the array klass for next time.
3457 PreserveJVMState pjvms(this);
3458 CallJavaNode* slow_call = NULL;
3459 if (uninitialized) {
3460 // Generate optimized virtual call (holder class 'Unsafe' is final)
3461 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3462 } else {
3463 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3464 }
3465 Node* slow_result = set_results_for_java_call(slow_call);
3466 // this->control() comes from set_results_for_java_call
3467 result_reg->set_req(_slow_path, control());
3468 result_val->set_req(_slow_path, slow_result);
3469 result_io ->set_req(_slow_path, i_o());
3470 result_mem->set_req(_slow_path, reset_memory());
3471 }
3472
3473 set_control(normal_ctl);
3474 if (!stopped()) {
3475 // Normal case: The array type has been cached in the java.lang.Class.
3476 // The following call works fine even if the array type is polymorphic.
3477 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3478 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3479 result_reg->init_req(_normal_path, control());
3480 result_val->init_req(_normal_path, obj);
3481 result_io ->init_req(_normal_path, i_o());
3482 result_mem->init_req(_normal_path, reset_memory());
3483
3484 if (uninitialized) {
3485 // Mark the allocation so that zeroing is skipped
3486 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3487 alloc->maybe_set_complete(&_gvn);
3488 }
3489 }
3490
3491 // Return the combined state.
3492 set_i_o( _gvn.transform(result_io) );
3493 set_all_memory( _gvn.transform(result_mem));
3494
3495 C->set_has_split_ifs(true); // Has chance for split-if optimization
3496 set_result(result_reg, result_val);
3497 return true;
3498 }
3499
3500 //----------------------inline_native_getLength--------------------------
3501 // public static native int java.lang.reflect.Array.getLength(Object array);
3502 bool LibraryCallKit::inline_native_getLength() {
3503 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3504
3505 Node* array = null_check(argument(0));
3506 // If array is dead, only null-path is taken.
3507 if (stopped()) return true;
3508
3509 // Deoptimize if it is a non-array.
3510 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3511
3512 if (non_array != NULL) {
3513 PreserveJVMState pjvms(this);
3514 set_control(non_array);
3515 uncommon_trap(Deoptimization::Reason_intrinsic,
3516 Deoptimization::Action_maybe_recompile);
3517 }
3518
3519 // If control is dead, only non-array-path is taken.
3520 if (stopped()) return true;
3521
3522 // The works fine even if the array type is polymorphic.
3523 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3524 Node* result = load_array_length(array);
3525
3526 C->set_has_split_ifs(true); // Has chance for split-if optimization
3527 set_result(result);
3528 return true;
3529 }
3530
3531 //------------------------inline_array_copyOf----------------------------
3532 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3533 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3534 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3535 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3536
3537 // Get the arguments.
3538 Node* original = argument(0);
3539 Node* start = is_copyOfRange? argument(1): intcon(0);
3540 Node* end = is_copyOfRange? argument(2): argument(1);
3541 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3542
3543 Node* newcopy = NULL;
3544
3545 // Set the original stack and the reexecute bit for the interpreter to reexecute
3546 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3547 { PreserveReexecuteState preexecs(this);
3548 jvms()->set_should_reexecute(true);
3549
3550 array_type_mirror = null_check(array_type_mirror);
3551 original = null_check(original);
3552
3553 // Check if a null path was taken unconditionally.
3554 if (stopped()) return true;
3555
3556 Node* orig_length = load_array_length(original);
3557
3558 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3559 klass_node = null_check(klass_node);
3560
3561 RegionNode* bailout = new RegionNode(1);
3562 record_for_igvn(bailout);
3563
3564 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3565 // Bail out if that is so.
3566 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3567 if (not_objArray != NULL) {
3568 // Improve the klass node's type from the new optimistic assumption:
3569 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3570 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3571 Node* cast = new CastPPNode(klass_node, akls);
3572 cast->init_req(0, control());
3573 klass_node = _gvn.transform(cast);
3574 }
3575
3576 // Bail out if either start or end is negative.
3577 generate_negative_guard(start, bailout, &start);
3578 generate_negative_guard(end, bailout, &end);
3579
3580 Node* length = end;
3581 if (_gvn.type(start) != TypeInt::ZERO) {
3582 length = _gvn.transform(new SubINode(end, start));
3583 }
3584
3585 // Bail out if length is negative.
3586 // Without this the new_array would throw
3587 // NegativeArraySizeException but IllegalArgumentException is what
3588 // should be thrown
3589 generate_negative_guard(length, bailout, &length);
3590
3591 if (bailout->req() > 1) {
3592 PreserveJVMState pjvms(this);
3593 set_control(_gvn.transform(bailout));
3594 uncommon_trap(Deoptimization::Reason_intrinsic,
3595 Deoptimization::Action_maybe_recompile);
3596 }
3597
3598 if (!stopped()) {
3599 // How many elements will we copy from the original?
3600 // The answer is MinI(orig_length - start, length).
3601 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3602 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3603
3604 // Generate a direct call to the right arraycopy function(s).
3605 // We know the copy is disjoint but we might not know if the
3606 // oop stores need checking.
3607 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3608 // This will fail a store-check if x contains any non-nulls.
3609
3610 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3611 // loads/stores but it is legal only if we're sure the
3612 // Arrays.copyOf would succeed. So we need all input arguments
3613 // to the copyOf to be validated, including that the copy to the
3614 // new array won't trigger an ArrayStoreException. That subtype
3615 // check can be optimized if we know something on the type of
3616 // the input array from type speculation.
3617 if (_gvn.type(klass_node)->singleton()) {
3618 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3619 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3620
3621 int test = C->static_subtype_check(superk, subk);
3622 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3623 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3624 if (t_original->speculative_type() != NULL) {
3625 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3626 }
3627 }
3628 }
3629
3630 bool validated = false;
3631 // Reason_class_check rather than Reason_intrinsic because we
3632 // want to intrinsify even if this traps.
3633 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3634 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3635
3636 if (not_subtype_ctrl != top()) {
3637 PreserveJVMState pjvms(this);
3638 set_control(not_subtype_ctrl);
3639 uncommon_trap(Deoptimization::Reason_class_check,
3640 Deoptimization::Action_make_not_entrant);
3641 assert(stopped(), "Should be stopped");
3642 }
3643 validated = true;
3644 }
3645
3646 if (!stopped()) {
3647 newcopy = new_array(klass_node, length, 0); // no arguments to push
3648
3649 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3650 load_object_klass(original), klass_node);
3651 if (!is_copyOfRange) {
3652 ac->set_copyof(validated);
3653 } else {
3654 ac->set_copyofrange(validated);
3655 }
3656 Node* n = _gvn.transform(ac);
3657 if (n == ac) {
3658 ac->connect_outputs(this);
3659 } else {
3660 assert(validated, "shouldn't transform if all arguments not validated");
3661 set_all_memory(n);
3662 }
3663 }
3664 }
3665 } // original reexecute is set back here
3666
3667 C->set_has_split_ifs(true); // Has chance for split-if optimization
3668 if (!stopped()) {
3669 set_result(newcopy);
3670 }
3671 return true;
3672 }
3673
3674
3675 //----------------------generate_virtual_guard---------------------------
3676 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3677 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3678 RegionNode* slow_region) {
3679 ciMethod* method = callee();
3680 int vtable_index = method->vtable_index();
3681 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3682 "bad index %d", vtable_index);
3683 // Get the Method* out of the appropriate vtable entry.
3684 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3685 vtable_index*vtableEntry::size_in_bytes() +
3686 vtableEntry::method_offset_in_bytes();
3687 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3688 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3689
3690 // Compare the target method with the expected method (e.g., Object.hashCode).
3691 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3692
3693 Node* native_call = makecon(native_call_addr);
3694 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3695 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3696
3697 return generate_slow_guard(test_native, slow_region);
3698 }
3699
3700 //-----------------------generate_method_call----------------------------
3701 // Use generate_method_call to make a slow-call to the real
3702 // method if the fast path fails. An alternative would be to
3703 // use a stub like OptoRuntime::slow_arraycopy_Java.
3704 // This only works for expanding the current library call,
3705 // not another intrinsic. (E.g., don't use this for making an
3706 // arraycopy call inside of the copyOf intrinsic.)
3707 CallJavaNode*
3708 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3709 // When compiling the intrinsic method itself, do not use this technique.
3710 guarantee(callee() != C->method(), "cannot make slow-call to self");
3711
3712 ciMethod* method = callee();
3713 // ensure the JVMS we have will be correct for this call
3714 guarantee(method_id == method->intrinsic_id(), "must match");
3715
3716 const TypeFunc* tf = TypeFunc::make(method);
3717 CallJavaNode* slow_call;
3718 if (is_static) {
3719 assert(!is_virtual, "");
3720 slow_call = new CallStaticJavaNode(C, tf,
3721 SharedRuntime::get_resolve_static_call_stub(), method);
3722 } else if (is_virtual) {
3723 null_check_receiver();
3724 int vtable_index = Method::invalid_vtable_index;
3725 if (UseInlineCaches) {
3726 // Suppress the vtable call
3727 } else {
3728 // hashCode and clone are not a miranda methods,
3729 // so the vtable index is fixed.
3730 // No need to use the linkResolver to get it.
3731 vtable_index = method->vtable_index();
3732 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3733 "bad index %d", vtable_index);
3734 }
3735 slow_call = new CallDynamicJavaNode(tf,
3736 SharedRuntime::get_resolve_virtual_call_stub(),
3737 method, vtable_index);
3738 } else { // neither virtual nor static: opt_virtual
3739 null_check_receiver();
3740 slow_call = new CallStaticJavaNode(C, tf,
3741 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
3742 slow_call->set_optimized_virtual(true);
3743 }
3744 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3745 // To be able to issue a direct call (optimized virtual or virtual)
3746 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3747 // about the method being invoked should be attached to the call site to
3748 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3749 slow_call->set_override_symbolic_info(true);
3750 }
3751 set_arguments_for_java_call(slow_call);
3752 set_edges_for_java_call(slow_call);
3753 return slow_call;
3754 }
3755
3756
3757 /**
3758 * Build special case code for calls to hashCode on an object. This call may
3759 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3760 * slightly different code.
3761 */
3762 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3763 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3764 assert(!(is_virtual && is_static), "either virtual, special, or static");
3765
3766 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3767
3768 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3769 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3770 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3771 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3772 Node* obj = NULL;
3773 if (!is_static) {
3774 // Check for hashing null object
3775 obj = null_check_receiver();
3776 if (stopped()) return true; // unconditionally null
3777 result_reg->init_req(_null_path, top());
3778 result_val->init_req(_null_path, top());
3779 } else {
3780 // Do a null check, and return zero if null.
3781 // System.identityHashCode(null) == 0
3782 obj = argument(0);
3783 Node* null_ctl = top();
3784 obj = null_check_oop(obj, &null_ctl);
3785 result_reg->init_req(_null_path, null_ctl);
3786 result_val->init_req(_null_path, _gvn.intcon(0));
3787 }
3788
3789 // Unconditionally null? Then return right away.
3790 if (stopped()) {
3791 set_control( result_reg->in(_null_path));
3792 if (!stopped())
3793 set_result(result_val->in(_null_path));
3794 return true;
3795 }
3796
3797 // We only go to the fast case code if we pass a number of guards. The
3798 // paths which do not pass are accumulated in the slow_region.
3799 RegionNode* slow_region = new RegionNode(1);
3800 record_for_igvn(slow_region);
3801
3802 // If this is a virtual call, we generate a funny guard. We pull out
3803 // the vtable entry corresponding to hashCode() from the target object.
3804 // If the target method which we are calling happens to be the native
3805 // Object hashCode() method, we pass the guard. We do not need this
3806 // guard for non-virtual calls -- the caller is known to be the native
3807 // Object hashCode().
3808 if (is_virtual) {
3809 // After null check, get the object's klass.
3810 Node* obj_klass = load_object_klass(obj);
3811 generate_virtual_guard(obj_klass, slow_region);
3812 }
3813
3814 // Get the header out of the object, use LoadMarkNode when available
3815 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3816 // The control of the load must be NULL. Otherwise, the load can move before
3817 // the null check after castPP removal.
3818 Node* no_ctrl = NULL;
3819 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3820
3821 // Test the header to see if it is unlocked.
3822 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
3823 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3824 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3825 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3826 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3827
3828 generate_slow_guard(test_unlocked, slow_region);
3829
3830 // Get the hash value and check to see that it has been properly assigned.
3831 // We depend on hash_mask being at most 32 bits and avoid the use of
3832 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3833 // vm: see markWord.hpp.
3834 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3835 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3836 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3837 // This hack lets the hash bits live anywhere in the mark object now, as long
3838 // as the shift drops the relevant bits into the low 32 bits. Note that
3839 // Java spec says that HashCode is an int so there's no point in capturing
3840 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3841 hshifted_header = ConvX2I(hshifted_header);
3842 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3843
3844 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3845 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3846 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3847
3848 generate_slow_guard(test_assigned, slow_region);
3849
3850 Node* init_mem = reset_memory();
3851 // fill in the rest of the null path:
3852 result_io ->init_req(_null_path, i_o());
3853 result_mem->init_req(_null_path, init_mem);
3854
3855 result_val->init_req(_fast_path, hash_val);
3856 result_reg->init_req(_fast_path, control());
3857 result_io ->init_req(_fast_path, i_o());
3858 result_mem->init_req(_fast_path, init_mem);
3859
3860 // Generate code for the slow case. We make a call to hashCode().
3861 set_control(_gvn.transform(slow_region));
3862 if (!stopped()) {
3863 // No need for PreserveJVMState, because we're using up the present state.
3864 set_all_memory(init_mem);
3865 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3866 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3867 Node* slow_result = set_results_for_java_call(slow_call);
3868 // this->control() comes from set_results_for_java_call
3869 result_reg->init_req(_slow_path, control());
3870 result_val->init_req(_slow_path, slow_result);
3871 result_io ->set_req(_slow_path, i_o());
3872 result_mem ->set_req(_slow_path, reset_memory());
3873 }
3874
3875 // Return the combined state.
3876 set_i_o( _gvn.transform(result_io) );
3877 set_all_memory( _gvn.transform(result_mem));
3878
3879 set_result(result_reg, result_val);
3880 return true;
3881 }
3882
3883 //---------------------------inline_native_getClass----------------------------
3884 // public final native Class<?> java.lang.Object.getClass();
3885 //
3886 // Build special case code for calls to getClass on an object.
3887 bool LibraryCallKit::inline_native_getClass() {
3888 Node* obj = null_check_receiver();
3889 if (stopped()) return true;
3890 set_result(load_mirror_from_klass(load_object_klass(obj)));
3891 return true;
3892 }
3893
3894 //-----------------inline_native_Reflection_getCallerClass---------------------
3895 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3896 //
3897 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3898 //
3899 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3900 // in that it must skip particular security frames and checks for
3901 // caller sensitive methods.
3902 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3903 #ifndef PRODUCT
3904 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3905 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3906 }
3907 #endif
3908
3909 if (!jvms()->has_method()) {
3910 #ifndef PRODUCT
3911 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3912 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
3913 }
3914 #endif
3915 return false;
3916 }
3917
3918 // Walk back up the JVM state to find the caller at the required
3919 // depth.
3920 JVMState* caller_jvms = jvms();
3921
3922 // Cf. JVM_GetCallerClass
3923 // NOTE: Start the loop at depth 1 because the current JVM state does
3924 // not include the Reflection.getCallerClass() frame.
3925 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
3926 ciMethod* m = caller_jvms->method();
3927 switch (n) {
3928 case 0:
3929 fatal("current JVM state does not include the Reflection.getCallerClass frame");
3930 break;
3931 case 1:
3932 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
3933 if (!m->caller_sensitive()) {
3934 #ifndef PRODUCT
3935 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3936 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
3937 }
3938 #endif
3939 return false; // bail-out; let JVM_GetCallerClass do the work
3940 }
3941 break;
3942 default:
3943 if (!m->is_ignored_by_security_stack_walk()) {
3944 // We have reached the desired frame; return the holder class.
3945 // Acquire method holder as java.lang.Class and push as constant.
3946 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
3947 ciInstance* caller_mirror = caller_klass->java_mirror();
3948 set_result(makecon(TypeInstPtr::make(caller_mirror)));
3949
3950 #ifndef PRODUCT
3951 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3952 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());
3953 tty->print_cr(" JVM state at this point:");
3954 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
3955 ciMethod* m = jvms()->of_depth(i)->method();
3956 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
3957 }
3958 }
3959 #endif
3960 return true;
3961 }
3962 break;
3963 }
3964 }
3965
3966 #ifndef PRODUCT
3967 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3968 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
3969 tty->print_cr(" JVM state at this point:");
3970 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
3971 ciMethod* m = jvms()->of_depth(i)->method();
3972 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
3973 }
3974 }
3975 #endif
3976
3977 return false; // bail-out; let JVM_GetCallerClass do the work
3978 }
3979
3980 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
3981 Node* arg = argument(0);
3982 Node* result = NULL;
3983
3984 switch (id) {
3985 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
3986 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
3987 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
3988 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
3989
3990 case vmIntrinsics::_doubleToLongBits: {
3991 // two paths (plus control) merge in a wood
3992 RegionNode *r = new RegionNode(3);
3993 Node *phi = new PhiNode(r, TypeLong::LONG);
3994
3995 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
3996 // Build the boolean node
3997 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
3998
3999 // Branch either way.
4000 // NaN case is less traveled, which makes all the difference.
4001 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4002 Node *opt_isnan = _gvn.transform(ifisnan);
4003 assert( opt_isnan->is_If(), "Expect an IfNode");
4004 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4005 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4006
4007 set_control(iftrue);
4008
4009 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4010 Node *slow_result = longcon(nan_bits); // return NaN
4011 phi->init_req(1, _gvn.transform( slow_result ));
4012 r->init_req(1, iftrue);
4013
4014 // Else fall through
4015 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4016 set_control(iffalse);
4017
4018 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4019 r->init_req(2, iffalse);
4020
4021 // Post merge
4022 set_control(_gvn.transform(r));
4023 record_for_igvn(r);
4024
4025 C->set_has_split_ifs(true); // Has chance for split-if optimization
4026 result = phi;
4027 assert(result->bottom_type()->isa_long(), "must be");
4028 break;
4029 }
4030
4031 case vmIntrinsics::_floatToIntBits: {
4032 // two paths (plus control) merge in a wood
4033 RegionNode *r = new RegionNode(3);
4034 Node *phi = new PhiNode(r, TypeInt::INT);
4035
4036 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4037 // Build the boolean node
4038 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4039
4040 // Branch either way.
4041 // NaN case is less traveled, which makes all the difference.
4042 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4043 Node *opt_isnan = _gvn.transform(ifisnan);
4044 assert( opt_isnan->is_If(), "Expect an IfNode");
4045 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4046 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4047
4048 set_control(iftrue);
4049
4050 static const jint nan_bits = 0x7fc00000;
4051 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4052 phi->init_req(1, _gvn.transform( slow_result ));
4053 r->init_req(1, iftrue);
4054
4055 // Else fall through
4056 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4057 set_control(iffalse);
4058
4059 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4060 r->init_req(2, iffalse);
4061
4062 // Post merge
4063 set_control(_gvn.transform(r));
4064 record_for_igvn(r);
4065
4066 C->set_has_split_ifs(true); // Has chance for split-if optimization
4067 result = phi;
4068 assert(result->bottom_type()->isa_int(), "must be");
4069 break;
4070 }
4071
4072 default:
4073 fatal_unexpected_iid(id);
4074 break;
4075 }
4076 set_result(_gvn.transform(result));
4077 return true;
4078 }
4079
4080 //----------------------inline_unsafe_copyMemory-------------------------
4081 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4082
4083 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4084 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4085 const Type* base_t = gvn.type(base);
4086
4087 bool in_native = (base_t == TypePtr::NULL_PTR);
4088 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4089 bool is_mixed = !in_heap && !in_native;
4090
4091 if (is_mixed) {
4092 return true; // mixed accesses can touch both on-heap and off-heap memory
4093 }
4094 if (in_heap) {
4095 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4096 if (!is_prim_array) {
4097 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4098 // there's not enough type information available to determine proper memory slice for it.
4099 return true;
4100 }
4101 }
4102 return false;
4103 }
4104
4105 bool LibraryCallKit::inline_unsafe_copyMemory() {
4106 if (callee()->is_static()) return false; // caller must have the capability!
4107 null_check_receiver(); // null-check receiver
4108 if (stopped()) return true;
4109
4110 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4111
4112 Node* src_base = argument(1); // type: oop
4113 Node* src_off = ConvL2X(argument(2)); // type: long
4114 Node* dst_base = argument(4); // type: oop
4115 Node* dst_off = ConvL2X(argument(5)); // type: long
4116 Node* size = ConvL2X(argument(7)); // type: long
4117
4118 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4119 "fieldOffset must be byte-scaled");
4120
4121 Node* src_addr = make_unsafe_address(src_base, src_off);
4122 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4123
4124 Node* thread = _gvn.transform(new ThreadLocalNode());
4125 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4126 BasicType doing_unsafe_access_bt = T_BYTE;
4127 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4128
4129 // update volatile field
4130 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4131
4132 int flags = RC_LEAF | RC_NO_FP;
4133
4134 const TypePtr* dst_type = TypePtr::BOTTOM;
4135
4136 // Adjust memory effects of the runtime call based on input values.
4137 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4138 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4139 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4140
4141 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4142 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4143 flags |= RC_NARROW_MEM; // narrow in memory
4144 }
4145 }
4146
4147 // Call it. Note that the length argument is not scaled.
4148 make_runtime_call(flags,
4149 OptoRuntime::fast_arraycopy_Type(),
4150 StubRoutines::unsafe_arraycopy(),
4151 "unsafe_arraycopy",
4152 dst_type,
4153 src_addr, dst_addr, size XTOP);
4154
4155 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4156
4157 return true;
4158 }
4159
4160 #undef XTOP
4161
4162 //------------------------clone_coping-----------------------------------
4163 // Helper function for inline_native_clone.
4164 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4165 assert(obj_size != NULL, "");
4166 Node* raw_obj = alloc_obj->in(1);
4167 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4168
4169 AllocateNode* alloc = NULL;
4170 if (ReduceBulkZeroing) {
4171 // We will be completely responsible for initializing this object -
4172 // mark Initialize node as complete.
4173 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4174 // The object was just allocated - there should be no any stores!
4175 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4176 // Mark as complete_with_arraycopy so that on AllocateNode
4177 // expansion, we know this AllocateNode is initialized by an array
4178 // copy and a StoreStore barrier exists after the array copy.
4179 alloc->initialization()->set_complete_with_arraycopy();
4180 }
4181
4182 Node* size = _gvn.transform(obj_size);
4183 access_clone(obj, alloc_obj, size, is_array);
4184
4185 // Do not let reads from the cloned object float above the arraycopy.
4186 if (alloc != NULL) {
4187 // Do not let stores that initialize this object be reordered with
4188 // a subsequent store that would make this object accessible by
4189 // other threads.
4190 // Record what AllocateNode this StoreStore protects so that
4191 // escape analysis can go from the MemBarStoreStoreNode to the
4192 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4193 // based on the escape status of the AllocateNode.
4194 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4195 } else {
4196 insert_mem_bar(Op_MemBarCPUOrder);
4197 }
4198 }
4199
4200 //------------------------inline_native_clone----------------------------
4201 // protected native Object java.lang.Object.clone();
4202 //
4203 // Here are the simple edge cases:
4204 // null receiver => normal trap
4205 // virtual and clone was overridden => slow path to out-of-line clone
4206 // not cloneable or finalizer => slow path to out-of-line Object.clone
4207 //
4208 // The general case has two steps, allocation and copying.
4209 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4210 //
4211 // Copying also has two cases, oop arrays and everything else.
4212 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4213 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4214 //
4215 // These steps fold up nicely if and when the cloned object's klass
4216 // can be sharply typed as an object array, a type array, or an instance.
4217 //
4218 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4219 PhiNode* result_val;
4220
4221 // Set the reexecute bit for the interpreter to reexecute
4222 // the bytecode that invokes Object.clone if deoptimization happens.
4223 { PreserveReexecuteState preexecs(this);
4224 jvms()->set_should_reexecute(true);
4225
4226 Node* obj = null_check_receiver();
4227 if (stopped()) return true;
4228
4229 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4230
4231 // If we are going to clone an instance, we need its exact type to
4232 // know the number and types of fields to convert the clone to
4233 // loads/stores. Maybe a speculative type can help us.
4234 if (!obj_type->klass_is_exact() &&
4235 obj_type->speculative_type() != NULL &&
4236 obj_type->speculative_type()->is_instance_klass()) {
4237 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4238 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4239 !spec_ik->has_injected_fields()) {
4240 ciKlass* k = obj_type->klass();
4241 if (!k->is_instance_klass() ||
4242 k->as_instance_klass()->is_interface() ||
4243 k->as_instance_klass()->has_subklass()) {
4244 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4245 }
4246 }
4247 }
4248
4249 // Conservatively insert a memory barrier on all memory slices.
4250 // Do not let writes into the original float below the clone.
4251 insert_mem_bar(Op_MemBarCPUOrder);
4252
4253 // paths into result_reg:
4254 enum {
4255 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4256 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4257 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4258 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4259 PATH_LIMIT
4260 };
4261 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4262 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4263 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4264 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4265 record_for_igvn(result_reg);
4266
4267 Node* obj_klass = load_object_klass(obj);
4268 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4269 if (array_ctl != NULL) {
4270 // It's an array.
4271 PreserveJVMState pjvms(this);
4272 set_control(array_ctl);
4273 Node* obj_length = load_array_length(obj);
4274 Node* obj_size = NULL;
4275 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4276
4277 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4278 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4279 // If it is an oop array, it requires very special treatment,
4280 // because gc barriers are required when accessing the array.
4281 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4282 if (is_obja != NULL) {
4283 PreserveJVMState pjvms2(this);
4284 set_control(is_obja);
4285 // Generate a direct call to the right arraycopy function(s).
4286 // Clones are always tightly coupled.
4287 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4288 ac->set_clone_oop_array();
4289 Node* n = _gvn.transform(ac);
4290 assert(n == ac, "cannot disappear");
4291 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4292
4293 result_reg->init_req(_objArray_path, control());
4294 result_val->init_req(_objArray_path, alloc_obj);
4295 result_i_o ->set_req(_objArray_path, i_o());
4296 result_mem ->set_req(_objArray_path, reset_memory());
4297 }
4298 }
4299 // Otherwise, there are no barriers to worry about.
4300 // (We can dispense with card marks if we know the allocation
4301 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4302 // causes the non-eden paths to take compensating steps to
4303 // simulate a fresh allocation, so that no further
4304 // card marks are required in compiled code to initialize
4305 // the object.)
4306
4307 if (!stopped()) {
4308 copy_to_clone(obj, alloc_obj, obj_size, true);
4309
4310 // Present the results of the copy.
4311 result_reg->init_req(_array_path, control());
4312 result_val->init_req(_array_path, alloc_obj);
4313 result_i_o ->set_req(_array_path, i_o());
4314 result_mem ->set_req(_array_path, reset_memory());
4315 }
4316 }
4317
4318 // We only go to the instance fast case code if we pass a number of guards.
4319 // The paths which do not pass are accumulated in the slow_region.
4320 RegionNode* slow_region = new RegionNode(1);
4321 record_for_igvn(slow_region);
4322 if (!stopped()) {
4323 // It's an instance (we did array above). Make the slow-path tests.
4324 // If this is a virtual call, we generate a funny guard. We grab
4325 // the vtable entry corresponding to clone() from the target object.
4326 // If the target method which we are calling happens to be the
4327 // Object clone() method, we pass the guard. We do not need this
4328 // guard for non-virtual calls; the caller is known to be the native
4329 // Object clone().
4330 if (is_virtual) {
4331 generate_virtual_guard(obj_klass, slow_region);
4332 }
4333
4334 // The object must be easily cloneable and must not have a finalizer.
4335 // Both of these conditions may be checked in a single test.
4336 // We could optimize the test further, but we don't care.
4337 generate_access_flags_guard(obj_klass,
4338 // Test both conditions:
4339 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4340 // Must be cloneable but not finalizer:
4341 JVM_ACC_IS_CLONEABLE_FAST,
4342 slow_region);
4343 }
4344
4345 if (!stopped()) {
4346 // It's an instance, and it passed the slow-path tests.
4347 PreserveJVMState pjvms(this);
4348 Node* obj_size = NULL;
4349 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4350 // is reexecuted if deoptimization occurs and there could be problems when merging
4351 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4352 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4353
4354 copy_to_clone(obj, alloc_obj, obj_size, false);
4355
4356 // Present the results of the slow call.
4357 result_reg->init_req(_instance_path, control());
4358 result_val->init_req(_instance_path, alloc_obj);
4359 result_i_o ->set_req(_instance_path, i_o());
4360 result_mem ->set_req(_instance_path, reset_memory());
4361 }
4362
4363 // Generate code for the slow case. We make a call to clone().
4364 set_control(_gvn.transform(slow_region));
4365 if (!stopped()) {
4366 PreserveJVMState pjvms(this);
4367 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4368 // We need to deoptimize on exception (see comment above)
4369 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4370 // this->control() comes from set_results_for_java_call
4371 result_reg->init_req(_slow_path, control());
4372 result_val->init_req(_slow_path, slow_result);
4373 result_i_o ->set_req(_slow_path, i_o());
4374 result_mem ->set_req(_slow_path, reset_memory());
4375 }
4376
4377 // Return the combined state.
4378 set_control( _gvn.transform(result_reg));
4379 set_i_o( _gvn.transform(result_i_o));
4380 set_all_memory( _gvn.transform(result_mem));
4381 } // original reexecute is set back here
4382
4383 set_result(_gvn.transform(result_val));
4384 return true;
4385 }
4386
4387 // If we have a tightly coupled allocation, the arraycopy may take care
4388 // of the array initialization. If one of the guards we insert between
4389 // the allocation and the arraycopy causes a deoptimization, an
4390 // unitialized array will escape the compiled method. To prevent that
4391 // we set the JVM state for uncommon traps between the allocation and
4392 // the arraycopy to the state before the allocation so, in case of
4393 // deoptimization, we'll reexecute the allocation and the
4394 // initialization.
4395 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4396 if (alloc != NULL) {
4397 ciMethod* trap_method = alloc->jvms()->method();
4398 int trap_bci = alloc->jvms()->bci();
4399
4400 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4401 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4402 // Make sure there's no store between the allocation and the
4403 // arraycopy otherwise visible side effects could be rexecuted
4404 // in case of deoptimization and cause incorrect execution.
4405 bool no_interfering_store = true;
4406 Node* mem = alloc->in(TypeFunc::Memory);
4407 if (mem->is_MergeMem()) {
4408 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4409 Node* n = mms.memory();
4410 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4411 assert(n->is_Store(), "what else?");
4412 no_interfering_store = false;
4413 break;
4414 }
4415 }
4416 } else {
4417 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4418 Node* n = mms.memory();
4419 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4420 assert(n->is_Store(), "what else?");
4421 no_interfering_store = false;
4422 break;
4423 }
4424 }
4425 }
4426
4427 if (no_interfering_store) {
4428 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4429 uint size = alloc->req();
4430 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4431 old_jvms->set_map(sfpt);
4432 for (uint i = 0; i < size; i++) {
4433 sfpt->init_req(i, alloc->in(i));
4434 }
4435 // re-push array length for deoptimization
4436 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4437 old_jvms->set_sp(old_jvms->sp()+1);
4438 old_jvms->set_monoff(old_jvms->monoff()+1);
4439 old_jvms->set_scloff(old_jvms->scloff()+1);
4440 old_jvms->set_endoff(old_jvms->endoff()+1);
4441 old_jvms->set_should_reexecute(true);
4442
4443 sfpt->set_i_o(map()->i_o());
4444 sfpt->set_memory(map()->memory());
4445 sfpt->set_control(map()->control());
4446
4447 JVMState* saved_jvms = jvms();
4448 saved_reexecute_sp = _reexecute_sp;
4449
4450 set_jvms(sfpt->jvms());
4451 _reexecute_sp = jvms()->sp();
4452
4453 return saved_jvms;
4454 }
4455 }
4456 }
4457 return NULL;
4458 }
4459
4460 // In case of a deoptimization, we restart execution at the
4461 // allocation, allocating a new array. We would leave an uninitialized
4462 // array in the heap that GCs wouldn't expect. Move the allocation
4463 // after the traps so we don't allocate the array if we
4464 // deoptimize. This is possible because tightly_coupled_allocation()
4465 // guarantees there's no observer of the allocated array at this point
4466 // and the control flow is simple enough.
4467 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4468 int saved_reexecute_sp, uint new_idx) {
4469 if (saved_jvms != NULL && !stopped()) {
4470 assert(alloc != NULL, "only with a tightly coupled allocation");
4471 // restore JVM state to the state at the arraycopy
4472 saved_jvms->map()->set_control(map()->control());
4473 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4474 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4475 // If we've improved the types of some nodes (null check) while
4476 // emitting the guards, propagate them to the current state
4477 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4478 set_jvms(saved_jvms);
4479 _reexecute_sp = saved_reexecute_sp;
4480
4481 // Remove the allocation from above the guards
4482 CallProjections callprojs;
4483 alloc->extract_projections(&callprojs, true);
4484 InitializeNode* init = alloc->initialization();
4485 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4486 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4487 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4488
4489 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4490 // the allocation (i.e. is only valid if the allocation succeeds):
4491 // 1) replace CastIINode with AllocateArrayNode's length here
4492 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4493 //
4494 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4495 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4496 Node* init_control = init->proj_out(TypeFunc::Control);
4497 Node* alloc_length = alloc->Ideal_length();
4498 #ifdef ASSERT
4499 Node* prev_cast = NULL;
4500 #endif
4501 for (uint i = 0; i < init_control->outcnt(); i++) {
4502 Node* init_out = init_control->raw_out(i);
4503 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4504 #ifdef ASSERT
4505 if (prev_cast == NULL) {
4506 prev_cast = init_out;
4507 } else {
4508 if (prev_cast->cmp(*init_out) == false) {
4509 prev_cast->dump();
4510 init_out->dump();
4511 assert(false, "not equal CastIINode");
4512 }
4513 }
4514 #endif
4515 C->gvn_replace_by(init_out, alloc_length);
4516 }
4517 }
4518 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4519
4520 // move the allocation here (after the guards)
4521 _gvn.hash_delete(alloc);
4522 alloc->set_req(TypeFunc::Control, control());
4523 alloc->set_req(TypeFunc::I_O, i_o());
4524 Node *mem = reset_memory();
4525 set_all_memory(mem);
4526 alloc->set_req(TypeFunc::Memory, mem);
4527 set_control(init->proj_out_or_null(TypeFunc::Control));
4528 set_i_o(callprojs.fallthrough_ioproj);
4529
4530 // Update memory as done in GraphKit::set_output_for_allocation()
4531 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4532 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4533 if (ary_type->isa_aryptr() && length_type != NULL) {
4534 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4535 }
4536 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4537 int elemidx = C->get_alias_index(telemref);
4538 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4539 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4540
4541 Node* allocx = _gvn.transform(alloc);
4542 assert(allocx == alloc, "where has the allocation gone?");
4543 assert(dest->is_CheckCastPP(), "not an allocation result?");
4544
4545 _gvn.hash_delete(dest);
4546 dest->set_req(0, control());
4547 Node* destx = _gvn.transform(dest);
4548 assert(destx == dest, "where has the allocation result gone?");
4549
4550 array_ideal_length(alloc, ary_type, true);
4551 }
4552 }
4553
4554
4555 //------------------------------inline_arraycopy-----------------------
4556 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4557 // Object dest, int destPos,
4558 // int length);
4559 bool LibraryCallKit::inline_arraycopy() {
4560 // Get the arguments.
4561 Node* src = argument(0); // type: oop
4562 Node* src_offset = argument(1); // type: int
4563 Node* dest = argument(2); // type: oop
4564 Node* dest_offset = argument(3); // type: int
4565 Node* length = argument(4); // type: int
4566
4567 uint new_idx = C->unique();
4568
4569 // Check for allocation before we add nodes that would confuse
4570 // tightly_coupled_allocation()
4571 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
4572
4573 int saved_reexecute_sp = -1;
4574 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4575 // See arraycopy_restore_alloc_state() comment
4576 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4577 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4578 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4579 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4580
4581 // The following tests must be performed
4582 // (1) src and dest are arrays.
4583 // (2) src and dest arrays must have elements of the same BasicType
4584 // (3) src and dest must not be null.
4585 // (4) src_offset must not be negative.
4586 // (5) dest_offset must not be negative.
4587 // (6) length must not be negative.
4588 // (7) src_offset + length must not exceed length of src.
4589 // (8) dest_offset + length must not exceed length of dest.
4590 // (9) each element of an oop array must be assignable
4591
4592 // (3) src and dest must not be null.
4593 // always do this here because we need the JVM state for uncommon traps
4594 Node* null_ctl = top();
4595 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4596 assert(null_ctl->is_top(), "no null control here");
4597 dest = null_check(dest, T_ARRAY);
4598
4599 if (!can_emit_guards) {
4600 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4601 // guards but the arraycopy node could still take advantage of a
4602 // tightly allocated allocation. tightly_coupled_allocation() is
4603 // called again to make sure it takes the null check above into
4604 // account: the null check is mandatory and if it caused an
4605 // uncommon trap to be emitted then the allocation can't be
4606 // considered tightly coupled in this context.
4607 alloc = tightly_coupled_allocation(dest);
4608 }
4609
4610 bool validated = false;
4611
4612 const Type* src_type = _gvn.type(src);
4613 const Type* dest_type = _gvn.type(dest);
4614 const TypeAryPtr* top_src = src_type->isa_aryptr();
4615 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4616
4617 // Do we have the type of src?
4618 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4619 // Do we have the type of dest?
4620 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4621 // Is the type for src from speculation?
4622 bool src_spec = false;
4623 // Is the type for dest from speculation?
4624 bool dest_spec = false;
4625
4626 if ((!has_src || !has_dest) && can_emit_guards) {
4627 // We don't have sufficient type information, let's see if
4628 // speculative types can help. We need to have types for both src
4629 // and dest so that it pays off.
4630
4631 // Do we already have or could we have type information for src
4632 bool could_have_src = has_src;
4633 // Do we already have or could we have type information for dest
4634 bool could_have_dest = has_dest;
4635
4636 ciKlass* src_k = NULL;
4637 if (!has_src) {
4638 src_k = src_type->speculative_type_not_null();
4639 if (src_k != NULL && src_k->is_array_klass()) {
4640 could_have_src = true;
4641 }
4642 }
4643
4644 ciKlass* dest_k = NULL;
4645 if (!has_dest) {
4646 dest_k = dest_type->speculative_type_not_null();
4647 if (dest_k != NULL && dest_k->is_array_klass()) {
4648 could_have_dest = true;
4649 }
4650 }
4651
4652 if (could_have_src && could_have_dest) {
4653 // This is going to pay off so emit the required guards
4654 if (!has_src) {
4655 src = maybe_cast_profiled_obj(src, src_k, true);
4656 src_type = _gvn.type(src);
4657 top_src = src_type->isa_aryptr();
4658 has_src = (top_src != NULL && top_src->klass() != NULL);
4659 src_spec = true;
4660 }
4661 if (!has_dest) {
4662 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4663 dest_type = _gvn.type(dest);
4664 top_dest = dest_type->isa_aryptr();
4665 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4666 dest_spec = true;
4667 }
4668 }
4669 }
4670
4671 if (has_src && has_dest && can_emit_guards) {
4672 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4673 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4674 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4675 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4676
4677 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4678 // If both arrays are object arrays then having the exact types
4679 // for both will remove the need for a subtype check at runtime
4680 // before the call and may make it possible to pick a faster copy
4681 // routine (without a subtype check on every element)
4682 // Do we have the exact type of src?
4683 bool could_have_src = src_spec;
4684 // Do we have the exact type of dest?
4685 bool could_have_dest = dest_spec;
4686 ciKlass* src_k = top_src->klass();
4687 ciKlass* dest_k = top_dest->klass();
4688 if (!src_spec) {
4689 src_k = src_type->speculative_type_not_null();
4690 if (src_k != NULL && src_k->is_array_klass()) {
4691 could_have_src = true;
4692 }
4693 }
4694 if (!dest_spec) {
4695 dest_k = dest_type->speculative_type_not_null();
4696 if (dest_k != NULL && dest_k->is_array_klass()) {
4697 could_have_dest = true;
4698 }
4699 }
4700 if (could_have_src && could_have_dest) {
4701 // If we can have both exact types, emit the missing guards
4702 if (could_have_src && !src_spec) {
4703 src = maybe_cast_profiled_obj(src, src_k, true);
4704 }
4705 if (could_have_dest && !dest_spec) {
4706 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4707 }
4708 }
4709 }
4710 }
4711
4712 ciMethod* trap_method = method();
4713 int trap_bci = bci();
4714 if (saved_jvms != NULL) {
4715 trap_method = alloc->jvms()->method();
4716 trap_bci = alloc->jvms()->bci();
4717 }
4718
4719 bool negative_length_guard_generated = false;
4720
4721 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4722 can_emit_guards &&
4723 !src->is_top() && !dest->is_top()) {
4724 // validate arguments: enables transformation the ArrayCopyNode
4725 validated = true;
4726
4727 RegionNode* slow_region = new RegionNode(1);
4728 record_for_igvn(slow_region);
4729
4730 // (1) src and dest are arrays.
4731 generate_non_array_guard(load_object_klass(src), slow_region);
4732 generate_non_array_guard(load_object_klass(dest), slow_region);
4733
4734 // (2) src and dest arrays must have elements of the same BasicType
4735 // done at macro expansion or at Ideal transformation time
4736
4737 // (4) src_offset must not be negative.
4738 generate_negative_guard(src_offset, slow_region);
4739
4740 // (5) dest_offset must not be negative.
4741 generate_negative_guard(dest_offset, slow_region);
4742
4743 // (7) src_offset + length must not exceed length of src.
4744 generate_limit_guard(src_offset, length,
4745 load_array_length(src),
4746 slow_region);
4747
4748 // (8) dest_offset + length must not exceed length of dest.
4749 generate_limit_guard(dest_offset, length,
4750 load_array_length(dest),
4751 slow_region);
4752
4753 // (6) length must not be negative.
4754 // This is also checked in generate_arraycopy() during macro expansion, but
4755 // we also have to check it here for the case where the ArrayCopyNode will
4756 // be eliminated by Escape Analysis.
4757 if (EliminateAllocations) {
4758 generate_negative_guard(length, slow_region);
4759 negative_length_guard_generated = true;
4760 }
4761
4762 // (9) each element of an oop array must be assignable
4763 Node* dest_klass = load_object_klass(dest);
4764 if (src != dest) {
4765 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4766
4767 if (not_subtype_ctrl != top()) {
4768 PreserveJVMState pjvms(this);
4769 set_control(not_subtype_ctrl);
4770 uncommon_trap(Deoptimization::Reason_intrinsic,
4771 Deoptimization::Action_make_not_entrant);
4772 assert(stopped(), "Should be stopped");
4773 }
4774 }
4775 {
4776 PreserveJVMState pjvms(this);
4777 set_control(_gvn.transform(slow_region));
4778 uncommon_trap(Deoptimization::Reason_intrinsic,
4779 Deoptimization::Action_make_not_entrant);
4780 assert(stopped(), "Should be stopped");
4781 }
4782
4783 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4784 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4785 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4786 }
4787
4788 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4789
4790 if (stopped()) {
4791 return true;
4792 }
4793
4794 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4795 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4796 // so the compiler has a chance to eliminate them: during macro expansion,
4797 // we have to set their control (CastPP nodes are eliminated).
4798 load_object_klass(src), load_object_klass(dest),
4799 load_array_length(src), load_array_length(dest));
4800
4801 ac->set_arraycopy(validated);
4802
4803 Node* n = _gvn.transform(ac);
4804 if (n == ac) {
4805 ac->connect_outputs(this);
4806 } else {
4807 assert(validated, "shouldn't transform if all arguments not validated");
4808 set_all_memory(n);
4809 }
4810 clear_upper_avx();
4811
4812
4813 return true;
4814 }
4815
4816
4817 // Helper function which determines if an arraycopy immediately follows
4818 // an allocation, with no intervening tests or other escapes for the object.
4819 AllocateArrayNode*
4820 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
4821 if (stopped()) return NULL; // no fast path
4822 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4823
4824 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4825 if (alloc == NULL) return NULL;
4826
4827 Node* rawmem = memory(Compile::AliasIdxRaw);
4828 // Is the allocation's memory state untouched?
4829 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4830 // Bail out if there have been raw-memory effects since the allocation.
4831 // (Example: There might have been a call or safepoint.)
4832 return NULL;
4833 }
4834 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4835 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4836 return NULL;
4837 }
4838
4839 // There must be no unexpected observers of this allocation.
4840 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4841 Node* obs = ptr->fast_out(i);
4842 if (obs != this->map()) {
4843 return NULL;
4844 }
4845 }
4846
4847 // This arraycopy must unconditionally follow the allocation of the ptr.
4848 Node* alloc_ctl = ptr->in(0);
4849 Node* ctl = control();
4850 while (ctl != alloc_ctl) {
4851 // There may be guards which feed into the slow_region.
4852 // Any other control flow means that we might not get a chance
4853 // to finish initializing the allocated object.
4854 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4855 IfNode* iff = ctl->in(0)->as_If();
4856 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4857 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4858 // One more try: Various low-level checks bottom out in
4859 // uncommon traps. If the debug-info of the trap omits
4860 // any reference to the allocation, as we've already
4861 // observed, then there can be no objection to the trap.
4862 bool found_trap = false;
4863 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4864 Node* obs = not_ctl->fast_out(j);
4865 if (obs->in(0) == not_ctl && obs->is_Call() &&
4866 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4867 found_trap = true; break;
4868 }
4869 }
4870 if (found_trap) {
4871 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4872 continue;
4873 }
4874 }
4875 return NULL;
4876 }
4877
4878 // If we get this far, we have an allocation which immediately
4879 // precedes the arraycopy, and we can take over zeroing the new object.
4880 // The arraycopy will finish the initialization, and provide
4881 // a new control state to which we will anchor the destination pointer.
4882
4883 return alloc;
4884 }
4885
4886 //-------------inline_encodeISOArray-----------------------------------
4887 // encode char[] to byte[] in ISO_8859_1 or ASCII
4888 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
4889 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4890 // no receiver since it is static method
4891 Node *src = argument(0);
4892 Node *src_offset = argument(1);
4893 Node *dst = argument(2);
4894 Node *dst_offset = argument(3);
4895 Node *length = argument(4);
4896
4897 src = must_be_not_null(src, true);
4898 dst = must_be_not_null(dst, true);
4899
4900 const Type* src_type = src->Value(&_gvn);
4901 const Type* dst_type = dst->Value(&_gvn);
4902 const TypeAryPtr* top_src = src_type->isa_aryptr();
4903 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4904 if (top_src == NULL || top_src->klass() == NULL ||
4905 top_dest == NULL || top_dest->klass() == NULL) {
4906 // failed array check
4907 return false;
4908 }
4909
4910 // Figure out the size and type of the elements we will be copying.
4911 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4912 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4913 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4914 return false;
4915 }
4916
4917 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4918 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4919 // 'src_start' points to src array + scaled offset
4920 // 'dst_start' points to dst array + scaled offset
4921
4922 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4923 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
4924 enc = _gvn.transform(enc);
4925 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4926 set_memory(res_mem, mtype);
4927 set_result(enc);
4928 clear_upper_avx();
4929
4930 return true;
4931 }
4932
4933 //-------------inline_multiplyToLen-----------------------------------
4934 bool LibraryCallKit::inline_multiplyToLen() {
4935 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4936
4937 address stubAddr = StubRoutines::multiplyToLen();
4938 if (stubAddr == NULL) {
4939 return false; // Intrinsic's stub is not implemented on this platform
4940 }
4941 const char* stubName = "multiplyToLen";
4942
4943 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4944
4945 // no receiver because it is a static method
4946 Node* x = argument(0);
4947 Node* xlen = argument(1);
4948 Node* y = argument(2);
4949 Node* ylen = argument(3);
4950 Node* z = argument(4);
4951
4952 x = must_be_not_null(x, true);
4953 y = must_be_not_null(y, true);
4954
4955 const Type* x_type = x->Value(&_gvn);
4956 const Type* y_type = y->Value(&_gvn);
4957 const TypeAryPtr* top_x = x_type->isa_aryptr();
4958 const TypeAryPtr* top_y = y_type->isa_aryptr();
4959 if (top_x == NULL || top_x->klass() == NULL ||
4960 top_y == NULL || top_y->klass() == NULL) {
4961 // failed array check
4962 return false;
4963 }
4964
4965 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4966 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4967 if (x_elem != T_INT || y_elem != T_INT) {
4968 return false;
4969 }
4970
4971 // Set the original stack and the reexecute bit for the interpreter to reexecute
4972 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
4973 // on the return from z array allocation in runtime.
4974 { PreserveReexecuteState preexecs(this);
4975 jvms()->set_should_reexecute(true);
4976
4977 Node* x_start = array_element_address(x, intcon(0), x_elem);
4978 Node* y_start = array_element_address(y, intcon(0), y_elem);
4979 // 'x_start' points to x array + scaled xlen
4980 // 'y_start' points to y array + scaled ylen
4981
4982 // Allocate the result array
4983 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
4984 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
4985 Node* klass_node = makecon(TypeKlassPtr::make(klass));
4986
4987 IdealKit ideal(this);
4988
4989 #define __ ideal.
4990 Node* one = __ ConI(1);
4991 Node* zero = __ ConI(0);
4992 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
4993 __ set(need_alloc, zero);
4994 __ set(z_alloc, z);
4995 __ if_then(z, BoolTest::eq, null()); {
4996 __ increment (need_alloc, one);
4997 } __ else_(); {
4998 // Update graphKit memory and control from IdealKit.
4999 sync_kit(ideal);
5000 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5001 cast->init_req(0, control());
5002 _gvn.set_type(cast, cast->bottom_type());
5003 C->record_for_igvn(cast);
5004
5005 Node* zlen_arg = load_array_length(cast);
5006 // Update IdealKit memory and control from graphKit.
5007 __ sync_kit(this);
5008 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5009 __ increment (need_alloc, one);
5010 } __ end_if();
5011 } __ end_if();
5012
5013 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5014 // Update graphKit memory and control from IdealKit.
5015 sync_kit(ideal);
5016 Node * narr = new_array(klass_node, zlen, 1);
5017 // Update IdealKit memory and control from graphKit.
5018 __ sync_kit(this);
5019 __ set(z_alloc, narr);
5020 } __ end_if();
5021
5022 sync_kit(ideal);
5023 z = __ value(z_alloc);
5024 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5025 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5026 // Final sync IdealKit and GraphKit.
5027 final_sync(ideal);
5028 #undef __
5029
5030 Node* z_start = array_element_address(z, intcon(0), T_INT);
5031
5032 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5033 OptoRuntime::multiplyToLen_Type(),
5034 stubAddr, stubName, TypePtr::BOTTOM,
5035 x_start, xlen, y_start, ylen, z_start, zlen);
5036 } // original reexecute is set back here
5037
5038 C->set_has_split_ifs(true); // Has chance for split-if optimization
5039 set_result(z);
5040 return true;
5041 }
5042
5043 //-------------inline_squareToLen------------------------------------
5044 bool LibraryCallKit::inline_squareToLen() {
5045 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5046
5047 address stubAddr = StubRoutines::squareToLen();
5048 if (stubAddr == NULL) {
5049 return false; // Intrinsic's stub is not implemented on this platform
5050 }
5051 const char* stubName = "squareToLen";
5052
5053 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5054
5055 Node* x = argument(0);
5056 Node* len = argument(1);
5057 Node* z = argument(2);
5058 Node* zlen = argument(3);
5059
5060 x = must_be_not_null(x, true);
5061 z = must_be_not_null(z, true);
5062
5063 const Type* x_type = x->Value(&_gvn);
5064 const Type* z_type = z->Value(&_gvn);
5065 const TypeAryPtr* top_x = x_type->isa_aryptr();
5066 const TypeAryPtr* top_z = z_type->isa_aryptr();
5067 if (top_x == NULL || top_x->klass() == NULL ||
5068 top_z == NULL || top_z->klass() == NULL) {
5069 // failed array check
5070 return false;
5071 }
5072
5073 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5074 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5075 if (x_elem != T_INT || z_elem != T_INT) {
5076 return false;
5077 }
5078
5079
5080 Node* x_start = array_element_address(x, intcon(0), x_elem);
5081 Node* z_start = array_element_address(z, intcon(0), z_elem);
5082
5083 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5084 OptoRuntime::squareToLen_Type(),
5085 stubAddr, stubName, TypePtr::BOTTOM,
5086 x_start, len, z_start, zlen);
5087
5088 set_result(z);
5089 return true;
5090 }
5091
5092 //-------------inline_mulAdd------------------------------------------
5093 bool LibraryCallKit::inline_mulAdd() {
5094 assert(UseMulAddIntrinsic, "not implemented on this platform");
5095
5096 address stubAddr = StubRoutines::mulAdd();
5097 if (stubAddr == NULL) {
5098 return false; // Intrinsic's stub is not implemented on this platform
5099 }
5100 const char* stubName = "mulAdd";
5101
5102 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5103
5104 Node* out = argument(0);
5105 Node* in = argument(1);
5106 Node* offset = argument(2);
5107 Node* len = argument(3);
5108 Node* k = argument(4);
5109
5110 out = must_be_not_null(out, true);
5111
5112 const Type* out_type = out->Value(&_gvn);
5113 const Type* in_type = in->Value(&_gvn);
5114 const TypeAryPtr* top_out = out_type->isa_aryptr();
5115 const TypeAryPtr* top_in = in_type->isa_aryptr();
5116 if (top_out == NULL || top_out->klass() == NULL ||
5117 top_in == NULL || top_in->klass() == NULL) {
5118 // failed array check
5119 return false;
5120 }
5121
5122 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5123 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5124 if (out_elem != T_INT || in_elem != T_INT) {
5125 return false;
5126 }
5127
5128 Node* outlen = load_array_length(out);
5129 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5130 Node* out_start = array_element_address(out, intcon(0), out_elem);
5131 Node* in_start = array_element_address(in, intcon(0), in_elem);
5132
5133 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5134 OptoRuntime::mulAdd_Type(),
5135 stubAddr, stubName, TypePtr::BOTTOM,
5136 out_start,in_start, new_offset, len, k);
5137 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5138 set_result(result);
5139 return true;
5140 }
5141
5142 //-------------inline_montgomeryMultiply-----------------------------------
5143 bool LibraryCallKit::inline_montgomeryMultiply() {
5144 address stubAddr = StubRoutines::montgomeryMultiply();
5145 if (stubAddr == NULL) {
5146 return false; // Intrinsic's stub is not implemented on this platform
5147 }
5148
5149 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5150 const char* stubName = "montgomery_multiply";
5151
5152 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5153
5154 Node* a = argument(0);
5155 Node* b = argument(1);
5156 Node* n = argument(2);
5157 Node* len = argument(3);
5158 Node* inv = argument(4);
5159 Node* m = argument(6);
5160
5161 const Type* a_type = a->Value(&_gvn);
5162 const TypeAryPtr* top_a = a_type->isa_aryptr();
5163 const Type* b_type = b->Value(&_gvn);
5164 const TypeAryPtr* top_b = b_type->isa_aryptr();
5165 const Type* n_type = a->Value(&_gvn);
5166 const TypeAryPtr* top_n = n_type->isa_aryptr();
5167 const Type* m_type = a->Value(&_gvn);
5168 const TypeAryPtr* top_m = m_type->isa_aryptr();
5169 if (top_a == NULL || top_a->klass() == NULL ||
5170 top_b == NULL || top_b->klass() == NULL ||
5171 top_n == NULL || top_n->klass() == NULL ||
5172 top_m == NULL || top_m->klass() == NULL) {
5173 // failed array check
5174 return false;
5175 }
5176
5177 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5178 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5179 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5180 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5181 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5182 return false;
5183 }
5184
5185 // Make the call
5186 {
5187 Node* a_start = array_element_address(a, intcon(0), a_elem);
5188 Node* b_start = array_element_address(b, intcon(0), b_elem);
5189 Node* n_start = array_element_address(n, intcon(0), n_elem);
5190 Node* m_start = array_element_address(m, intcon(0), m_elem);
5191
5192 Node* call = make_runtime_call(RC_LEAF,
5193 OptoRuntime::montgomeryMultiply_Type(),
5194 stubAddr, stubName, TypePtr::BOTTOM,
5195 a_start, b_start, n_start, len, inv, top(),
5196 m_start);
5197 set_result(m);
5198 }
5199
5200 return true;
5201 }
5202
5203 bool LibraryCallKit::inline_montgomerySquare() {
5204 address stubAddr = StubRoutines::montgomerySquare();
5205 if (stubAddr == NULL) {
5206 return false; // Intrinsic's stub is not implemented on this platform
5207 }
5208
5209 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5210 const char* stubName = "montgomery_square";
5211
5212 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5213
5214 Node* a = argument(0);
5215 Node* n = argument(1);
5216 Node* len = argument(2);
5217 Node* inv = argument(3);
5218 Node* m = argument(5);
5219
5220 const Type* a_type = a->Value(&_gvn);
5221 const TypeAryPtr* top_a = a_type->isa_aryptr();
5222 const Type* n_type = a->Value(&_gvn);
5223 const TypeAryPtr* top_n = n_type->isa_aryptr();
5224 const Type* m_type = a->Value(&_gvn);
5225 const TypeAryPtr* top_m = m_type->isa_aryptr();
5226 if (top_a == NULL || top_a->klass() == NULL ||
5227 top_n == NULL || top_n->klass() == NULL ||
5228 top_m == NULL || top_m->klass() == NULL) {
5229 // failed array check
5230 return false;
5231 }
5232
5233 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5234 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5235 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5236 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5237 return false;
5238 }
5239
5240 // Make the call
5241 {
5242 Node* a_start = array_element_address(a, intcon(0), a_elem);
5243 Node* n_start = array_element_address(n, intcon(0), n_elem);
5244 Node* m_start = array_element_address(m, intcon(0), m_elem);
5245
5246 Node* call = make_runtime_call(RC_LEAF,
5247 OptoRuntime::montgomerySquare_Type(),
5248 stubAddr, stubName, TypePtr::BOTTOM,
5249 a_start, n_start, len, inv, top(),
5250 m_start);
5251 set_result(m);
5252 }
5253
5254 return true;
5255 }
5256
5257 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5258 address stubAddr = NULL;
5259 const char* stubName = NULL;
5260
5261 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5262 if (stubAddr == NULL) {
5263 return false; // Intrinsic's stub is not implemented on this platform
5264 }
5265
5266 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5267
5268 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5269
5270 Node* newArr = argument(0);
5271 Node* oldArr = argument(1);
5272 Node* newIdx = argument(2);
5273 Node* shiftCount = argument(3);
5274 Node* numIter = argument(4);
5275
5276 const Type* newArr_type = newArr->Value(&_gvn);
5277 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5278 const Type* oldArr_type = oldArr->Value(&_gvn);
5279 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5280 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5281 || top_oldArr->klass() == NULL) {
5282 return false;
5283 }
5284
5285 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5286 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5287 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5288 return false;
5289 }
5290
5291 // Make the call
5292 {
5293 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5294 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5295
5296 Node* call = make_runtime_call(RC_LEAF,
5297 OptoRuntime::bigIntegerShift_Type(),
5298 stubAddr,
5299 stubName,
5300 TypePtr::BOTTOM,
5301 newArr_start,
5302 oldArr_start,
5303 newIdx,
5304 shiftCount,
5305 numIter);
5306 }
5307
5308 return true;
5309 }
5310
5311 //-------------inline_vectorizedMismatch------------------------------
5312 bool LibraryCallKit::inline_vectorizedMismatch() {
5313 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5314
5315 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5316 Node* obja = argument(0); // Object
5317 Node* aoffset = argument(1); // long
5318 Node* objb = argument(3); // Object
5319 Node* boffset = argument(4); // long
5320 Node* length = argument(6); // int
5321 Node* scale = argument(7); // int
5322
5323 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5324 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5325 if (obja_t == NULL || obja_t->klass() == NULL ||
5326 objb_t == NULL || objb_t->klass() == NULL ||
5327 scale == top()) {
5328 return false; // failed input validation
5329 }
5330
5331 Node* obja_adr = make_unsafe_address(obja, aoffset);
5332 Node* objb_adr = make_unsafe_address(objb, boffset);
5333
5334 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5335 //
5336 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5337 // if (length <= inline_limit) {
5338 // inline_path:
5339 // vmask = VectorMaskGen length
5340 // vload1 = LoadVectorMasked obja, vmask
5341 // vload2 = LoadVectorMasked objb, vmask
5342 // result1 = VectorCmpMasked vload1, vload2, vmask
5343 // } else {
5344 // call_stub_path:
5345 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5346 // }
5347 // exit_block:
5348 // return Phi(result1, result2);
5349 //
5350 enum { inline_path = 1, // input is small enough to process it all at once
5351 stub_path = 2, // input is too large; call into the VM
5352 PATH_LIMIT = 3
5353 };
5354
5355 Node* exit_block = new RegionNode(PATH_LIMIT);
5356 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5357 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5358
5359 Node* call_stub_path = control();
5360
5361 BasicType elem_bt = T_ILLEGAL;
5362
5363 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5364 if (scale_t->is_con()) {
5365 switch (scale_t->get_con()) {
5366 case 0: elem_bt = T_BYTE; break;
5367 case 1: elem_bt = T_SHORT; break;
5368 case 2: elem_bt = T_INT; break;
5369 case 3: elem_bt = T_LONG; break;
5370
5371 default: elem_bt = T_ILLEGAL; break; // not supported
5372 }
5373 }
5374
5375 int inline_limit = 0;
5376 bool do_partial_inline = false;
5377
5378 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5379 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5380 do_partial_inline = inline_limit >= 16;
5381 }
5382
5383 if (do_partial_inline) {
5384 assert(elem_bt != T_ILLEGAL, "sanity");
5385
5386 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5387 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5388 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5389
5390 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5391 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5392 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5393
5394 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5395
5396 if (!stopped()) {
5397 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5398
5399 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5400 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5401 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5402 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5403
5404 Node* vmask = _gvn.transform(new VectorMaskGenNode(ConvI2X(casted_length), TypeVect::VECTMASK, elem_bt));
5405 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5406 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5407 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5408
5409 exit_block->init_req(inline_path, control());
5410 memory_phi->init_req(inline_path, map()->memory());
5411 result_phi->init_req(inline_path, result);
5412
5413 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5414 clear_upper_avx();
5415 }
5416 }
5417 }
5418
5419 if (call_stub_path != NULL) {
5420 set_control(call_stub_path);
5421
5422 Node* call = make_runtime_call(RC_LEAF,
5423 OptoRuntime::vectorizedMismatch_Type(),
5424 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5425 obja_adr, objb_adr, length, scale);
5426
5427 exit_block->init_req(stub_path, control());
5428 memory_phi->init_req(stub_path, map()->memory());
5429 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5430 }
5431
5432 exit_block = _gvn.transform(exit_block);
5433 memory_phi = _gvn.transform(memory_phi);
5434 result_phi = _gvn.transform(result_phi);
5435
5436 set_control(exit_block);
5437 set_all_memory(memory_phi);
5438 set_result(result_phi);
5439
5440 return true;
5441 }
5442
5443 /**
5444 * Calculate CRC32 for byte.
5445 * int java.util.zip.CRC32.update(int crc, int b)
5446 */
5447 bool LibraryCallKit::inline_updateCRC32() {
5448 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5449 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5450 // no receiver since it is static method
5451 Node* crc = argument(0); // type: int
5452 Node* b = argument(1); // type: int
5453
5454 /*
5455 * int c = ~ crc;
5456 * b = timesXtoThe32[(b ^ c) & 0xFF];
5457 * b = b ^ (c >>> 8);
5458 * crc = ~b;
5459 */
5460
5461 Node* M1 = intcon(-1);
5462 crc = _gvn.transform(new XorINode(crc, M1));
5463 Node* result = _gvn.transform(new XorINode(crc, b));
5464 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5465
5466 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5467 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5468 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5469 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5470
5471 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5472 result = _gvn.transform(new XorINode(crc, result));
5473 result = _gvn.transform(new XorINode(result, M1));
5474 set_result(result);
5475 return true;
5476 }
5477
5478 /**
5479 * Calculate CRC32 for byte[] array.
5480 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5481 */
5482 bool LibraryCallKit::inline_updateBytesCRC32() {
5483 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5484 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5485 // no receiver since it is static method
5486 Node* crc = argument(0); // type: int
5487 Node* src = argument(1); // type: oop
5488 Node* offset = argument(2); // type: int
5489 Node* length = argument(3); // type: int
5490
5491 const Type* src_type = src->Value(&_gvn);
5492 const TypeAryPtr* top_src = src_type->isa_aryptr();
5493 if (top_src == NULL || top_src->klass() == NULL) {
5494 // failed array check
5495 return false;
5496 }
5497
5498 // Figure out the size and type of the elements we will be copying.
5499 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5500 if (src_elem != T_BYTE) {
5501 return false;
5502 }
5503
5504 // 'src_start' points to src array + scaled offset
5505 src = must_be_not_null(src, true);
5506 Node* src_start = array_element_address(src, offset, src_elem);
5507
5508 // We assume that range check is done by caller.
5509 // TODO: generate range check (offset+length < src.length) in debug VM.
5510
5511 // Call the stub.
5512 address stubAddr = StubRoutines::updateBytesCRC32();
5513 const char *stubName = "updateBytesCRC32";
5514
5515 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5516 stubAddr, stubName, TypePtr::BOTTOM,
5517 crc, src_start, length);
5518 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5519 set_result(result);
5520 return true;
5521 }
5522
5523 /**
5524 * Calculate CRC32 for ByteBuffer.
5525 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5526 */
5527 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5528 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5529 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5530 // no receiver since it is static method
5531 Node* crc = argument(0); // type: int
5532 Node* src = argument(1); // type: long
5533 Node* offset = argument(3); // type: int
5534 Node* length = argument(4); // type: int
5535
5536 src = ConvL2X(src); // adjust Java long to machine word
5537 Node* base = _gvn.transform(new CastX2PNode(src));
5538 offset = ConvI2X(offset);
5539
5540 // 'src_start' points to src array + scaled offset
5541 Node* src_start = basic_plus_adr(top(), base, offset);
5542
5543 // Call the stub.
5544 address stubAddr = StubRoutines::updateBytesCRC32();
5545 const char *stubName = "updateBytesCRC32";
5546
5547 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5548 stubAddr, stubName, TypePtr::BOTTOM,
5549 crc, src_start, length);
5550 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5551 set_result(result);
5552 return true;
5553 }
5554
5555 //------------------------------get_table_from_crc32c_class-----------------------
5556 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5557 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
5558 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5559
5560 return table;
5561 }
5562
5563 //------------------------------inline_updateBytesCRC32C-----------------------
5564 //
5565 // Calculate CRC32C for byte[] array.
5566 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5567 //
5568 bool LibraryCallKit::inline_updateBytesCRC32C() {
5569 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5570 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5571 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5572 // no receiver since it is a static method
5573 Node* crc = argument(0); // type: int
5574 Node* src = argument(1); // type: oop
5575 Node* offset = argument(2); // type: int
5576 Node* end = argument(3); // type: int
5577
5578 Node* length = _gvn.transform(new SubINode(end, offset));
5579
5580 const Type* src_type = src->Value(&_gvn);
5581 const TypeAryPtr* top_src = src_type->isa_aryptr();
5582 if (top_src == NULL || top_src->klass() == NULL) {
5583 // failed array check
5584 return false;
5585 }
5586
5587 // Figure out the size and type of the elements we will be copying.
5588 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5589 if (src_elem != T_BYTE) {
5590 return false;
5591 }
5592
5593 // 'src_start' points to src array + scaled offset
5594 src = must_be_not_null(src, true);
5595 Node* src_start = array_element_address(src, offset, src_elem);
5596
5597 // static final int[] byteTable in class CRC32C
5598 Node* table = get_table_from_crc32c_class(callee()->holder());
5599 table = must_be_not_null(table, true);
5600 Node* table_start = array_element_address(table, intcon(0), T_INT);
5601
5602 // We assume that range check is done by caller.
5603 // TODO: generate range check (offset+length < src.length) in debug VM.
5604
5605 // Call the stub.
5606 address stubAddr = StubRoutines::updateBytesCRC32C();
5607 const char *stubName = "updateBytesCRC32C";
5608
5609 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5610 stubAddr, stubName, TypePtr::BOTTOM,
5611 crc, src_start, length, table_start);
5612 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5613 set_result(result);
5614 return true;
5615 }
5616
5617 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5618 //
5619 // Calculate CRC32C for DirectByteBuffer.
5620 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5621 //
5622 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5623 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5624 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5625 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5626 // no receiver since it is a static method
5627 Node* crc = argument(0); // type: int
5628 Node* src = argument(1); // type: long
5629 Node* offset = argument(3); // type: int
5630 Node* end = argument(4); // type: int
5631
5632 Node* length = _gvn.transform(new SubINode(end, offset));
5633
5634 src = ConvL2X(src); // adjust Java long to machine word
5635 Node* base = _gvn.transform(new CastX2PNode(src));
5636 offset = ConvI2X(offset);
5637
5638 // 'src_start' points to src array + scaled offset
5639 Node* src_start = basic_plus_adr(top(), base, offset);
5640
5641 // static final int[] byteTable in class CRC32C
5642 Node* table = get_table_from_crc32c_class(callee()->holder());
5643 table = must_be_not_null(table, true);
5644 Node* table_start = array_element_address(table, intcon(0), T_INT);
5645
5646 // Call the stub.
5647 address stubAddr = StubRoutines::updateBytesCRC32C();
5648 const char *stubName = "updateBytesCRC32C";
5649
5650 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5651 stubAddr, stubName, TypePtr::BOTTOM,
5652 crc, src_start, length, table_start);
5653 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5654 set_result(result);
5655 return true;
5656 }
5657
5658 //------------------------------inline_updateBytesAdler32----------------------
5659 //
5660 // Calculate Adler32 checksum for byte[] array.
5661 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5662 //
5663 bool LibraryCallKit::inline_updateBytesAdler32() {
5664 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5665 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5666 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5667 // no receiver since it is static method
5668 Node* crc = argument(0); // type: int
5669 Node* src = argument(1); // type: oop
5670 Node* offset = argument(2); // type: int
5671 Node* length = argument(3); // type: int
5672
5673 const Type* src_type = src->Value(&_gvn);
5674 const TypeAryPtr* top_src = src_type->isa_aryptr();
5675 if (top_src == NULL || top_src->klass() == NULL) {
5676 // failed array check
5677 return false;
5678 }
5679
5680 // Figure out the size and type of the elements we will be copying.
5681 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5682 if (src_elem != T_BYTE) {
5683 return false;
5684 }
5685
5686 // 'src_start' points to src array + scaled offset
5687 Node* src_start = array_element_address(src, offset, src_elem);
5688
5689 // We assume that range check is done by caller.
5690 // TODO: generate range check (offset+length < src.length) in debug VM.
5691
5692 // Call the stub.
5693 address stubAddr = StubRoutines::updateBytesAdler32();
5694 const char *stubName = "updateBytesAdler32";
5695
5696 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5697 stubAddr, stubName, TypePtr::BOTTOM,
5698 crc, src_start, length);
5699 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5700 set_result(result);
5701 return true;
5702 }
5703
5704 //------------------------------inline_updateByteBufferAdler32---------------
5705 //
5706 // Calculate Adler32 checksum for DirectByteBuffer.
5707 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5708 //
5709 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5710 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5711 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5712 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5713 // no receiver since it is static method
5714 Node* crc = argument(0); // type: int
5715 Node* src = argument(1); // type: long
5716 Node* offset = argument(3); // type: int
5717 Node* length = argument(4); // type: int
5718
5719 src = ConvL2X(src); // adjust Java long to machine word
5720 Node* base = _gvn.transform(new CastX2PNode(src));
5721 offset = ConvI2X(offset);
5722
5723 // 'src_start' points to src array + scaled offset
5724 Node* src_start = basic_plus_adr(top(), base, offset);
5725
5726 // Call the stub.
5727 address stubAddr = StubRoutines::updateBytesAdler32();
5728 const char *stubName = "updateBytesAdler32";
5729
5730 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5731 stubAddr, stubName, TypePtr::BOTTOM,
5732 crc, src_start, length);
5733
5734 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5735 set_result(result);
5736 return true;
5737 }
5738
5739 //----------------------------inline_reference_get----------------------------
5740 // public T java.lang.ref.Reference.get();
5741 bool LibraryCallKit::inline_reference_get() {
5742 const int referent_offset = java_lang_ref_Reference::referent_offset();
5743
5744 // Get the argument:
5745 Node* reference_obj = null_check_receiver();
5746 if (stopped()) return true;
5747
5748 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5749 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
5750 decorators, /*is_static*/ false, NULL);
5751 if (result == NULL) return false;
5752
5753 // Add memory barrier to prevent commoning reads from this field
5754 // across safepoint since GC can change its value.
5755 insert_mem_bar(Op_MemBarCPUOrder);
5756
5757 set_result(result);
5758 return true;
5759 }
5760
5761 //----------------------------inline_reference_refersTo0----------------------------
5762 // bool java.lang.ref.Reference.refersTo0();
5763 // bool java.lang.ref.PhantomReference.refersTo0();
5764 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
5765 // Get arguments:
5766 Node* reference_obj = null_check_receiver();
5767 Node* other_obj = argument(1);
5768 if (stopped()) return true;
5769
5770 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
5771 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
5772 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
5773 decorators, /*is_static*/ false, NULL);
5774 if (referent == NULL) return false;
5775
5776 // Add memory barrier to prevent commoning reads from this field
5777 // across safepoint since GC can change its value.
5778 insert_mem_bar(Op_MemBarCPUOrder);
5779
5780 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
5781 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
5782 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
5783
5784 RegionNode* region = new RegionNode(3);
5785 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
5786
5787 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
5788 region->init_req(1, if_true);
5789 phi->init_req(1, intcon(1));
5790
5791 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
5792 region->init_req(2, if_false);
5793 phi->init_req(2, intcon(0));
5794
5795 set_control(_gvn.transform(region));
5796 record_for_igvn(region);
5797 set_result(_gvn.transform(phi));
5798 return true;
5799 }
5800
5801
5802 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
5803 DecoratorSet decorators = IN_HEAP, bool is_static = false,
5804 ciInstanceKlass* fromKls = NULL) {
5805 if (fromKls == NULL) {
5806 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5807 assert(tinst != NULL, "obj is null");
5808 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5809 fromKls = tinst->klass()->as_instance_klass();
5810 } else {
5811 assert(is_static, "only for static field access");
5812 }
5813 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5814 ciSymbol::make(fieldTypeString),
5815 is_static);
5816
5817 assert (field != NULL, "undefined field");
5818 if (field == NULL) return (Node *) NULL;
5819
5820 if (is_static) {
5821 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5822 fromObj = makecon(tip);
5823 }
5824
5825 // Next code copied from Parse::do_get_xxx():
5826
5827 // Compute address and memory type.
5828 int offset = field->offset_in_bytes();
5829 bool is_vol = field->is_volatile();
5830 ciType* field_klass = field->type();
5831 assert(field_klass->is_loaded(), "should be loaded");
5832 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5833 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5834 BasicType bt = field->layout_type();
5835
5836 // Build the resultant type of the load
5837 const Type *type;
5838 if (bt == T_OBJECT) {
5839 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5840 } else {
5841 type = Type::get_const_basic_type(bt);
5842 }
5843
5844 if (is_vol) {
5845 decorators |= MO_SEQ_CST;
5846 }
5847
5848 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5849 }
5850
5851 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5852 bool is_exact = true, bool is_static = false,
5853 ciInstanceKlass * fromKls = NULL) {
5854 if (fromKls == NULL) {
5855 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5856 assert(tinst != NULL, "obj is null");
5857 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5858 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5859 fromKls = tinst->klass()->as_instance_klass();
5860 }
5861 else {
5862 assert(is_static, "only for static field access");
5863 }
5864 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5865 ciSymbol::make(fieldTypeString),
5866 is_static);
5867
5868 assert(field != NULL, "undefined field");
5869 assert(!field->is_volatile(), "not defined for volatile fields");
5870
5871 if (is_static) {
5872 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5873 fromObj = makecon(tip);
5874 }
5875
5876 // Next code copied from Parse::do_get_xxx():
5877
5878 // Compute address and memory type.
5879 int offset = field->offset_in_bytes();
5880 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5881
5882 return adr;
5883 }
5884
5885 //------------------------------inline_aescrypt_Block-----------------------
5886 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5887 address stubAddr = NULL;
5888 const char *stubName;
5889 assert(UseAES, "need AES instruction support");
5890
5891 switch(id) {
5892 case vmIntrinsics::_aescrypt_encryptBlock:
5893 stubAddr = StubRoutines::aescrypt_encryptBlock();
5894 stubName = "aescrypt_encryptBlock";
5895 break;
5896 case vmIntrinsics::_aescrypt_decryptBlock:
5897 stubAddr = StubRoutines::aescrypt_decryptBlock();
5898 stubName = "aescrypt_decryptBlock";
5899 break;
5900 default:
5901 break;
5902 }
5903 if (stubAddr == NULL) return false;
5904
5905 Node* aescrypt_object = argument(0);
5906 Node* src = argument(1);
5907 Node* src_offset = argument(2);
5908 Node* dest = argument(3);
5909 Node* dest_offset = argument(4);
5910
5911 src = must_be_not_null(src, true);
5912 dest = must_be_not_null(dest, true);
5913
5914 // (1) src and dest are arrays.
5915 const Type* src_type = src->Value(&_gvn);
5916 const Type* dest_type = dest->Value(&_gvn);
5917 const TypeAryPtr* top_src = src_type->isa_aryptr();
5918 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5919 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5920
5921 // for the quick and dirty code we will skip all the checks.
5922 // we are just trying to get the call to be generated.
5923 Node* src_start = src;
5924 Node* dest_start = dest;
5925 if (src_offset != NULL || dest_offset != NULL) {
5926 assert(src_offset != NULL && dest_offset != NULL, "");
5927 src_start = array_element_address(src, src_offset, T_BYTE);
5928 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5929 }
5930
5931 // now need to get the start of its expanded key array
5932 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5933 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5934 if (k_start == NULL) return false;
5935
5936 // Call the stub.
5937 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5938 stubAddr, stubName, TypePtr::BOTTOM,
5939 src_start, dest_start, k_start);
5940
5941 return true;
5942 }
5943
5944 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5945 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5946 address stubAddr = NULL;
5947 const char *stubName = NULL;
5948
5949 assert(UseAES, "need AES instruction support");
5950
5951 switch(id) {
5952 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5953 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5954 stubName = "cipherBlockChaining_encryptAESCrypt";
5955 break;
5956 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5957 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5958 stubName = "cipherBlockChaining_decryptAESCrypt";
5959 break;
5960 default:
5961 break;
5962 }
5963 if (stubAddr == NULL) return false;
5964
5965 Node* cipherBlockChaining_object = argument(0);
5966 Node* src = argument(1);
5967 Node* src_offset = argument(2);
5968 Node* len = argument(3);
5969 Node* dest = argument(4);
5970 Node* dest_offset = argument(5);
5971
5972 src = must_be_not_null(src, false);
5973 dest = must_be_not_null(dest, false);
5974
5975 // (1) src and dest are arrays.
5976 const Type* src_type = src->Value(&_gvn);
5977 const Type* dest_type = dest->Value(&_gvn);
5978 const TypeAryPtr* top_src = src_type->isa_aryptr();
5979 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5980 assert (top_src != NULL && top_src->klass() != NULL
5981 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5982
5983 // checks are the responsibility of the caller
5984 Node* src_start = src;
5985 Node* dest_start = dest;
5986 if (src_offset != NULL || dest_offset != NULL) {
5987 assert(src_offset != NULL && dest_offset != NULL, "");
5988 src_start = array_element_address(src, src_offset, T_BYTE);
5989 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5990 }
5991
5992 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5993 // (because of the predicated logic executed earlier).
5994 // so we cast it here safely.
5995 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5996
5997 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
5998 if (embeddedCipherObj == NULL) return false;
5999
6000 // cast it to what we know it will be at runtime
6001 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6002 assert(tinst != NULL, "CBC obj is null");
6003 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6004 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6005 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6006
6007 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6008 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6009 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6010 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6011 aescrypt_object = _gvn.transform(aescrypt_object);
6012
6013 // we need to get the start of the aescrypt_object's expanded key array
6014 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6015 if (k_start == NULL) return false;
6016
6017 // similarly, get the start address of the r vector
6018 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6019 if (objRvec == NULL) return false;
6020 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6021
6022 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6023 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6024 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6025 stubAddr, stubName, TypePtr::BOTTOM,
6026 src_start, dest_start, k_start, r_start, len);
6027
6028 // return cipher length (int)
6029 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6030 set_result(retvalue);
6031 return true;
6032 }
6033
6034 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6035 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6036 address stubAddr = NULL;
6037 const char *stubName = NULL;
6038
6039 assert(UseAES, "need AES instruction support");
6040
6041 switch (id) {
6042 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6043 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6044 stubName = "electronicCodeBook_encryptAESCrypt";
6045 break;
6046 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6047 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6048 stubName = "electronicCodeBook_decryptAESCrypt";
6049 break;
6050 default:
6051 break;
6052 }
6053
6054 if (stubAddr == NULL) return false;
6055
6056 Node* electronicCodeBook_object = argument(0);
6057 Node* src = argument(1);
6058 Node* src_offset = argument(2);
6059 Node* len = argument(3);
6060 Node* dest = argument(4);
6061 Node* dest_offset = argument(5);
6062
6063 // (1) src and dest are arrays.
6064 const Type* src_type = src->Value(&_gvn);
6065 const Type* dest_type = dest->Value(&_gvn);
6066 const TypeAryPtr* top_src = src_type->isa_aryptr();
6067 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6068 assert(top_src != NULL && top_src->klass() != NULL
6069 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6070
6071 // checks are the responsibility of the caller
6072 Node* src_start = src;
6073 Node* dest_start = dest;
6074 if (src_offset != NULL || dest_offset != NULL) {
6075 assert(src_offset != NULL && dest_offset != NULL, "");
6076 src_start = array_element_address(src, src_offset, T_BYTE);
6077 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6078 }
6079
6080 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6081 // (because of the predicated logic executed earlier).
6082 // so we cast it here safely.
6083 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6084
6085 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6086 if (embeddedCipherObj == NULL) return false;
6087
6088 // cast it to what we know it will be at runtime
6089 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6090 assert(tinst != NULL, "ECB obj is null");
6091 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6092 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6093 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6094
6095 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6096 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6097 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6098 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6099 aescrypt_object = _gvn.transform(aescrypt_object);
6100
6101 // we need to get the start of the aescrypt_object's expanded key array
6102 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6103 if (k_start == NULL) return false;
6104
6105 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6106 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6107 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6108 stubAddr, stubName, TypePtr::BOTTOM,
6109 src_start, dest_start, k_start, len);
6110
6111 // return cipher length (int)
6112 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6113 set_result(retvalue);
6114 return true;
6115 }
6116
6117 //------------------------------inline_counterMode_AESCrypt-----------------------
6118 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6119 assert(UseAES, "need AES instruction support");
6120 if (!UseAESCTRIntrinsics) return false;
6121
6122 address stubAddr = NULL;
6123 const char *stubName = NULL;
6124 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6125 stubAddr = StubRoutines::counterMode_AESCrypt();
6126 stubName = "counterMode_AESCrypt";
6127 }
6128 if (stubAddr == NULL) return false;
6129
6130 Node* counterMode_object = argument(0);
6131 Node* src = argument(1);
6132 Node* src_offset = argument(2);
6133 Node* len = argument(3);
6134 Node* dest = argument(4);
6135 Node* dest_offset = argument(5);
6136
6137 // (1) src and dest are arrays.
6138 const Type* src_type = src->Value(&_gvn);
6139 const Type* dest_type = dest->Value(&_gvn);
6140 const TypeAryPtr* top_src = src_type->isa_aryptr();
6141 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6142 assert(top_src != NULL && top_src->klass() != NULL &&
6143 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6144
6145 // checks are the responsibility of the caller
6146 Node* src_start = src;
6147 Node* dest_start = dest;
6148 if (src_offset != NULL || dest_offset != NULL) {
6149 assert(src_offset != NULL && dest_offset != NULL, "");
6150 src_start = array_element_address(src, src_offset, T_BYTE);
6151 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6152 }
6153
6154 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6155 // (because of the predicated logic executed earlier).
6156 // so we cast it here safely.
6157 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6158 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6159 if (embeddedCipherObj == NULL) return false;
6160 // cast it to what we know it will be at runtime
6161 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6162 assert(tinst != NULL, "CTR obj is null");
6163 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6164 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6165 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6166 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6167 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6168 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6169 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6170 aescrypt_object = _gvn.transform(aescrypt_object);
6171 // we need to get the start of the aescrypt_object's expanded key array
6172 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6173 if (k_start == NULL) return false;
6174 // similarly, get the start address of the r vector
6175 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6176 if (obj_counter == NULL) return false;
6177 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6178
6179 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6180 if (saved_encCounter == NULL) return false;
6181 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6182 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6183
6184 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6185 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6186 OptoRuntime::counterMode_aescrypt_Type(),
6187 stubAddr, stubName, TypePtr::BOTTOM,
6188 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6189
6190 // return cipher length (int)
6191 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6192 set_result(retvalue);
6193 return true;
6194 }
6195
6196 //------------------------------get_key_start_from_aescrypt_object-----------------------
6197 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6198 #if defined(PPC64) || defined(S390)
6199 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6200 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6201 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6202 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6203 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6204 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6205 if (objSessionK == NULL) {
6206 return (Node *) NULL;
6207 }
6208 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6209 #else
6210 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6211 #endif // PPC64
6212 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6213 if (objAESCryptKey == NULL) return (Node *) NULL;
6214
6215 // now have the array, need to get the start address of the K array
6216 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6217 return k_start;
6218 }
6219
6220 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6221 // Return node representing slow path of predicate check.
6222 // the pseudo code we want to emulate with this predicate is:
6223 // for encryption:
6224 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6225 // for decryption:
6226 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6227 // note cipher==plain is more conservative than the original java code but that's OK
6228 //
6229 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6230 // The receiver was checked for NULL already.
6231 Node* objCBC = argument(0);
6232
6233 Node* src = argument(1);
6234 Node* dest = argument(4);
6235
6236 // Load embeddedCipher field of CipherBlockChaining object.
6237 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6238
6239 // get AESCrypt klass for instanceOf check
6240 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6241 // will have same classloader as CipherBlockChaining object
6242 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6243 assert(tinst != NULL, "CBCobj is null");
6244 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6245
6246 // we want to do an instanceof comparison against the AESCrypt class
6247 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6248 if (!klass_AESCrypt->is_loaded()) {
6249 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6250 Node* ctrl = control();
6251 set_control(top()); // no regular fast path
6252 return ctrl;
6253 }
6254
6255 src = must_be_not_null(src, true);
6256 dest = must_be_not_null(dest, true);
6257
6258 // Resolve oops to stable for CmpP below.
6259 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6260
6261 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6262 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6263 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6264
6265 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6266
6267 // for encryption, we are done
6268 if (!decrypting)
6269 return instof_false; // even if it is NULL
6270
6271 // for decryption, we need to add a further check to avoid
6272 // taking the intrinsic path when cipher and plain are the same
6273 // see the original java code for why.
6274 RegionNode* region = new RegionNode(3);
6275 region->init_req(1, instof_false);
6276
6277 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6278 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6279 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6280 region->init_req(2, src_dest_conjoint);
6281
6282 record_for_igvn(region);
6283 return _gvn.transform(region);
6284 }
6285
6286 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6287 // Return node representing slow path of predicate check.
6288 // the pseudo code we want to emulate with this predicate is:
6289 // for encryption:
6290 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6291 // for decryption:
6292 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6293 // note cipher==plain is more conservative than the original java code but that's OK
6294 //
6295 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6296 // The receiver was checked for NULL already.
6297 Node* objECB = argument(0);
6298
6299 // Load embeddedCipher field of ElectronicCodeBook object.
6300 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6301
6302 // get AESCrypt klass for instanceOf check
6303 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6304 // will have same classloader as ElectronicCodeBook object
6305 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6306 assert(tinst != NULL, "ECBobj is null");
6307 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6308
6309 // we want to do an instanceof comparison against the AESCrypt class
6310 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6311 if (!klass_AESCrypt->is_loaded()) {
6312 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6313 Node* ctrl = control();
6314 set_control(top()); // no regular fast path
6315 return ctrl;
6316 }
6317 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6318
6319 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6320 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6321 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6322
6323 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6324
6325 // for encryption, we are done
6326 if (!decrypting)
6327 return instof_false; // even if it is NULL
6328
6329 // for decryption, we need to add a further check to avoid
6330 // taking the intrinsic path when cipher and plain are the same
6331 // see the original java code for why.
6332 RegionNode* region = new RegionNode(3);
6333 region->init_req(1, instof_false);
6334 Node* src = argument(1);
6335 Node* dest = argument(4);
6336 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6337 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6338 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6339 region->init_req(2, src_dest_conjoint);
6340
6341 record_for_igvn(region);
6342 return _gvn.transform(region);
6343 }
6344
6345 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6346 // Return node representing slow path of predicate check.
6347 // the pseudo code we want to emulate with this predicate is:
6348 // for encryption:
6349 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6350 // for decryption:
6351 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6352 // note cipher==plain is more conservative than the original java code but that's OK
6353 //
6354
6355 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6356 // The receiver was checked for NULL already.
6357 Node* objCTR = argument(0);
6358
6359 // Load embeddedCipher field of CipherBlockChaining object.
6360 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6361
6362 // get AESCrypt klass for instanceOf check
6363 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6364 // will have same classloader as CipherBlockChaining object
6365 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6366 assert(tinst != NULL, "CTRobj is null");
6367 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6368
6369 // we want to do an instanceof comparison against the AESCrypt class
6370 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6371 if (!klass_AESCrypt->is_loaded()) {
6372 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6373 Node* ctrl = control();
6374 set_control(top()); // no regular fast path
6375 return ctrl;
6376 }
6377
6378 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6379 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6380 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6381 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6382 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6383
6384 return instof_false; // even if it is NULL
6385 }
6386
6387 //------------------------------inline_ghash_processBlocks
6388 bool LibraryCallKit::inline_ghash_processBlocks() {
6389 address stubAddr;
6390 const char *stubName;
6391 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6392
6393 stubAddr = StubRoutines::ghash_processBlocks();
6394 stubName = "ghash_processBlocks";
6395
6396 Node* data = argument(0);
6397 Node* offset = argument(1);
6398 Node* len = argument(2);
6399 Node* state = argument(3);
6400 Node* subkeyH = argument(4);
6401
6402 state = must_be_not_null(state, true);
6403 subkeyH = must_be_not_null(subkeyH, true);
6404 data = must_be_not_null(data, true);
6405
6406 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6407 assert(state_start, "state is NULL");
6408 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6409 assert(subkeyH_start, "subkeyH is NULL");
6410 Node* data_start = array_element_address(data, offset, T_BYTE);
6411 assert(data_start, "data is NULL");
6412
6413 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6414 OptoRuntime::ghash_processBlocks_Type(),
6415 stubAddr, stubName, TypePtr::BOTTOM,
6416 state_start, subkeyH_start, data_start, len);
6417 return true;
6418 }
6419
6420 bool LibraryCallKit::inline_base64_encodeBlock() {
6421 address stubAddr;
6422 const char *stubName;
6423 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6424 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6425 stubAddr = StubRoutines::base64_encodeBlock();
6426 stubName = "encodeBlock";
6427
6428 if (!stubAddr) return false;
6429 Node* base64obj = argument(0);
6430 Node* src = argument(1);
6431 Node* offset = argument(2);
6432 Node* len = argument(3);
6433 Node* dest = argument(4);
6434 Node* dp = argument(5);
6435 Node* isURL = argument(6);
6436
6437 src = must_be_not_null(src, true);
6438 dest = must_be_not_null(dest, true);
6439
6440 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6441 assert(src_start, "source array is NULL");
6442 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6443 assert(dest_start, "destination array is NULL");
6444
6445 Node* base64 = make_runtime_call(RC_LEAF,
6446 OptoRuntime::base64_encodeBlock_Type(),
6447 stubAddr, stubName, TypePtr::BOTTOM,
6448 src_start, offset, len, dest_start, dp, isURL);
6449 return true;
6450 }
6451
6452 bool LibraryCallKit::inline_base64_decodeBlock() {
6453 address stubAddr;
6454 const char *stubName;
6455 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6456 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6457 stubAddr = StubRoutines::base64_decodeBlock();
6458 stubName = "decodeBlock";
6459
6460 if (!stubAddr) return false;
6461 Node* base64obj = argument(0);
6462 Node* src = argument(1);
6463 Node* src_offset = argument(2);
6464 Node* len = argument(3);
6465 Node* dest = argument(4);
6466 Node* dest_offset = argument(5);
6467 Node* isURL = argument(6);
6468 Node* isMIME = argument(7);
6469
6470 src = must_be_not_null(src, true);
6471 dest = must_be_not_null(dest, true);
6472
6473 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6474 assert(src_start, "source array is NULL");
6475 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6476 assert(dest_start, "destination array is NULL");
6477
6478 Node* call = make_runtime_call(RC_LEAF,
6479 OptoRuntime::base64_decodeBlock_Type(),
6480 stubAddr, stubName, TypePtr::BOTTOM,
6481 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6482 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6483 set_result(result);
6484 return true;
6485 }
6486
6487 //------------------------------inline_digestBase_implCompress-----------------------
6488 //
6489 // Calculate MD5 for single-block byte[] array.
6490 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6491 //
6492 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6493 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6494 //
6495 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6496 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6497 //
6498 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6499 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6500 //
6501 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6502 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6503 //
6504 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6505 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6506
6507 Node* digestBase_obj = argument(0);
6508 Node* src = argument(1); // type oop
6509 Node* ofs = argument(2); // type int
6510
6511 const Type* src_type = src->Value(&_gvn);
6512 const TypeAryPtr* top_src = src_type->isa_aryptr();
6513 if (top_src == NULL || top_src->klass() == NULL) {
6514 // failed array check
6515 return false;
6516 }
6517 // Figure out the size and type of the elements we will be copying.
6518 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6519 if (src_elem != T_BYTE) {
6520 return false;
6521 }
6522 // 'src_start' points to src array + offset
6523 src = must_be_not_null(src, true);
6524 Node* src_start = array_element_address(src, ofs, src_elem);
6525 Node* state = NULL;
6526 Node* digest_length = NULL;
6527 address stubAddr;
6528 const char *stubName;
6529
6530 switch(id) {
6531 case vmIntrinsics::_md5_implCompress:
6532 assert(UseMD5Intrinsics, "need MD5 instruction support");
6533 state = get_state_from_digest_object(digestBase_obj, "[I");
6534 stubAddr = StubRoutines::md5_implCompress();
6535 stubName = "md5_implCompress";
6536 break;
6537 case vmIntrinsics::_sha_implCompress:
6538 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6539 state = get_state_from_digest_object(digestBase_obj, "[I");
6540 stubAddr = StubRoutines::sha1_implCompress();
6541 stubName = "sha1_implCompress";
6542 break;
6543 case vmIntrinsics::_sha2_implCompress:
6544 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6545 state = get_state_from_digest_object(digestBase_obj, "[I");
6546 stubAddr = StubRoutines::sha256_implCompress();
6547 stubName = "sha256_implCompress";
6548 break;
6549 case vmIntrinsics::_sha5_implCompress:
6550 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6551 state = get_state_from_digest_object(digestBase_obj, "[J");
6552 stubAddr = StubRoutines::sha512_implCompress();
6553 stubName = "sha512_implCompress";
6554 break;
6555 case vmIntrinsics::_sha3_implCompress:
6556 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
6557 state = get_state_from_digest_object(digestBase_obj, "[B");
6558 stubAddr = StubRoutines::sha3_implCompress();
6559 stubName = "sha3_implCompress";
6560 digest_length = get_digest_length_from_digest_object(digestBase_obj);
6561 if (digest_length == NULL) return false;
6562 break;
6563 default:
6564 fatal_unexpected_iid(id);
6565 return false;
6566 }
6567 if (state == NULL) return false;
6568
6569 assert(stubAddr != NULL, "Stub is generated");
6570 if (stubAddr == NULL) return false;
6571
6572 // Call the stub.
6573 Node* call;
6574 if (digest_length == NULL) {
6575 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
6576 stubAddr, stubName, TypePtr::BOTTOM,
6577 src_start, state);
6578 } else {
6579 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
6580 stubAddr, stubName, TypePtr::BOTTOM,
6581 src_start, state, digest_length);
6582 }
6583
6584 return true;
6585 }
6586
6587 //------------------------------inline_digestBase_implCompressMB-----------------------
6588 //
6589 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
6590 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6591 //
6592 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6593 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6594 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6595 assert((uint)predicate < 5, "sanity");
6596 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6597
6598 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6599 Node* src = argument(1); // byte[] array
6600 Node* ofs = argument(2); // type int
6601 Node* limit = argument(3); // type int
6602
6603 const Type* src_type = src->Value(&_gvn);
6604 const TypeAryPtr* top_src = src_type->isa_aryptr();
6605 if (top_src == NULL || top_src->klass() == NULL) {
6606 // failed array check
6607 return false;
6608 }
6609 // Figure out the size and type of the elements we will be copying.
6610 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6611 if (src_elem != T_BYTE) {
6612 return false;
6613 }
6614 // 'src_start' points to src array + offset
6615 src = must_be_not_null(src, false);
6616 Node* src_start = array_element_address(src, ofs, src_elem);
6617
6618 const char* klass_digestBase_name = NULL;
6619 const char* stub_name = NULL;
6620 address stub_addr = NULL;
6621 const char* state_type = "[I";
6622
6623 switch (predicate) {
6624 case 0:
6625 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
6626 klass_digestBase_name = "sun/security/provider/MD5";
6627 stub_name = "md5_implCompressMB";
6628 stub_addr = StubRoutines::md5_implCompressMB();
6629 }
6630 break;
6631 case 1:
6632 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
6633 klass_digestBase_name = "sun/security/provider/SHA";
6634 stub_name = "sha1_implCompressMB";
6635 stub_addr = StubRoutines::sha1_implCompressMB();
6636 }
6637 break;
6638 case 2:
6639 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
6640 klass_digestBase_name = "sun/security/provider/SHA2";
6641 stub_name = "sha256_implCompressMB";
6642 stub_addr = StubRoutines::sha256_implCompressMB();
6643 }
6644 break;
6645 case 3:
6646 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
6647 klass_digestBase_name = "sun/security/provider/SHA5";
6648 stub_name = "sha512_implCompressMB";
6649 stub_addr = StubRoutines::sha512_implCompressMB();
6650 state_type = "[J";
6651 }
6652 break;
6653 case 4:
6654 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
6655 klass_digestBase_name = "sun/security/provider/SHA3";
6656 stub_name = "sha3_implCompressMB";
6657 stub_addr = StubRoutines::sha3_implCompressMB();
6658 state_type = "[B";
6659 }
6660 break;
6661 default:
6662 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
6663 }
6664 if (klass_digestBase_name != NULL) {
6665 assert(stub_addr != NULL, "Stub is generated");
6666 if (stub_addr == NULL) return false;
6667
6668 // get DigestBase klass to lookup for SHA klass
6669 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6670 assert(tinst != NULL, "digestBase_obj is not instance???");
6671 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6672
6673 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
6674 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
6675 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
6676 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, state_type, stub_addr, stub_name, src_start, ofs, limit);
6677 }
6678 return false;
6679 }
6680
6681 //------------------------------inline_digestBase_implCompressMB-----------------------
6682 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
6683 const char* state_type, address stubAddr, const char *stubName,
6684 Node* src_start, Node* ofs, Node* limit) {
6685 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
6686 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6687 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6688 digest_obj = _gvn.transform(digest_obj);
6689
6690 Node* state = get_state_from_digest_object(digest_obj, state_type);
6691 if (state == NULL) return false;
6692
6693 Node* digest_length = NULL;
6694 if (strcmp("sha3_implCompressMB", stubName) == 0) {
6695 digest_length = get_digest_length_from_digest_object(digest_obj);
6696 if (digest_length == NULL) return false;
6697 }
6698
6699 // Call the stub.
6700 Node* call;
6701 if (digest_length == NULL) {
6702 call = make_runtime_call(RC_LEAF|RC_NO_FP,
6703 OptoRuntime::digestBase_implCompressMB_Type(false),
6704 stubAddr, stubName, TypePtr::BOTTOM,
6705 src_start, state, ofs, limit);
6706 } else {
6707 call = make_runtime_call(RC_LEAF|RC_NO_FP,
6708 OptoRuntime::digestBase_implCompressMB_Type(true),
6709 stubAddr, stubName, TypePtr::BOTTOM,
6710 src_start, state, digest_length, ofs, limit);
6711 }
6712
6713 // return ofs (int)
6714 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6715 set_result(result);
6716
6717 return true;
6718 }
6719
6720 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
6721 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
6722 assert(UseAES, "need AES instruction support");
6723 address stubAddr = NULL;
6724 const char *stubName = NULL;
6725 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
6726 stubName = "galoisCounterMode_AESCrypt";
6727
6728 if (stubAddr == NULL) return false;
6729
6730 Node* in = argument(0);
6731 Node* inOfs = argument(1);
6732 Node* len = argument(2);
6733 Node* ct = argument(3);
6734 Node* ctOfs = argument(4);
6735 Node* out = argument(5);
6736 Node* outOfs = argument(6);
6737 Node* gctr_object = argument(7);
6738 Node* ghash_object = argument(8);
6739
6740 // (1) in, ct and out are arrays.
6741 const Type* in_type = in->Value(&_gvn);
6742 const Type* ct_type = ct->Value(&_gvn);
6743 const Type* out_type = out->Value(&_gvn);
6744 const TypeAryPtr* top_in = in_type->isa_aryptr();
6745 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
6746 const TypeAryPtr* top_out = out_type->isa_aryptr();
6747 assert(top_in != NULL && top_in->klass() != NULL &&
6748 top_ct != NULL && top_ct->klass() != NULL &&
6749 top_out != NULL && top_out->klass() != NULL, "args are strange");
6750
6751 // checks are the responsibility of the caller
6752 Node* in_start = in;
6753 Node* ct_start = ct;
6754 Node* out_start = out;
6755 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
6756 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
6757 in_start = array_element_address(in, inOfs, T_BYTE);
6758 ct_start = array_element_address(ct, ctOfs, T_BYTE);
6759 out_start = array_element_address(out, outOfs, T_BYTE);
6760 }
6761
6762 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6763 // (because of the predicated logic executed earlier).
6764 // so we cast it here safely.
6765 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6766 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6767 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
6768 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
6769 Node* state = load_field_from_object(ghash_object, "state", "[J");
6770
6771 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
6772 return false;
6773 }
6774 // cast it to what we know it will be at runtime
6775 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
6776 assert(tinst != NULL, "GCTR obj is null");
6777 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
6778 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6779 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6780 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6781 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6782 const TypeOopPtr* xtype = aklass->as_instance_type();
6783 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6784 aescrypt_object = _gvn.transform(aescrypt_object);
6785 // we need to get the start of the aescrypt_object's expanded key array
6786 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6787 if (k_start == NULL) return false;
6788
6789 // similarly, get the start address of the r vector
6790 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
6791 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6792 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
6793
6794 ciKlass* klass = ciTypeArrayKlass::make(T_LONG);
6795 Node* klass_node = makecon(TypeKlassPtr::make(klass));
6796
6797 // htbl entries is set to 96 only fox x86-64
6798 if (Matcher::htbl_entries == -1) return false;
6799
6800 // new array to hold 48 computed htbl entries
6801 Node* subkeyHtbl_48_entries = new_array(klass_node, intcon(Matcher::htbl_entries), 0);
6802 if (subkeyHtbl_48_entries == NULL) return false;
6803
6804 Node* subkeyHtbl_48_entries_start = array_element_address(subkeyHtbl_48_entries, intcon(0), T_LONG);
6805
6806 // Call the stub, passing params
6807 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6808 OptoRuntime::galoisCounterMode_aescrypt_Type(),
6809 stubAddr, stubName, TypePtr::BOTTOM,
6810 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, subkeyHtbl_48_entries_start, cnt_start);
6811
6812 // return cipher length (int)
6813 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
6814 set_result(retvalue);
6815 return true;
6816 }
6817
6818 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
6819 // Return node representing slow path of predicate check.
6820 // the pseudo code we want to emulate with this predicate is:
6821 // for encryption:
6822 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6823 // for decryption:
6824 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6825 // note cipher==plain is more conservative than the original java code but that's OK
6826 //
6827
6828 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
6829 // The receiver was checked for NULL already.
6830 Node* objGCTR = argument(7);
6831 // Load embeddedCipher field of GCTR object.
6832 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6833 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
6834
6835 // get AESCrypt klass for instanceOf check
6836 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6837 // will have same classloader as CipherBlockChaining object
6838 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
6839 assert(tinst != NULL, "GCTR obj is null");
6840 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
6841
6842 // we want to do an instanceof comparison against the AESCrypt class
6843 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6844 if (!klass_AESCrypt->is_loaded()) {
6845 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6846 Node* ctrl = control();
6847 set_control(top()); // no regular fast path
6848 return ctrl;
6849 }
6850
6851 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6852 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6853 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6854 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6855 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6856
6857 return instof_false; // even if it is NULL
6858 }
6859
6860 //------------------------------get_state_from_digest_object-----------------------
6861 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, const char *state_type) {
6862 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
6863 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
6864 if (digest_state == NULL) return (Node *) NULL;
6865
6866 // now have the array, need to get the start address of the state array
6867 Node* state = array_element_address(digest_state, intcon(0), T_INT);
6868 return state;
6869 }
6870
6871 //------------------------------get_digest_length_from_sha3_object----------------------------------
6872 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
6873 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
6874 assert (digest_length != NULL, "sanity");
6875 return digest_length;
6876 }
6877
6878 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6879 // Return node representing slow path of predicate check.
6880 // the pseudo code we want to emulate with this predicate is:
6881 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
6882 //
6883 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6884 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6885 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6886 assert((uint)predicate < 5, "sanity");
6887
6888 // The receiver was checked for NULL already.
6889 Node* digestBaseObj = argument(0);
6890
6891 // get DigestBase klass for instanceOf check
6892 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6893 assert(tinst != NULL, "digestBaseObj is null");
6894 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6895
6896 const char* klass_name = NULL;
6897 switch (predicate) {
6898 case 0:
6899 if (UseMD5Intrinsics) {
6900 // we want to do an instanceof comparison against the MD5 class
6901 klass_name = "sun/security/provider/MD5";
6902 }
6903 break;
6904 case 1:
6905 if (UseSHA1Intrinsics) {
6906 // we want to do an instanceof comparison against the SHA class
6907 klass_name = "sun/security/provider/SHA";
6908 }
6909 break;
6910 case 2:
6911 if (UseSHA256Intrinsics) {
6912 // we want to do an instanceof comparison against the SHA2 class
6913 klass_name = "sun/security/provider/SHA2";
6914 }
6915 break;
6916 case 3:
6917 if (UseSHA512Intrinsics) {
6918 // we want to do an instanceof comparison against the SHA5 class
6919 klass_name = "sun/security/provider/SHA5";
6920 }
6921 break;
6922 case 4:
6923 if (UseSHA3Intrinsics) {
6924 // we want to do an instanceof comparison against the SHA3 class
6925 klass_name = "sun/security/provider/SHA3";
6926 }
6927 break;
6928 default:
6929 fatal("unknown SHA intrinsic predicate: %d", predicate);
6930 }
6931
6932 ciKlass* klass = NULL;
6933 if (klass_name != NULL) {
6934 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
6935 }
6936 if ((klass == NULL) || !klass->is_loaded()) {
6937 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6938 Node* ctrl = control();
6939 set_control(top()); // no intrinsic path
6940 return ctrl;
6941 }
6942 ciInstanceKlass* instklass = klass->as_instance_klass();
6943
6944 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
6945 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6946 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6947 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6948
6949 return instof_false; // even if it is NULL
6950 }
6951
6952 //-------------inline_fma-----------------------------------
6953 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6954 Node *a = NULL;
6955 Node *b = NULL;
6956 Node *c = NULL;
6957 Node* result = NULL;
6958 switch (id) {
6959 case vmIntrinsics::_fmaD:
6960 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6961 // no receiver since it is static method
6962 a = round_double_node(argument(0));
6963 b = round_double_node(argument(2));
6964 c = round_double_node(argument(4));
6965 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6966 break;
6967 case vmIntrinsics::_fmaF:
6968 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6969 a = argument(0);
6970 b = argument(1);
6971 c = argument(2);
6972 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6973 break;
6974 default:
6975 fatal_unexpected_iid(id); break;
6976 }
6977 set_result(result);
6978 return true;
6979 }
6980
6981 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6982 // argument(0) is receiver
6983 Node* codePoint = argument(1);
6984 Node* n = NULL;
6985
6986 switch (id) {
6987 case vmIntrinsics::_isDigit :
6988 n = new DigitNode(control(), codePoint);
6989 break;
6990 case vmIntrinsics::_isLowerCase :
6991 n = new LowerCaseNode(control(), codePoint);
6992 break;
6993 case vmIntrinsics::_isUpperCase :
6994 n = new UpperCaseNode(control(), codePoint);
6995 break;
6996 case vmIntrinsics::_isWhitespace :
6997 n = new WhitespaceNode(control(), codePoint);
6998 break;
6999 default:
7000 fatal_unexpected_iid(id);
7001 }
7002
7003 set_result(_gvn.transform(n));
7004 return true;
7005 }
7006
7007 //------------------------------inline_fp_min_max------------------------------
7008 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7009 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7010
7011 // The intrinsic should be used only when the API branches aren't predictable,
7012 // the last one performing the most important comparison. The following heuristic
7013 // uses the branch statistics to eventually bail out if necessary.
7014
7015 ciMethodData *md = callee()->method_data();
7016
7017 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7018 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7019
7020 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7021 // Bail out if the call-site didn't contribute enough to the statistics.
7022 return false;
7023 }
7024
7025 uint taken = 0, not_taken = 0;
7026
7027 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7028 if (p->is_BranchData()) {
7029 taken = ((ciBranchData*)p)->taken();
7030 not_taken = ((ciBranchData*)p)->not_taken();
7031 }
7032 }
7033
7034 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7035 balance = balance < 0 ? -balance : balance;
7036 if ( balance > 0.2 ) {
7037 // Bail out if the most important branch is predictable enough.
7038 return false;
7039 }
7040 }
7041 */
7042
7043 Node *a = NULL;
7044 Node *b = NULL;
7045 Node *n = NULL;
7046 switch (id) {
7047 case vmIntrinsics::_maxF:
7048 case vmIntrinsics::_minF:
7049 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7050 a = argument(0);
7051 b = argument(1);
7052 break;
7053 case vmIntrinsics::_maxD:
7054 case vmIntrinsics::_minD:
7055 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7056 a = round_double_node(argument(0));
7057 b = round_double_node(argument(2));
7058 break;
7059 default:
7060 fatal_unexpected_iid(id);
7061 break;
7062 }
7063 switch (id) {
7064 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
7065 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
7066 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
7067 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
7068 default: fatal_unexpected_iid(id); break;
7069 }
7070 set_result(_gvn.transform(n));
7071 return true;
7072 }
7073
7074 bool LibraryCallKit::inline_profileBoolean() {
7075 Node* counts = argument(1);
7076 const TypeAryPtr* ary = NULL;
7077 ciArray* aobj = NULL;
7078 if (counts->is_Con()
7079 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7080 && (aobj = ary->const_oop()->as_array()) != NULL
7081 && (aobj->length() == 2)) {
7082 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7083 jint false_cnt = aobj->element_value(0).as_int();
7084 jint true_cnt = aobj->element_value(1).as_int();
7085
7086 if (C->log() != NULL) {
7087 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7088 false_cnt, true_cnt);
7089 }
7090
7091 if (false_cnt + true_cnt == 0) {
7092 // According to profile, never executed.
7093 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7094 Deoptimization::Action_reinterpret);
7095 return true;
7096 }
7097
7098 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7099 // is a number of each value occurrences.
7100 Node* result = argument(0);
7101 if (false_cnt == 0 || true_cnt == 0) {
7102 // According to profile, one value has been never seen.
7103 int expected_val = (false_cnt == 0) ? 1 : 0;
7104
7105 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7106 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7107
7108 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7109 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7110 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7111
7112 { // Slow path: uncommon trap for never seen value and then reexecute
7113 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7114 // the value has been seen at least once.
7115 PreserveJVMState pjvms(this);
7116 PreserveReexecuteState preexecs(this);
7117 jvms()->set_should_reexecute(true);
7118
7119 set_control(slow_path);
7120 set_i_o(i_o());
7121
7122 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7123 Deoptimization::Action_reinterpret);
7124 }
7125 // The guard for never seen value enables sharpening of the result and
7126 // returning a constant. It allows to eliminate branches on the same value
7127 // later on.
7128 set_control(fast_path);
7129 result = intcon(expected_val);
7130 }
7131 // Stop profiling.
7132 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7133 // By replacing method body with profile data (represented as ProfileBooleanNode
7134 // on IR level) we effectively disable profiling.
7135 // It enables full speed execution once optimized code is generated.
7136 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7137 C->record_for_igvn(profile);
7138 set_result(profile);
7139 return true;
7140 } else {
7141 // Continue profiling.
7142 // Profile data isn't available at the moment. So, execute method's bytecode version.
7143 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7144 // is compiled and counters aren't available since corresponding MethodHandle
7145 // isn't a compile-time constant.
7146 return false;
7147 }
7148 }
7149
7150 bool LibraryCallKit::inline_isCompileConstant() {
7151 Node* n = argument(0);
7152 set_result(n->is_Con() ? intcon(1) : intcon(0));
7153 return true;
7154 }
7155
7156 //------------------------------- inline_getObjectSize --------------------------------------
7157 //
7158 // Calculate the runtime size of the object/array.
7159 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7160 //
7161 bool LibraryCallKit::inline_getObjectSize() {
7162 Node* obj = argument(3);
7163 Node* klass_node = load_object_klass(obj);
7164
7165 jint layout_con = Klass::_lh_neutral_value;
7166 Node* layout_val = get_layout_helper(klass_node, layout_con);
7167 int layout_is_con = (layout_val == NULL);
7168
7169 if (layout_is_con) {
7170 // Layout helper is constant, can figure out things at compile time.
7171
7172 if (Klass::layout_helper_is_instance(layout_con)) {
7173 // Instance case: layout_con contains the size itself.
7174 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7175 set_result(size);
7176 } else {
7177 // Array case: size is round(header + element_size*arraylength).
7178 // Since arraylength is different for every array instance, we have to
7179 // compute the whole thing at runtime.
7180
7181 Node* arr_length = load_array_length(obj);
7182
7183 int round_mask = MinObjAlignmentInBytes - 1;
7184 int hsize = Klass::layout_helper_header_size(layout_con);
7185 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7186
7187 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7188 round_mask = 0; // strength-reduce it if it goes away completely
7189 }
7190 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7191 Node* header_size = intcon(hsize + round_mask);
7192
7193 Node* lengthx = ConvI2X(arr_length);
7194 Node* headerx = ConvI2X(header_size);
7195
7196 Node* abody = lengthx;
7197 if (eshift != 0) {
7198 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7199 }
7200 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7201 if (round_mask != 0) {
7202 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7203 }
7204 size = ConvX2L(size);
7205 set_result(size);
7206 }
7207 } else {
7208 // Layout helper is not constant, need to test for array-ness at runtime.
7209
7210 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7211 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7212 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7213 record_for_igvn(result_reg);
7214
7215 Node* array_ctl = generate_array_guard(klass_node, NULL);
7216 if (array_ctl != NULL) {
7217 // Array case: size is round(header + element_size*arraylength).
7218 // Since arraylength is different for every array instance, we have to
7219 // compute the whole thing at runtime.
7220
7221 PreserveJVMState pjvms(this);
7222 set_control(array_ctl);
7223 Node* arr_length = load_array_length(obj);
7224
7225 int round_mask = MinObjAlignmentInBytes - 1;
7226 Node* mask = intcon(round_mask);
7227
7228 Node* hss = intcon(Klass::_lh_header_size_shift);
7229 Node* hsm = intcon(Klass::_lh_header_size_mask);
7230 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7231 header_size = _gvn.transform(new AndINode(header_size, hsm));
7232 header_size = _gvn.transform(new AddINode(header_size, mask));
7233
7234 // There is no need to mask or shift this value.
7235 // The semantics of LShiftINode include an implicit mask to 0x1F.
7236 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7237 Node* elem_shift = layout_val;
7238
7239 Node* lengthx = ConvI2X(arr_length);
7240 Node* headerx = ConvI2X(header_size);
7241
7242 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7243 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7244 if (round_mask != 0) {
7245 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7246 }
7247 size = ConvX2L(size);
7248
7249 result_reg->init_req(_array_path, control());
7250 result_val->init_req(_array_path, size);
7251 }
7252
7253 if (!stopped()) {
7254 // Instance case: the layout helper gives us instance size almost directly,
7255 // but we need to mask out the _lh_instance_slow_path_bit.
7256 Node* size = ConvI2X(layout_val);
7257 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7258 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7259 size = _gvn.transform(new AndXNode(size, mask));
7260 size = ConvX2L(size);
7261
7262 result_reg->init_req(_instance_path, control());
7263 result_val->init_req(_instance_path, size);
7264 }
7265
7266 set_result(result_reg, result_val);
7267 }
7268
7269 return true;
7270 }
7271
7272 //------------------------------- inline_blackhole --------------------------------------
7273 //
7274 // Make sure all arguments to this node are alive.
7275 // This matches methods that were requested to be blackholed through compile commands.
7276 //
7277 bool LibraryCallKit::inline_blackhole() {
7278 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7279 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7280 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7281
7282 // Bind call arguments as blackhole arguments to keep them alive
7283 Node* bh = insert_mem_bar(Op_Blackhole);
7284 uint nargs = callee()->arg_size();
7285 for (uint i = 0; i < nargs; i++) {
7286 bh->add_req(argument(i));
7287 }
7288
7289 return true;
7290 }