1 /*
2 * Copyright (c) 1999, 2022, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/unsafe.hpp"
54 #include "runtime/objectMonitor.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "utilities/macros.hpp"
58 #include "utilities/powerOfTwo.hpp"
59
60 //---------------------------make_vm_intrinsic----------------------------
61 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
62 vmIntrinsicID id = m->intrinsic_id();
63 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
64
65 if (!m->is_loaded()) {
66 // Do not attempt to inline unloaded methods.
67 return NULL;
68 }
69
70 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
71 bool is_available = false;
72
73 {
74 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
75 // the compiler must transition to '_thread_in_vm' state because both
76 // methods access VM-internal data.
77 VM_ENTRY_MARK;
78 methodHandle mh(THREAD, m->get_Method());
79 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
80 !C->directive()->is_intrinsic_disabled(mh) &&
81 !vmIntrinsics::is_disabled_by_flags(mh);
82
83 }
84
85 if (is_available) {
86 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
87 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
88 return new LibraryIntrinsic(m, is_virtual,
89 vmIntrinsics::predicates_needed(id),
90 vmIntrinsics::does_virtual_dispatch(id),
91 id);
92 } else {
93 return NULL;
94 }
95 }
96
97 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
98 LibraryCallKit kit(jvms, this);
99 Compile* C = kit.C;
100 int nodes = C->unique();
101 #ifndef PRODUCT
102 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
103 char buf[1000];
104 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
105 tty->print_cr("Intrinsic %s", str);
106 }
107 #endif
108 ciMethod* callee = kit.callee();
109 const int bci = kit.bci();
110 #ifdef ASSERT
111 Node* ctrl = kit.control();
112 #endif
113 // Try to inline the intrinsic.
114 if (callee->check_intrinsic_candidate() &&
115 kit.try_to_inline(_last_predicate)) {
116 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
117 : "(intrinsic)";
118 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
119 if (C->print_intrinsics() || C->print_inlining()) {
120 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
121 }
122 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
123 if (C->log()) {
124 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
125 vmIntrinsics::name_at(intrinsic_id()),
126 (is_virtual() ? " virtual='1'" : ""),
127 C->unique() - nodes);
128 }
129 // Push the result from the inlined method onto the stack.
130 kit.push_result();
131 C->print_inlining_update(this);
132 return kit.transfer_exceptions_into_jvms();
133 }
134
135 // The intrinsic bailed out
136 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
137 if (jvms->has_method()) {
138 // Not a root compile.
139 const char* msg;
140 if (callee->intrinsic_candidate()) {
141 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
142 } else {
143 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
144 : "failed to inline (intrinsic), method not annotated";
145 }
146 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
147 if (C->print_intrinsics() || C->print_inlining()) {
148 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
149 }
150 } else {
151 // Root compile
152 ResourceMark rm;
153 stringStream msg_stream;
154 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
155 vmIntrinsics::name_at(intrinsic_id()),
156 is_virtual() ? " (virtual)" : "", bci);
157 const char *msg = msg_stream.as_string();
158 log_debug(jit, inlining)("%s", msg);
159 if (C->print_intrinsics() || C->print_inlining()) {
160 tty->print("%s", msg);
161 }
162 }
163 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
164 C->print_inlining_update(this);
165
166 return NULL;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
190 if (C->print_intrinsics() || C->print_inlining()) {
191 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
192 }
193 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
194 if (C->log()) {
195 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
196 vmIntrinsics::name_at(intrinsic_id()),
197 (is_virtual() ? " virtual='1'" : ""),
198 C->unique() - nodes);
199 }
200 return slow_ctl; // Could be NULL if the check folds.
201 }
202
203 // The intrinsic bailed out
204 if (jvms->has_method()) {
205 // Not a root compile.
206 const char* msg = "failed to generate predicate for intrinsic";
207 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
208 if (C->print_intrinsics() || C->print_inlining()) {
209 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
210 }
211 } else {
212 // Root compile
213 ResourceMark rm;
214 stringStream msg_stream;
215 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
216 vmIntrinsics::name_at(intrinsic_id()),
217 is_virtual() ? " (virtual)" : "", bci);
218 const char *msg = msg_stream.as_string();
219 log_debug(jit, inlining)("%s", msg);
220 if (C->print_intrinsics() || C->print_inlining()) {
221 C->print_inlining_stream()->print("%s", msg);
222 }
223 }
224 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
225 return NULL;
226 }
227
228 bool LibraryCallKit::try_to_inline(int predicate) {
229 // Handle symbolic names for otherwise undistinguished boolean switches:
230 const bool is_store = true;
231 const bool is_compress = true;
232 const bool is_static = true;
233 const bool is_volatile = true;
234
235 if (!jvms()->has_method()) {
236 // Root JVMState has a null method.
237 assert(map()->memory()->Opcode() == Op_Parm, "");
238 // Insert the memory aliasing node
239 set_all_memory(reset_memory());
240 }
241 assert(merged_memory(), "");
242
243 switch (intrinsic_id()) {
244 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
245 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
246 case vmIntrinsics::_getClass: return inline_native_getClass();
247
248 case vmIntrinsics::_ceil:
249 case vmIntrinsics::_floor:
250 case vmIntrinsics::_rint:
251 case vmIntrinsics::_dsin:
252 case vmIntrinsics::_dcos:
253 case vmIntrinsics::_dtan:
254 case vmIntrinsics::_dabs:
255 case vmIntrinsics::_fabs:
256 case vmIntrinsics::_iabs:
257 case vmIntrinsics::_labs:
258 case vmIntrinsics::_datan2:
259 case vmIntrinsics::_dsqrt:
260 case vmIntrinsics::_dsqrt_strict:
261 case vmIntrinsics::_dexp:
262 case vmIntrinsics::_dlog:
263 case vmIntrinsics::_dlog10:
264 case vmIntrinsics::_dpow:
265 case vmIntrinsics::_dcopySign:
266 case vmIntrinsics::_fcopySign:
267 case vmIntrinsics::_dsignum:
268 case vmIntrinsics::_roundF:
269 case vmIntrinsics::_roundD:
270 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
271
272 case vmIntrinsics::_notify:
273 case vmIntrinsics::_notifyAll:
274 return inline_notify(intrinsic_id());
275
276 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
277 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
278 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
279 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
280 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
281 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
282 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
283 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
284 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
285 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
286 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
287 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
288 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
289 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
290
291 case vmIntrinsics::_arraycopy: return inline_arraycopy();
292
293 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
294 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
295 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
296 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
297
298 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
299 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
300 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
301 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
305 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
306
307 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
308 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
309
310 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
311 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
312 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
313 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
314
315 case vmIntrinsics::_compressStringC:
316 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
317 case vmIntrinsics::_inflateStringC:
318 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
319
320 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
321 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
322 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
323 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
324 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
325 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
326 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
327 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
328 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
329
330 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
331 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
332 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
333 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
334 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
335 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
336 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
337 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
338 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
339
340 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
341 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
342 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
343 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
344 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
345 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
346 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
347 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
348 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
349
350 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
351 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
352 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
353 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
354 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
355 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
356 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
357 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
358 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
359
360 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
361 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
362 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
363 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
364
365 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
366 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
367 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
368 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
369
370 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
371 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
372 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
373 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
374 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
375 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
376 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
377 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
378 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
379
380 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
381 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
382 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
383 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
384 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
385 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
386 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
387 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
388 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
389
390 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
391 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
392 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
393 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
394 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
395 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
396 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
397 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
398 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
399
400 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
401 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
402 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
403 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
404 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
405 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
406 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
407 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
408 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
409
410 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
415
416 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
417 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
418 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
419 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
420 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
421 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
422 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
423 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
424 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
436
437 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
438 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
439 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
440 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
441 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
442 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
443 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
444 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
445 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
446 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
452
453 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
455 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
457
458 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
463
464 case vmIntrinsics::_loadFence:
465 case vmIntrinsics::_storeFence:
466 case vmIntrinsics::_storeStoreFence:
467 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
468
469 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
470
471 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
472 case vmIntrinsics::_currentThread: return inline_native_currentThread();
473 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
474
475 case vmIntrinsics::_extentLocalCache: return inline_native_extentLocalCache();
476 case vmIntrinsics::_setExtentLocalCache: return inline_native_setExtentLocalCache();
477
478 #ifdef JFR_HAVE_INTRINSICS
479 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
480 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
481 #endif
482 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
483 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
484 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
485 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
486 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
487 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
488 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
489 case vmIntrinsics::_getLength: return inline_native_getLength();
490 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
491 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
492 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
493 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
494 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
495 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
496 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
497
498 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
499 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
500
501 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
502
503 case vmIntrinsics::_isInstance:
504 case vmIntrinsics::_getModifiers:
505 case vmIntrinsics::_isInterface:
506 case vmIntrinsics::_isArray:
507 case vmIntrinsics::_isPrimitive:
508 case vmIntrinsics::_isHidden:
509 case vmIntrinsics::_getSuperclass:
510 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
511
512 case vmIntrinsics::_floatToRawIntBits:
513 case vmIntrinsics::_floatToIntBits:
514 case vmIntrinsics::_intBitsToFloat:
515 case vmIntrinsics::_doubleToRawLongBits:
516 case vmIntrinsics::_doubleToLongBits:
517 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
518
519 case vmIntrinsics::_floatIsInfinite:
520 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
521
522 case vmIntrinsics::_numberOfLeadingZeros_i:
523 case vmIntrinsics::_numberOfLeadingZeros_l:
524 case vmIntrinsics::_numberOfTrailingZeros_i:
525 case vmIntrinsics::_numberOfTrailingZeros_l:
526 case vmIntrinsics::_bitCount_i:
527 case vmIntrinsics::_bitCount_l:
528 case vmIntrinsics::_reverse_i:
529 case vmIntrinsics::_reverse_l:
530 case vmIntrinsics::_reverseBytes_i:
531 case vmIntrinsics::_reverseBytes_l:
532 case vmIntrinsics::_reverseBytes_s:
533 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
534
535 case vmIntrinsics::_compress_i:
536 case vmIntrinsics::_compress_l:
537 case vmIntrinsics::_expand_i:
538 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
539
540 case vmIntrinsics::_compareUnsigned_i:
541 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
542
543 case vmIntrinsics::_divideUnsigned_i:
544 case vmIntrinsics::_divideUnsigned_l:
545 case vmIntrinsics::_remainderUnsigned_i:
546 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
547
548 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
549
550 case vmIntrinsics::_Reference_get: return inline_reference_get();
551 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
552 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
553
554 case vmIntrinsics::_Class_cast: return inline_Class_cast();
555
556 case vmIntrinsics::_aescrypt_encryptBlock:
557 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
558
559 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
560 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
561 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
562
563 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
564 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
565 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
566
567 case vmIntrinsics::_counterMode_AESCrypt:
568 return inline_counterMode_AESCrypt(intrinsic_id());
569
570 case vmIntrinsics::_galoisCounterMode_AESCrypt:
571 return inline_galoisCounterMode_AESCrypt();
572
573 case vmIntrinsics::_md5_implCompress:
574 case vmIntrinsics::_sha_implCompress:
575 case vmIntrinsics::_sha2_implCompress:
576 case vmIntrinsics::_sha5_implCompress:
577 case vmIntrinsics::_sha3_implCompress:
578 return inline_digestBase_implCompress(intrinsic_id());
579
580 case vmIntrinsics::_digestBase_implCompressMB:
581 return inline_digestBase_implCompressMB(predicate);
582
583 case vmIntrinsics::_multiplyToLen:
584 return inline_multiplyToLen();
585
586 case vmIntrinsics::_squareToLen:
587 return inline_squareToLen();
588
589 case vmIntrinsics::_mulAdd:
590 return inline_mulAdd();
591
592 case vmIntrinsics::_montgomeryMultiply:
593 return inline_montgomeryMultiply();
594 case vmIntrinsics::_montgomerySquare:
595 return inline_montgomerySquare();
596
597 case vmIntrinsics::_bigIntegerRightShiftWorker:
598 return inline_bigIntegerShift(true);
599 case vmIntrinsics::_bigIntegerLeftShiftWorker:
600 return inline_bigIntegerShift(false);
601
602 case vmIntrinsics::_vectorizedMismatch:
603 return inline_vectorizedMismatch();
604
605 case vmIntrinsics::_ghash_processBlocks:
606 return inline_ghash_processBlocks();
607 case vmIntrinsics::_base64_encodeBlock:
608 return inline_base64_encodeBlock();
609 case vmIntrinsics::_base64_decodeBlock:
610 return inline_base64_decodeBlock();
611
612 case vmIntrinsics::_encodeISOArray:
613 case vmIntrinsics::_encodeByteISOArray:
614 return inline_encodeISOArray(false);
615 case vmIntrinsics::_encodeAsciiArray:
616 return inline_encodeISOArray(true);
617
618 case vmIntrinsics::_updateCRC32:
619 return inline_updateCRC32();
620 case vmIntrinsics::_updateBytesCRC32:
621 return inline_updateBytesCRC32();
622 case vmIntrinsics::_updateByteBufferCRC32:
623 return inline_updateByteBufferCRC32();
624
625 case vmIntrinsics::_updateBytesCRC32C:
626 return inline_updateBytesCRC32C();
627 case vmIntrinsics::_updateDirectByteBufferCRC32C:
628 return inline_updateDirectByteBufferCRC32C();
629
630 case vmIntrinsics::_updateBytesAdler32:
631 return inline_updateBytesAdler32();
632 case vmIntrinsics::_updateByteBufferAdler32:
633 return inline_updateByteBufferAdler32();
634
635 case vmIntrinsics::_profileBoolean:
636 return inline_profileBoolean();
637 case vmIntrinsics::_isCompileConstant:
638 return inline_isCompileConstant();
639
640 case vmIntrinsics::_countPositives:
641 return inline_countPositives();
642
643 case vmIntrinsics::_fmaD:
644 case vmIntrinsics::_fmaF:
645 return inline_fma(intrinsic_id());
646
647 case vmIntrinsics::_Continuation_doYield:
648 return inline_continuation_do_yield();
649
650 case vmIntrinsics::_isDigit:
651 case vmIntrinsics::_isLowerCase:
652 case vmIntrinsics::_isUpperCase:
653 case vmIntrinsics::_isWhitespace:
654 return inline_character_compare(intrinsic_id());
655
656 case vmIntrinsics::_min:
657 case vmIntrinsics::_max:
658 case vmIntrinsics::_min_strict:
659 case vmIntrinsics::_max_strict:
660 return inline_min_max(intrinsic_id());
661
662 case vmIntrinsics::_maxF:
663 case vmIntrinsics::_minF:
664 case vmIntrinsics::_maxD:
665 case vmIntrinsics::_minD:
666 case vmIntrinsics::_maxF_strict:
667 case vmIntrinsics::_minF_strict:
668 case vmIntrinsics::_maxD_strict:
669 case vmIntrinsics::_minD_strict:
670 return inline_fp_min_max(intrinsic_id());
671
672 case vmIntrinsics::_VectorUnaryOp:
673 return inline_vector_nary_operation(1);
674 case vmIntrinsics::_VectorBinaryOp:
675 return inline_vector_nary_operation(2);
676 case vmIntrinsics::_VectorTernaryOp:
677 return inline_vector_nary_operation(3);
678 case vmIntrinsics::_VectorFromBitsCoerced:
679 return inline_vector_frombits_coerced();
680 case vmIntrinsics::_VectorShuffleIota:
681 return inline_vector_shuffle_iota();
682 case vmIntrinsics::_VectorMaskOp:
683 return inline_vector_mask_operation();
684 case vmIntrinsics::_VectorShuffleToVector:
685 return inline_vector_shuffle_to_vector();
686 case vmIntrinsics::_VectorLoadOp:
687 return inline_vector_mem_operation(/*is_store=*/false);
688 case vmIntrinsics::_VectorLoadMaskedOp:
689 return inline_vector_mem_masked_operation(/*is_store*/false);
690 case vmIntrinsics::_VectorStoreOp:
691 return inline_vector_mem_operation(/*is_store=*/true);
692 case vmIntrinsics::_VectorStoreMaskedOp:
693 return inline_vector_mem_masked_operation(/*is_store=*/true);
694 case vmIntrinsics::_VectorGatherOp:
695 return inline_vector_gather_scatter(/*is_scatter*/ false);
696 case vmIntrinsics::_VectorScatterOp:
697 return inline_vector_gather_scatter(/*is_scatter*/ true);
698 case vmIntrinsics::_VectorReductionCoerced:
699 return inline_vector_reduction();
700 case vmIntrinsics::_VectorTest:
701 return inline_vector_test();
702 case vmIntrinsics::_VectorBlend:
703 return inline_vector_blend();
704 case vmIntrinsics::_VectorRearrange:
705 return inline_vector_rearrange();
706 case vmIntrinsics::_VectorCompare:
707 return inline_vector_compare();
708 case vmIntrinsics::_VectorBroadcastInt:
709 return inline_vector_broadcast_int();
710 case vmIntrinsics::_VectorConvert:
711 return inline_vector_convert();
712 case vmIntrinsics::_VectorInsert:
713 return inline_vector_insert();
714 case vmIntrinsics::_VectorExtract:
715 return inline_vector_extract();
716 case vmIntrinsics::_VectorCompressExpand:
717 return inline_vector_compress_expand();
718
719 case vmIntrinsics::_getObjectSize:
720 return inline_getObjectSize();
721
722 case vmIntrinsics::_blackhole:
723 return inline_blackhole();
724
725 case vmIntrinsics::_shipilev_magic_timestamp:
726 return inline_timestamp(false);
727 case vmIntrinsics::_shipilev_magic_timestamp_serial:
728 return inline_timestamp(true);
729 case vmIntrinsics::_shipilev_magic_sizeOf:
730 return inline_sizeOf();
731 case vmIntrinsics::_shipilev_magic_addressOf:
732 return inline_addressOf();
733
734 default:
735 // If you get here, it may be that someone has added a new intrinsic
736 // to the list in vmIntrinsics.hpp without implementing it here.
737 #ifndef PRODUCT
738 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
739 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
740 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
741 }
742 #endif
743 return false;
744 }
745 }
746
747 Node* LibraryCallKit::try_to_predicate(int predicate) {
748 if (!jvms()->has_method()) {
749 // Root JVMState has a null method.
750 assert(map()->memory()->Opcode() == Op_Parm, "");
751 // Insert the memory aliasing node
752 set_all_memory(reset_memory());
753 }
754 assert(merged_memory(), "");
755
756 switch (intrinsic_id()) {
757 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
758 return inline_cipherBlockChaining_AESCrypt_predicate(false);
759 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
760 return inline_cipherBlockChaining_AESCrypt_predicate(true);
761 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
762 return inline_electronicCodeBook_AESCrypt_predicate(false);
763 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
764 return inline_electronicCodeBook_AESCrypt_predicate(true);
765 case vmIntrinsics::_counterMode_AESCrypt:
766 return inline_counterMode_AESCrypt_predicate();
767 case vmIntrinsics::_digestBase_implCompressMB:
768 return inline_digestBase_implCompressMB_predicate(predicate);
769 case vmIntrinsics::_galoisCounterMode_AESCrypt:
770 return inline_galoisCounterMode_AESCrypt_predicate();
771
772 default:
773 // If you get here, it may be that someone has added a new intrinsic
774 // to the list in vmIntrinsics.hpp without implementing it here.
775 #ifndef PRODUCT
776 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
777 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
778 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
779 }
780 #endif
781 Node* slow_ctl = control();
782 set_control(top()); // No fast path intrinsic
783 return slow_ctl;
784 }
785 }
786
787 //------------------------------set_result-------------------------------
788 // Helper function for finishing intrinsics.
789 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
790 record_for_igvn(region);
791 set_control(_gvn.transform(region));
792 set_result( _gvn.transform(value));
793 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
794 }
795
796 //------------------------------generate_guard---------------------------
797 // Helper function for generating guarded fast-slow graph structures.
798 // The given 'test', if true, guards a slow path. If the test fails
799 // then a fast path can be taken. (We generally hope it fails.)
800 // In all cases, GraphKit::control() is updated to the fast path.
801 // The returned value represents the control for the slow path.
802 // The return value is never 'top'; it is either a valid control
803 // or NULL if it is obvious that the slow path can never be taken.
804 // Also, if region and the slow control are not NULL, the slow edge
805 // is appended to the region.
806 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
807 if (stopped()) {
808 // Already short circuited.
809 return NULL;
810 }
811
812 // Build an if node and its projections.
813 // If test is true we take the slow path, which we assume is uncommon.
814 if (_gvn.type(test) == TypeInt::ZERO) {
815 // The slow branch is never taken. No need to build this guard.
816 return NULL;
817 }
818
819 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
820
821 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
822 if (if_slow == top()) {
823 // The slow branch is never taken. No need to build this guard.
824 return NULL;
825 }
826
827 if (region != NULL)
828 region->add_req(if_slow);
829
830 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
831 set_control(if_fast);
832
833 return if_slow;
834 }
835
836 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
837 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
838 }
839 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
840 return generate_guard(test, region, PROB_FAIR);
841 }
842
843 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
844 Node* *pos_index) {
845 if (stopped())
846 return NULL; // already stopped
847 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
848 return NULL; // index is already adequately typed
849 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
850 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
851 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
852 if (is_neg != NULL && pos_index != NULL) {
853 // Emulate effect of Parse::adjust_map_after_if.
854 Node* ccast = new CastIINode(index, TypeInt::POS);
855 ccast->set_req(0, control());
856 (*pos_index) = _gvn.transform(ccast);
857 }
858 return is_neg;
859 }
860
861 // Make sure that 'position' is a valid limit index, in [0..length].
862 // There are two equivalent plans for checking this:
863 // A. (offset + copyLength) unsigned<= arrayLength
864 // B. offset <= (arrayLength - copyLength)
865 // We require that all of the values above, except for the sum and
866 // difference, are already known to be non-negative.
867 // Plan A is robust in the face of overflow, if offset and copyLength
868 // are both hugely positive.
869 //
870 // Plan B is less direct and intuitive, but it does not overflow at
871 // all, since the difference of two non-negatives is always
872 // representable. Whenever Java methods must perform the equivalent
873 // check they generally use Plan B instead of Plan A.
874 // For the moment we use Plan A.
875 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
876 Node* subseq_length,
877 Node* array_length,
878 RegionNode* region) {
879 if (stopped())
880 return NULL; // already stopped
881 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
882 if (zero_offset && subseq_length->eqv_uncast(array_length))
883 return NULL; // common case of whole-array copy
884 Node* last = subseq_length;
885 if (!zero_offset) // last += offset
886 last = _gvn.transform(new AddINode(last, offset));
887 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
888 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
889 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
890 return is_over;
891 }
892
893 // Emit range checks for the given String.value byte array
894 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
895 if (stopped()) {
896 return; // already stopped
897 }
898 RegionNode* bailout = new RegionNode(1);
899 record_for_igvn(bailout);
900 if (char_count) {
901 // Convert char count to byte count
902 count = _gvn.transform(new LShiftINode(count, intcon(1)));
903 }
904
905 // Offset and count must not be negative
906 generate_negative_guard(offset, bailout);
907 generate_negative_guard(count, bailout);
908 // Offset + count must not exceed length of array
909 generate_limit_guard(offset, count, load_array_length(array), bailout);
910
911 if (bailout->req() > 1) {
912 PreserveJVMState pjvms(this);
913 set_control(_gvn.transform(bailout));
914 uncommon_trap(Deoptimization::Reason_intrinsic,
915 Deoptimization::Action_maybe_recompile);
916 }
917 }
918
919 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
920 bool is_immutable) {
921 ciKlass* thread_klass = env()->Thread_klass();
922 const Type* thread_type
923 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
924
925 Node* thread = _gvn.transform(new ThreadLocalNode());
926 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
927 tls_output = thread;
928
929 Node* thread_obj_handle
930 = (is_immutable
931 ? LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
932 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
933 : make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
934 thread_obj_handle = _gvn.transform(thread_obj_handle);
935
936 DecoratorSet decorators = IN_NATIVE;
937 if (is_immutable) {
938 decorators |= C2_IMMUTABLE_MEMORY;
939 }
940 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
941 }
942
943 //--------------------------generate_current_thread--------------------
944 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
945 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
946 /*is_immutable*/false);
947 }
948
949 //--------------------------generate_virtual_thread--------------------
950 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
951 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
952 !C->method()->changes_current_thread());
953 }
954
955 //------------------------------make_string_method_node------------------------
956 // Helper method for String intrinsic functions. This version is called with
957 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
958 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
959 // containing the lengths of str1 and str2.
960 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
961 Node* result = NULL;
962 switch (opcode) {
963 case Op_StrIndexOf:
964 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
965 str1_start, cnt1, str2_start, cnt2, ae);
966 break;
967 case Op_StrComp:
968 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
969 str1_start, cnt1, str2_start, cnt2, ae);
970 break;
971 case Op_StrEquals:
972 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
973 // Use the constant length if there is one because optimized match rule may exist.
974 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
975 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
976 break;
977 default:
978 ShouldNotReachHere();
979 return NULL;
980 }
981
982 // All these intrinsics have checks.
983 C->set_has_split_ifs(true); // Has chance for split-if optimization
984 clear_upper_avx();
985
986 return _gvn.transform(result);
987 }
988
989 //------------------------------inline_string_compareTo------------------------
990 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
991 Node* arg1 = argument(0);
992 Node* arg2 = argument(1);
993
994 arg1 = must_be_not_null(arg1, true);
995 arg2 = must_be_not_null(arg2, true);
996
997 // Get start addr and length of first argument
998 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
999 Node* arg1_cnt = load_array_length(arg1);
1000
1001 // Get start addr and length of second argument
1002 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1003 Node* arg2_cnt = load_array_length(arg2);
1004
1005 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1006 set_result(result);
1007 return true;
1008 }
1009
1010 //------------------------------inline_string_equals------------------------
1011 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1012 Node* arg1 = argument(0);
1013 Node* arg2 = argument(1);
1014
1015 // paths (plus control) merge
1016 RegionNode* region = new RegionNode(3);
1017 Node* phi = new PhiNode(region, TypeInt::BOOL);
1018
1019 if (!stopped()) {
1020
1021 arg1 = must_be_not_null(arg1, true);
1022 arg2 = must_be_not_null(arg2, true);
1023
1024 // Get start addr and length of first argument
1025 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1026 Node* arg1_cnt = load_array_length(arg1);
1027
1028 // Get start addr and length of second argument
1029 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1030 Node* arg2_cnt = load_array_length(arg2);
1031
1032 // Check for arg1_cnt != arg2_cnt
1033 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1034 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1035 Node* if_ne = generate_slow_guard(bol, NULL);
1036 if (if_ne != NULL) {
1037 phi->init_req(2, intcon(0));
1038 region->init_req(2, if_ne);
1039 }
1040
1041 // Check for count == 0 is done by assembler code for StrEquals.
1042
1043 if (!stopped()) {
1044 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1045 phi->init_req(1, equals);
1046 region->init_req(1, control());
1047 }
1048 }
1049
1050 // post merge
1051 set_control(_gvn.transform(region));
1052 record_for_igvn(region);
1053
1054 set_result(_gvn.transform(phi));
1055 return true;
1056 }
1057
1058 //------------------------------inline_array_equals----------------------------
1059 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1060 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1061 Node* arg1 = argument(0);
1062 Node* arg2 = argument(1);
1063
1064 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1065 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1066 clear_upper_avx();
1067
1068 return true;
1069 }
1070
1071 //------------------------------inline_countPositives------------------------------
1072 bool LibraryCallKit::inline_countPositives() {
1073 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1074 return false;
1075 }
1076
1077 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1078 // no receiver since it is static method
1079 Node* ba = argument(0);
1080 Node* offset = argument(1);
1081 Node* len = argument(2);
1082
1083 ba = must_be_not_null(ba, true);
1084
1085 // Range checks
1086 generate_string_range_check(ba, offset, len, false);
1087 if (stopped()) {
1088 return true;
1089 }
1090 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1091 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1092 set_result(_gvn.transform(result));
1093 return true;
1094 }
1095
1096 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1097 Node* index = argument(0);
1098 Node* length = bt == T_INT ? argument(1) : argument(2);
1099 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1100 return false;
1101 }
1102
1103 // check that length is positive
1104 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1105 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1106
1107 {
1108 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1109 uncommon_trap(Deoptimization::Reason_intrinsic,
1110 Deoptimization::Action_make_not_entrant);
1111 }
1112
1113 if (stopped()) {
1114 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1115 return true;
1116 }
1117
1118 // length is now known positive, add a cast node to make this explicit
1119 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1120 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1121 casted_length = _gvn.transform(casted_length);
1122 replace_in_map(length, casted_length);
1123 length = casted_length;
1124
1125 // Use an unsigned comparison for the range check itself
1126 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1127 BoolTest::mask btest = BoolTest::lt;
1128 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1129 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1130 _gvn.set_type(rc, rc->Value(&_gvn));
1131 if (!rc_bool->is_Con()) {
1132 record_for_igvn(rc);
1133 }
1134 set_control(_gvn.transform(new IfTrueNode(rc)));
1135 {
1136 PreserveJVMState pjvms(this);
1137 set_control(_gvn.transform(new IfFalseNode(rc)));
1138 uncommon_trap(Deoptimization::Reason_range_check,
1139 Deoptimization::Action_make_not_entrant);
1140 }
1141
1142 if (stopped()) {
1143 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1144 return true;
1145 }
1146
1147 // index is now known to be >= 0 and < length, cast it
1148 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1149 result = _gvn.transform(result);
1150 set_result(result);
1151 replace_in_map(index, result);
1152 return true;
1153 }
1154
1155 //------------------------------inline_string_indexOf------------------------
1156 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1157 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1158 return false;
1159 }
1160 Node* src = argument(0);
1161 Node* tgt = argument(1);
1162
1163 // Make the merge point
1164 RegionNode* result_rgn = new RegionNode(4);
1165 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1166
1167 src = must_be_not_null(src, true);
1168 tgt = must_be_not_null(tgt, true);
1169
1170 // Get start addr and length of source string
1171 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1172 Node* src_count = load_array_length(src);
1173
1174 // Get start addr and length of substring
1175 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1176 Node* tgt_count = load_array_length(tgt);
1177
1178 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1179 // Divide src size by 2 if String is UTF16 encoded
1180 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1181 }
1182 if (ae == StrIntrinsicNode::UU) {
1183 // Divide substring size by 2 if String is UTF16 encoded
1184 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1185 }
1186
1187 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1188 if (result != NULL) {
1189 result_phi->init_req(3, result);
1190 result_rgn->init_req(3, control());
1191 }
1192 set_control(_gvn.transform(result_rgn));
1193 record_for_igvn(result_rgn);
1194 set_result(_gvn.transform(result_phi));
1195
1196 return true;
1197 }
1198
1199 //-----------------------------inline_string_indexOf-----------------------
1200 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1201 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1202 return false;
1203 }
1204 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1205 return false;
1206 }
1207 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1208 Node* src = argument(0); // byte[]
1209 Node* src_count = argument(1); // char count
1210 Node* tgt = argument(2); // byte[]
1211 Node* tgt_count = argument(3); // char count
1212 Node* from_index = argument(4); // char index
1213
1214 src = must_be_not_null(src, true);
1215 tgt = must_be_not_null(tgt, true);
1216
1217 // Multiply byte array index by 2 if String is UTF16 encoded
1218 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1219 src_count = _gvn.transform(new SubINode(src_count, from_index));
1220 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1221 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1222
1223 // Range checks
1224 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1225 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1226 if (stopped()) {
1227 return true;
1228 }
1229
1230 RegionNode* region = new RegionNode(5);
1231 Node* phi = new PhiNode(region, TypeInt::INT);
1232
1233 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1234 if (result != NULL) {
1235 // The result is index relative to from_index if substring was found, -1 otherwise.
1236 // Generate code which will fold into cmove.
1237 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1238 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1239
1240 Node* if_lt = generate_slow_guard(bol, NULL);
1241 if (if_lt != NULL) {
1242 // result == -1
1243 phi->init_req(3, result);
1244 region->init_req(3, if_lt);
1245 }
1246 if (!stopped()) {
1247 result = _gvn.transform(new AddINode(result, from_index));
1248 phi->init_req(4, result);
1249 region->init_req(4, control());
1250 }
1251 }
1252
1253 set_control(_gvn.transform(region));
1254 record_for_igvn(region);
1255 set_result(_gvn.transform(phi));
1256 clear_upper_avx();
1257
1258 return true;
1259 }
1260
1261 // Create StrIndexOfNode with fast path checks
1262 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1263 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1264 // Check for substr count > string count
1265 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1266 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1267 Node* if_gt = generate_slow_guard(bol, NULL);
1268 if (if_gt != NULL) {
1269 phi->init_req(1, intcon(-1));
1270 region->init_req(1, if_gt);
1271 }
1272 if (!stopped()) {
1273 // Check for substr count == 0
1274 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1275 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1276 Node* if_zero = generate_slow_guard(bol, NULL);
1277 if (if_zero != NULL) {
1278 phi->init_req(2, intcon(0));
1279 region->init_req(2, if_zero);
1280 }
1281 }
1282 if (!stopped()) {
1283 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1284 }
1285 return NULL;
1286 }
1287
1288 //-----------------------------inline_string_indexOfChar-----------------------
1289 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1290 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1291 return false;
1292 }
1293 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1294 return false;
1295 }
1296 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1297 Node* src = argument(0); // byte[]
1298 Node* tgt = argument(1); // tgt is int ch
1299 Node* from_index = argument(2);
1300 Node* max = argument(3);
1301
1302 src = must_be_not_null(src, true);
1303
1304 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1305 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1306 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1307
1308 // Range checks
1309 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1310 if (stopped()) {
1311 return true;
1312 }
1313
1314 RegionNode* region = new RegionNode(3);
1315 Node* phi = new PhiNode(region, TypeInt::INT);
1316
1317 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1318 C->set_has_split_ifs(true); // Has chance for split-if optimization
1319 _gvn.transform(result);
1320
1321 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1322 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1323
1324 Node* if_lt = generate_slow_guard(bol, NULL);
1325 if (if_lt != NULL) {
1326 // result == -1
1327 phi->init_req(2, result);
1328 region->init_req(2, if_lt);
1329 }
1330 if (!stopped()) {
1331 result = _gvn.transform(new AddINode(result, from_index));
1332 phi->init_req(1, result);
1333 region->init_req(1, control());
1334 }
1335 set_control(_gvn.transform(region));
1336 record_for_igvn(region);
1337 set_result(_gvn.transform(phi));
1338
1339 return true;
1340 }
1341 //---------------------------inline_string_copy---------------------
1342 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1343 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1344 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1345 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1346 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1347 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1348 bool LibraryCallKit::inline_string_copy(bool compress) {
1349 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1350 return false;
1351 }
1352 int nargs = 5; // 2 oops, 3 ints
1353 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1354
1355 Node* src = argument(0);
1356 Node* src_offset = argument(1);
1357 Node* dst = argument(2);
1358 Node* dst_offset = argument(3);
1359 Node* length = argument(4);
1360
1361 // Check for allocation before we add nodes that would confuse
1362 // tightly_coupled_allocation()
1363 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1364
1365 // Figure out the size and type of the elements we will be copying.
1366 const Type* src_type = src->Value(&_gvn);
1367 const Type* dst_type = dst->Value(&_gvn);
1368 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
1369 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
1370 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1371 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1372 "Unsupported array types for inline_string_copy");
1373
1374 src = must_be_not_null(src, true);
1375 dst = must_be_not_null(dst, true);
1376
1377 // Convert char[] offsets to byte[] offsets
1378 bool convert_src = (compress && src_elem == T_BYTE);
1379 bool convert_dst = (!compress && dst_elem == T_BYTE);
1380 if (convert_src) {
1381 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1382 } else if (convert_dst) {
1383 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1384 }
1385
1386 // Range checks
1387 generate_string_range_check(src, src_offset, length, convert_src);
1388 generate_string_range_check(dst, dst_offset, length, convert_dst);
1389 if (stopped()) {
1390 return true;
1391 }
1392
1393 Node* src_start = array_element_address(src, src_offset, src_elem);
1394 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1395 // 'src_start' points to src array + scaled offset
1396 // 'dst_start' points to dst array + scaled offset
1397 Node* count = NULL;
1398 if (compress) {
1399 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1400 } else {
1401 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1402 }
1403
1404 if (alloc != NULL) {
1405 if (alloc->maybe_set_complete(&_gvn)) {
1406 // "You break it, you buy it."
1407 InitializeNode* init = alloc->initialization();
1408 assert(init->is_complete(), "we just did this");
1409 init->set_complete_with_arraycopy();
1410 assert(dst->is_CheckCastPP(), "sanity");
1411 assert(dst->in(0)->in(0) == init, "dest pinned");
1412 }
1413 // Do not let stores that initialize this object be reordered with
1414 // a subsequent store that would make this object accessible by
1415 // other threads.
1416 // Record what AllocateNode this StoreStore protects so that
1417 // escape analysis can go from the MemBarStoreStoreNode to the
1418 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1419 // based on the escape status of the AllocateNode.
1420 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1421 }
1422 if (compress) {
1423 set_result(_gvn.transform(count));
1424 }
1425 clear_upper_avx();
1426
1427 return true;
1428 }
1429
1430 #ifdef _LP64
1431 #define XTOP ,top() /*additional argument*/
1432 #else //_LP64
1433 #define XTOP /*no additional argument*/
1434 #endif //_LP64
1435
1436 //------------------------inline_string_toBytesU--------------------------
1437 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1438 bool LibraryCallKit::inline_string_toBytesU() {
1439 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1440 return false;
1441 }
1442 // Get the arguments.
1443 Node* value = argument(0);
1444 Node* offset = argument(1);
1445 Node* length = argument(2);
1446
1447 Node* newcopy = NULL;
1448
1449 // Set the original stack and the reexecute bit for the interpreter to reexecute
1450 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1451 { PreserveReexecuteState preexecs(this);
1452 jvms()->set_should_reexecute(true);
1453
1454 // Check if a null path was taken unconditionally.
1455 value = null_check(value);
1456
1457 RegionNode* bailout = new RegionNode(1);
1458 record_for_igvn(bailout);
1459
1460 // Range checks
1461 generate_negative_guard(offset, bailout);
1462 generate_negative_guard(length, bailout);
1463 generate_limit_guard(offset, length, load_array_length(value), bailout);
1464 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1465 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1466
1467 if (bailout->req() > 1) {
1468 PreserveJVMState pjvms(this);
1469 set_control(_gvn.transform(bailout));
1470 uncommon_trap(Deoptimization::Reason_intrinsic,
1471 Deoptimization::Action_maybe_recompile);
1472 }
1473 if (stopped()) {
1474 return true;
1475 }
1476
1477 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1478 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1479 newcopy = new_array(klass_node, size, 0); // no arguments to push
1480 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1481 guarantee(alloc != NULL, "created above");
1482
1483 // Calculate starting addresses.
1484 Node* src_start = array_element_address(value, offset, T_CHAR);
1485 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1486
1487 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1488 const TypeInt* toffset = gvn().type(offset)->is_int();
1489 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1490
1491 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1492 const char* copyfunc_name = "arraycopy";
1493 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1494 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1495 OptoRuntime::fast_arraycopy_Type(),
1496 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1497 src_start, dst_start, ConvI2X(length) XTOP);
1498 // Do not let reads from the cloned object float above the arraycopy.
1499 if (alloc->maybe_set_complete(&_gvn)) {
1500 // "You break it, you buy it."
1501 InitializeNode* init = alloc->initialization();
1502 assert(init->is_complete(), "we just did this");
1503 init->set_complete_with_arraycopy();
1504 assert(newcopy->is_CheckCastPP(), "sanity");
1505 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1506 }
1507 // Do not let stores that initialize this object be reordered with
1508 // a subsequent store that would make this object accessible by
1509 // other threads.
1510 // Record what AllocateNode this StoreStore protects so that
1511 // escape analysis can go from the MemBarStoreStoreNode to the
1512 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1513 // based on the escape status of the AllocateNode.
1514 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1515 } // original reexecute is set back here
1516
1517 C->set_has_split_ifs(true); // Has chance for split-if optimization
1518 if (!stopped()) {
1519 set_result(newcopy);
1520 }
1521 clear_upper_avx();
1522
1523 return true;
1524 }
1525
1526 //------------------------inline_string_getCharsU--------------------------
1527 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1528 bool LibraryCallKit::inline_string_getCharsU() {
1529 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1530 return false;
1531 }
1532
1533 // Get the arguments.
1534 Node* src = argument(0);
1535 Node* src_begin = argument(1);
1536 Node* src_end = argument(2); // exclusive offset (i < src_end)
1537 Node* dst = argument(3);
1538 Node* dst_begin = argument(4);
1539
1540 // Check for allocation before we add nodes that would confuse
1541 // tightly_coupled_allocation()
1542 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1543
1544 // Check if a null path was taken unconditionally.
1545 src = null_check(src);
1546 dst = null_check(dst);
1547 if (stopped()) {
1548 return true;
1549 }
1550
1551 // Get length and convert char[] offset to byte[] offset
1552 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1553 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1554
1555 // Range checks
1556 generate_string_range_check(src, src_begin, length, true);
1557 generate_string_range_check(dst, dst_begin, length, false);
1558 if (stopped()) {
1559 return true;
1560 }
1561
1562 if (!stopped()) {
1563 // Calculate starting addresses.
1564 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1565 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1566
1567 // Check if array addresses are aligned to HeapWordSize
1568 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1569 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1570 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1571 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1572
1573 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1574 const char* copyfunc_name = "arraycopy";
1575 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1576 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1577 OptoRuntime::fast_arraycopy_Type(),
1578 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1579 src_start, dst_start, ConvI2X(length) XTOP);
1580 // Do not let reads from the cloned object float above the arraycopy.
1581 if (alloc != NULL) {
1582 if (alloc->maybe_set_complete(&_gvn)) {
1583 // "You break it, you buy it."
1584 InitializeNode* init = alloc->initialization();
1585 assert(init->is_complete(), "we just did this");
1586 init->set_complete_with_arraycopy();
1587 assert(dst->is_CheckCastPP(), "sanity");
1588 assert(dst->in(0)->in(0) == init, "dest pinned");
1589 }
1590 // Do not let stores that initialize this object be reordered with
1591 // a subsequent store that would make this object accessible by
1592 // other threads.
1593 // Record what AllocateNode this StoreStore protects so that
1594 // escape analysis can go from the MemBarStoreStoreNode to the
1595 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1596 // based on the escape status of the AllocateNode.
1597 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1598 } else {
1599 insert_mem_bar(Op_MemBarCPUOrder);
1600 }
1601 }
1602
1603 C->set_has_split_ifs(true); // Has chance for split-if optimization
1604 return true;
1605 }
1606
1607 //----------------------inline_string_char_access----------------------------
1608 // Store/Load char to/from byte[] array.
1609 // static void StringUTF16.putChar(byte[] val, int index, int c)
1610 // static char StringUTF16.getChar(byte[] val, int index)
1611 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1612 Node* value = argument(0);
1613 Node* index = argument(1);
1614 Node* ch = is_store ? argument(2) : NULL;
1615
1616 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1617 // correctly requires matched array shapes.
1618 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1619 "sanity: byte[] and char[] bases agree");
1620 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1621 "sanity: byte[] and char[] scales agree");
1622
1623 // Bail when getChar over constants is requested: constant folding would
1624 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1625 // Java method would constant fold nicely instead.
1626 if (!is_store && value->is_Con() && index->is_Con()) {
1627 return false;
1628 }
1629
1630 value = must_be_not_null(value, true);
1631
1632 Node* adr = array_element_address(value, index, T_CHAR);
1633 if (adr->is_top()) {
1634 return false;
1635 }
1636 if (is_store) {
1637 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1638 } else {
1639 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
1640 set_result(ch);
1641 }
1642 return true;
1643 }
1644
1645 //--------------------------round_double_node--------------------------------
1646 // Round a double node if necessary.
1647 Node* LibraryCallKit::round_double_node(Node* n) {
1648 if (Matcher::strict_fp_requires_explicit_rounding) {
1649 #ifdef IA32
1650 if (UseSSE < 2) {
1651 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1652 }
1653 #else
1654 Unimplemented();
1655 #endif // IA32
1656 }
1657 return n;
1658 }
1659
1660 //------------------------------inline_math-----------------------------------
1661 // public static double Math.abs(double)
1662 // public static double Math.sqrt(double)
1663 // public static double Math.log(double)
1664 // public static double Math.log10(double)
1665 // public static double Math.round(double)
1666 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1667 Node* arg = round_double_node(argument(0));
1668 Node* n = NULL;
1669 switch (id) {
1670 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1671 case vmIntrinsics::_dsqrt:
1672 case vmIntrinsics::_dsqrt_strict:
1673 n = new SqrtDNode(C, control(), arg); break;
1674 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1675 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1676 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1677 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1678 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1679 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1680 default: fatal_unexpected_iid(id); break;
1681 }
1682 set_result(_gvn.transform(n));
1683 return true;
1684 }
1685
1686 //------------------------------inline_math-----------------------------------
1687 // public static float Math.abs(float)
1688 // public static int Math.abs(int)
1689 // public static long Math.abs(long)
1690 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1691 Node* arg = argument(0);
1692 Node* n = NULL;
1693 switch (id) {
1694 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1695 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1696 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1697 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1698 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1699 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1700 default: fatal_unexpected_iid(id); break;
1701 }
1702 set_result(_gvn.transform(n));
1703 return true;
1704 }
1705
1706 //------------------------------runtime_math-----------------------------
1707 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1708 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1709 "must be (DD)D or (D)D type");
1710
1711 // Inputs
1712 Node* a = round_double_node(argument(0));
1713 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1714
1715 const TypePtr* no_memory_effects = NULL;
1716 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1717 no_memory_effects,
1718 a, top(), b, b ? top() : NULL);
1719 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1720 #ifdef ASSERT
1721 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1722 assert(value_top == top(), "second value must be top");
1723 #endif
1724
1725 set_result(value);
1726 return true;
1727 }
1728
1729 //------------------------------inline_math_pow-----------------------------
1730 bool LibraryCallKit::inline_math_pow() {
1731 Node* exp = round_double_node(argument(2));
1732 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1733 if (d != NULL) {
1734 if (d->getd() == 2.0) {
1735 // Special case: pow(x, 2.0) => x * x
1736 Node* base = round_double_node(argument(0));
1737 set_result(_gvn.transform(new MulDNode(base, base)));
1738 return true;
1739 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1740 // Special case: pow(x, 0.5) => sqrt(x)
1741 Node* base = round_double_node(argument(0));
1742 Node* zero = _gvn.zerocon(T_DOUBLE);
1743
1744 RegionNode* region = new RegionNode(3);
1745 Node* phi = new PhiNode(region, Type::DOUBLE);
1746
1747 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1748 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1749 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1750 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1751 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1752
1753 Node* if_pow = generate_slow_guard(test, NULL);
1754 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1755 phi->init_req(1, value_sqrt);
1756 region->init_req(1, control());
1757
1758 if (if_pow != NULL) {
1759 set_control(if_pow);
1760 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1761 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1762 const TypePtr* no_memory_effects = NULL;
1763 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1764 no_memory_effects, base, top(), exp, top());
1765 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1766 #ifdef ASSERT
1767 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1768 assert(value_top == top(), "second value must be top");
1769 #endif
1770 phi->init_req(2, value_pow);
1771 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1772 }
1773
1774 C->set_has_split_ifs(true); // Has chance for split-if optimization
1775 set_control(_gvn.transform(region));
1776 record_for_igvn(region);
1777 set_result(_gvn.transform(phi));
1778
1779 return true;
1780 }
1781 }
1782
1783 return StubRoutines::dpow() != NULL ?
1784 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1785 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1786 }
1787
1788 //------------------------------inline_math_native-----------------------------
1789 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1790 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1791 switch (id) {
1792 // These intrinsics are not properly supported on all hardware
1793 case vmIntrinsics::_dsin:
1794 return StubRoutines::dsin() != NULL ?
1795 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1796 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1797 case vmIntrinsics::_dcos:
1798 return StubRoutines::dcos() != NULL ?
1799 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1800 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1801 case vmIntrinsics::_dtan:
1802 return StubRoutines::dtan() != NULL ?
1803 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1804 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1805 case vmIntrinsics::_dlog:
1806 return StubRoutines::dlog() != NULL ?
1807 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1808 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1809 case vmIntrinsics::_dlog10:
1810 return StubRoutines::dlog10() != NULL ?
1811 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1812 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1813
1814 // These intrinsics are supported on all hardware
1815 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1816 case vmIntrinsics::_ceil:
1817 case vmIntrinsics::_floor:
1818 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1819
1820 case vmIntrinsics::_dsqrt:
1821 case vmIntrinsics::_dsqrt_strict:
1822 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1823 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1824 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1825 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1826 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1827
1828 case vmIntrinsics::_dexp:
1829 return StubRoutines::dexp() != NULL ?
1830 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1831 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1832 #undef FN_PTR
1833
1834 case vmIntrinsics::_dpow: return inline_math_pow();
1835 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1836 case vmIntrinsics::_fcopySign: return inline_math(id);
1837 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1838 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1839 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1840
1841 // These intrinsics are not yet correctly implemented
1842 case vmIntrinsics::_datan2:
1843 return false;
1844
1845 default:
1846 fatal_unexpected_iid(id);
1847 return false;
1848 }
1849 }
1850
1851 //----------------------------inline_notify-----------------------------------*
1852 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1853 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1854 address func;
1855 if (id == vmIntrinsics::_notify) {
1856 func = OptoRuntime::monitor_notify_Java();
1857 } else {
1858 func = OptoRuntime::monitor_notifyAll_Java();
1859 }
1860 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1861 make_slow_call_ex(call, env()->Throwable_klass(), false);
1862 return true;
1863 }
1864
1865
1866 //----------------------------inline_min_max-----------------------------------
1867 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1868 set_result(generate_min_max(id, argument(0), argument(1)));
1869 return true;
1870 }
1871
1872 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1873 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1874 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1875 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1876 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1877
1878 {
1879 PreserveJVMState pjvms(this);
1880 PreserveReexecuteState preexecs(this);
1881 jvms()->set_should_reexecute(true);
1882
1883 set_control(slow_path);
1884 set_i_o(i_o());
1885
1886 uncommon_trap(Deoptimization::Reason_intrinsic,
1887 Deoptimization::Action_none);
1888 }
1889
1890 set_control(fast_path);
1891 set_result(math);
1892 }
1893
1894 template <typename OverflowOp>
1895 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1896 typedef typename OverflowOp::MathOp MathOp;
1897
1898 MathOp* mathOp = new MathOp(arg1, arg2);
1899 Node* operation = _gvn.transform( mathOp );
1900 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1901 inline_math_mathExact(operation, ofcheck);
1902 return true;
1903 }
1904
1905 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1906 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1907 }
1908
1909 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1910 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1911 }
1912
1913 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1914 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1915 }
1916
1917 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1918 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1919 }
1920
1921 bool LibraryCallKit::inline_math_negateExactI() {
1922 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1923 }
1924
1925 bool LibraryCallKit::inline_math_negateExactL() {
1926 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1927 }
1928
1929 bool LibraryCallKit::inline_math_multiplyExactI() {
1930 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1931 }
1932
1933 bool LibraryCallKit::inline_math_multiplyExactL() {
1934 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1935 }
1936
1937 bool LibraryCallKit::inline_math_multiplyHigh() {
1938 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1939 return true;
1940 }
1941
1942 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1943 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1944 return true;
1945 }
1946
1947 Node*
1948 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1949 Node* result_val = NULL;
1950 switch (id) {
1951 case vmIntrinsics::_min:
1952 case vmIntrinsics::_min_strict:
1953 result_val = _gvn.transform(new MinINode(x0, y0));
1954 break;
1955 case vmIntrinsics::_max:
1956 case vmIntrinsics::_max_strict:
1957 result_val = _gvn.transform(new MaxINode(x0, y0));
1958 break;
1959 default:
1960 fatal_unexpected_iid(id);
1961 break;
1962 }
1963 return result_val;
1964 }
1965
1966 inline int
1967 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
1968 const TypePtr* base_type = TypePtr::NULL_PTR;
1969 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
1970 if (base_type == NULL) {
1971 // Unknown type.
1972 return Type::AnyPtr;
1973 } else if (base_type == TypePtr::NULL_PTR) {
1974 // Since this is a NULL+long form, we have to switch to a rawptr.
1975 base = _gvn.transform(new CastX2PNode(offset));
1976 offset = MakeConX(0);
1977 return Type::RawPtr;
1978 } else if (base_type->base() == Type::RawPtr) {
1979 return Type::RawPtr;
1980 } else if (base_type->isa_oopptr()) {
1981 // Base is never null => always a heap address.
1982 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
1983 return Type::OopPtr;
1984 }
1985 // Offset is small => always a heap address.
1986 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
1987 if (offset_type != NULL &&
1988 base_type->offset() == 0 && // (should always be?)
1989 offset_type->_lo >= 0 &&
1990 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
1991 return Type::OopPtr;
1992 } else if (type == T_OBJECT) {
1993 // off heap access to an oop doesn't make any sense. Has to be on
1994 // heap.
1995 return Type::OopPtr;
1996 }
1997 // Otherwise, it might either be oop+off or NULL+addr.
1998 return Type::AnyPtr;
1999 } else {
2000 // No information:
2001 return Type::AnyPtr;
2002 }
2003 }
2004
2005 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2006 Node* uncasted_base = base;
2007 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2008 if (kind == Type::RawPtr) {
2009 return basic_plus_adr(top(), uncasted_base, offset);
2010 } else if (kind == Type::AnyPtr) {
2011 assert(base == uncasted_base, "unexpected base change");
2012 if (can_cast) {
2013 if (!_gvn.type(base)->speculative_maybe_null() &&
2014 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2015 // According to profiling, this access is always on
2016 // heap. Casting the base to not null and thus avoiding membars
2017 // around the access should allow better optimizations
2018 Node* null_ctl = top();
2019 base = null_check_oop(base, &null_ctl, true, true, true);
2020 assert(null_ctl->is_top(), "no null control here");
2021 return basic_plus_adr(base, offset);
2022 } else if (_gvn.type(base)->speculative_always_null() &&
2023 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2024 // According to profiling, this access is always off
2025 // heap.
2026 base = null_assert(base);
2027 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2028 offset = MakeConX(0);
2029 return basic_plus_adr(top(), raw_base, offset);
2030 }
2031 }
2032 // We don't know if it's an on heap or off heap access. Fall back
2033 // to raw memory access.
2034 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2035 return basic_plus_adr(top(), raw, offset);
2036 } else {
2037 assert(base == uncasted_base, "unexpected base change");
2038 // We know it's an on heap access so base can't be null
2039 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2040 base = must_be_not_null(base, true);
2041 }
2042 return basic_plus_adr(base, offset);
2043 }
2044 }
2045
2046 //--------------------------inline_number_methods-----------------------------
2047 // inline int Integer.numberOfLeadingZeros(int)
2048 // inline int Long.numberOfLeadingZeros(long)
2049 //
2050 // inline int Integer.numberOfTrailingZeros(int)
2051 // inline int Long.numberOfTrailingZeros(long)
2052 //
2053 // inline int Integer.bitCount(int)
2054 // inline int Long.bitCount(long)
2055 //
2056 // inline char Character.reverseBytes(char)
2057 // inline short Short.reverseBytes(short)
2058 // inline int Integer.reverseBytes(int)
2059 // inline long Long.reverseBytes(long)
2060 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2061 Node* arg = argument(0);
2062 Node* n = NULL;
2063 switch (id) {
2064 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2065 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2066 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2067 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2068 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2069 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2070 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2071 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2072 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2073 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2074 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2075 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2076 default: fatal_unexpected_iid(id); break;
2077 }
2078 set_result(_gvn.transform(n));
2079 return true;
2080 }
2081
2082 //--------------------------inline_bitshuffle_methods-----------------------------
2083 // inline int Integer.compress(int, int)
2084 // inline int Integer.expand(int, int)
2085 // inline long Long.compress(long, long)
2086 // inline long Long.expand(long, long)
2087 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2088 Node* n = NULL;
2089 switch (id) {
2090 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2091 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2092 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2093 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2094 default: fatal_unexpected_iid(id); break;
2095 }
2096 set_result(_gvn.transform(n));
2097 return true;
2098 }
2099
2100 //--------------------------inline_number_methods-----------------------------
2101 // inline int Integer.compareUnsigned(int, int)
2102 // inline int Long.compareUnsigned(long, long)
2103 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2104 Node* arg1 = argument(0);
2105 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2106 Node* n = NULL;
2107 switch (id) {
2108 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2109 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2110 default: fatal_unexpected_iid(id); break;
2111 }
2112 set_result(_gvn.transform(n));
2113 return true;
2114 }
2115
2116 //--------------------------inline_unsigned_divmod_methods-----------------------------
2117 // inline int Integer.divideUnsigned(int, int)
2118 // inline int Integer.remainderUnsigned(int, int)
2119 // inline long Long.divideUnsigned(long, long)
2120 // inline long Long.remainderUnsigned(long, long)
2121 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2122 Node* n = NULL;
2123 switch (id) {
2124 case vmIntrinsics::_divideUnsigned_i: {
2125 zero_check_int(argument(1));
2126 // Compile-time detect of null-exception
2127 if (stopped()) {
2128 return true; // keep the graph constructed so far
2129 }
2130 n = new UDivINode(control(), argument(0), argument(1));
2131 break;
2132 }
2133 case vmIntrinsics::_divideUnsigned_l: {
2134 zero_check_long(argument(2));
2135 // Compile-time detect of null-exception
2136 if (stopped()) {
2137 return true; // keep the graph constructed so far
2138 }
2139 n = new UDivLNode(control(), argument(0), argument(2));
2140 break;
2141 }
2142 case vmIntrinsics::_remainderUnsigned_i: {
2143 zero_check_int(argument(1));
2144 // Compile-time detect of null-exception
2145 if (stopped()) {
2146 return true; // keep the graph constructed so far
2147 }
2148 n = new UModINode(control(), argument(0), argument(1));
2149 break;
2150 }
2151 case vmIntrinsics::_remainderUnsigned_l: {
2152 zero_check_long(argument(2));
2153 // Compile-time detect of null-exception
2154 if (stopped()) {
2155 return true; // keep the graph constructed so far
2156 }
2157 n = new UModLNode(control(), argument(0), argument(2));
2158 break;
2159 }
2160 default: fatal_unexpected_iid(id); break;
2161 }
2162 set_result(_gvn.transform(n));
2163 return true;
2164 }
2165
2166 //----------------------------inline_unsafe_access----------------------------
2167
2168 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2169 // Attempt to infer a sharper value type from the offset and base type.
2170 ciKlass* sharpened_klass = NULL;
2171
2172 // See if it is an instance field, with an object type.
2173 if (alias_type->field() != NULL) {
2174 if (alias_type->field()->type()->is_klass()) {
2175 sharpened_klass = alias_type->field()->type()->as_klass();
2176 }
2177 }
2178
2179 const TypeOopPtr* result = NULL;
2180 // See if it is a narrow oop array.
2181 if (adr_type->isa_aryptr()) {
2182 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2183 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2184 if (elem_type != NULL && elem_type->is_loaded()) {
2185 // Sharpen the value type.
2186 result = elem_type;
2187 }
2188 }
2189 }
2190
2191 // The sharpened class might be unloaded if there is no class loader
2192 // contraint in place.
2193 if (result == NULL && sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2194 // Sharpen the value type.
2195 result = TypeOopPtr::make_from_klass(sharpened_klass);
2196 }
2197 if (result != NULL) {
2198 #ifndef PRODUCT
2199 if (C->print_intrinsics() || C->print_inlining()) {
2200 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2201 tty->print(" sharpened value: "); result->dump(); tty->cr();
2202 }
2203 #endif
2204 }
2205 return result;
2206 }
2207
2208 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2209 switch (kind) {
2210 case Relaxed:
2211 return MO_UNORDERED;
2212 case Opaque:
2213 return MO_RELAXED;
2214 case Acquire:
2215 return MO_ACQUIRE;
2216 case Release:
2217 return MO_RELEASE;
2218 case Volatile:
2219 return MO_SEQ_CST;
2220 default:
2221 ShouldNotReachHere();
2222 return 0;
2223 }
2224 }
2225
2226 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2227 if (callee()->is_static()) return false; // caller must have the capability!
2228 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2229 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2230 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2231 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2232
2233 if (is_reference_type(type)) {
2234 decorators |= ON_UNKNOWN_OOP_REF;
2235 }
2236
2237 if (unaligned) {
2238 decorators |= C2_UNALIGNED;
2239 }
2240
2241 #ifndef PRODUCT
2242 {
2243 ResourceMark rm;
2244 // Check the signatures.
2245 ciSignature* sig = callee()->signature();
2246 #ifdef ASSERT
2247 if (!is_store) {
2248 // Object getReference(Object base, int/long offset), etc.
2249 BasicType rtype = sig->return_type()->basic_type();
2250 assert(rtype == type, "getter must return the expected value");
2251 assert(sig->count() == 2, "oop getter has 2 arguments");
2252 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2253 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2254 } else {
2255 // void putReference(Object base, int/long offset, Object x), etc.
2256 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2257 assert(sig->count() == 3, "oop putter has 3 arguments");
2258 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2259 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2260 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2261 assert(vtype == type, "putter must accept the expected value");
2262 }
2263 #endif // ASSERT
2264 }
2265 #endif //PRODUCT
2266
2267 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2268
2269 Node* receiver = argument(0); // type: oop
2270
2271 // Build address expression.
2272 Node* heap_base_oop = top();
2273
2274 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2275 Node* base = argument(1); // type: oop
2276 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2277 Node* offset = argument(2); // type: long
2278 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2279 // to be plain byte offsets, which are also the same as those accepted
2280 // by oopDesc::field_addr.
2281 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2282 "fieldOffset must be byte-scaled");
2283 // 32-bit machines ignore the high half!
2284 offset = ConvL2X(offset);
2285
2286 // Save state and restore on bailout
2287 uint old_sp = sp();
2288 SafePointNode* old_map = clone_map();
2289
2290 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2291
2292 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2293 if (type != T_OBJECT) {
2294 decorators |= IN_NATIVE; // off-heap primitive access
2295 } else {
2296 set_map(old_map);
2297 set_sp(old_sp);
2298 return false; // off-heap oop accesses are not supported
2299 }
2300 } else {
2301 heap_base_oop = base; // on-heap or mixed access
2302 }
2303
2304 // Can base be NULL? Otherwise, always on-heap access.
2305 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2306
2307 if (!can_access_non_heap) {
2308 decorators |= IN_HEAP;
2309 }
2310
2311 Node* val = is_store ? argument(4) : NULL;
2312
2313 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2314 if (adr_type == TypePtr::NULL_PTR) {
2315 set_map(old_map);
2316 set_sp(old_sp);
2317 return false; // off-heap access with zero address
2318 }
2319
2320 // Try to categorize the address.
2321 Compile::AliasType* alias_type = C->alias_type(adr_type);
2322 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2323
2324 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2325 alias_type->adr_type() == TypeAryPtr::RANGE) {
2326 set_map(old_map);
2327 set_sp(old_sp);
2328 return false; // not supported
2329 }
2330
2331 bool mismatched = false;
2332 BasicType bt = alias_type->basic_type();
2333 if (bt != T_ILLEGAL) {
2334 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2335 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2336 // Alias type doesn't differentiate between byte[] and boolean[]).
2337 // Use address type to get the element type.
2338 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2339 }
2340 if (is_reference_type(bt, true)) {
2341 // accessing an array field with getReference is not a mismatch
2342 bt = T_OBJECT;
2343 }
2344 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2345 // Don't intrinsify mismatched object accesses
2346 set_map(old_map);
2347 set_sp(old_sp);
2348 return false;
2349 }
2350 mismatched = (bt != type);
2351 } else if (alias_type->adr_type()->isa_oopptr()) {
2352 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2353 }
2354
2355 old_map->destruct(&_gvn);
2356 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2357
2358 if (mismatched) {
2359 decorators |= C2_MISMATCHED;
2360 }
2361
2362 // First guess at the value type.
2363 const Type *value_type = Type::get_const_basic_type(type);
2364
2365 // Figure out the memory ordering.
2366 decorators |= mo_decorator_for_access_kind(kind);
2367
2368 if (!is_store && type == T_OBJECT) {
2369 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2370 if (tjp != NULL) {
2371 value_type = tjp;
2372 }
2373 }
2374
2375 receiver = null_check(receiver);
2376 if (stopped()) {
2377 return true;
2378 }
2379 // Heap pointers get a null-check from the interpreter,
2380 // as a courtesy. However, this is not guaranteed by Unsafe,
2381 // and it is not possible to fully distinguish unintended nulls
2382 // from intended ones in this API.
2383
2384 if (!is_store) {
2385 Node* p = NULL;
2386 // Try to constant fold a load from a constant field
2387 ciField* field = alias_type->field();
2388 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2389 // final or stable field
2390 p = make_constant_from_field(field, heap_base_oop);
2391 }
2392
2393 if (p == NULL) { // Could not constant fold the load
2394 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2395 // Normalize the value returned by getBoolean in the following cases
2396 if (type == T_BOOLEAN &&
2397 (mismatched ||
2398 heap_base_oop == top() || // - heap_base_oop is NULL or
2399 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2400 // and the unsafe access is made to large offset
2401 // (i.e., larger than the maximum offset necessary for any
2402 // field access)
2403 ) {
2404 IdealKit ideal = IdealKit(this);
2405 #define __ ideal.
2406 IdealVariable normalized_result(ideal);
2407 __ declarations_done();
2408 __ set(normalized_result, p);
2409 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2410 __ set(normalized_result, ideal.ConI(1));
2411 ideal.end_if();
2412 final_sync(ideal);
2413 p = __ value(normalized_result);
2414 #undef __
2415 }
2416 }
2417 if (type == T_ADDRESS) {
2418 p = gvn().transform(new CastP2XNode(NULL, p));
2419 p = ConvX2UL(p);
2420 }
2421 // The load node has the control of the preceding MemBarCPUOrder. All
2422 // following nodes will have the control of the MemBarCPUOrder inserted at
2423 // the end of this method. So, pushing the load onto the stack at a later
2424 // point is fine.
2425 set_result(p);
2426 } else {
2427 if (bt == T_ADDRESS) {
2428 // Repackage the long as a pointer.
2429 val = ConvL2X(val);
2430 val = gvn().transform(new CastX2PNode(val));
2431 }
2432 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2433 }
2434
2435 return true;
2436 }
2437
2438 //----------------------------inline_unsafe_load_store----------------------------
2439 // This method serves a couple of different customers (depending on LoadStoreKind):
2440 //
2441 // LS_cmp_swap:
2442 //
2443 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2444 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2445 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2446 //
2447 // LS_cmp_swap_weak:
2448 //
2449 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2450 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2451 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2452 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2453 //
2454 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2455 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2456 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2457 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2458 //
2459 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2460 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2461 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2462 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2463 //
2464 // LS_cmp_exchange:
2465 //
2466 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2467 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2468 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2469 //
2470 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2471 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2472 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2473 //
2474 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2475 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2476 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2477 //
2478 // LS_get_add:
2479 //
2480 // int getAndAddInt( Object o, long offset, int delta)
2481 // long getAndAddLong(Object o, long offset, long delta)
2482 //
2483 // LS_get_set:
2484 //
2485 // int getAndSet(Object o, long offset, int newValue)
2486 // long getAndSet(Object o, long offset, long newValue)
2487 // Object getAndSet(Object o, long offset, Object newValue)
2488 //
2489 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2490 // This basic scheme here is the same as inline_unsafe_access, but
2491 // differs in enough details that combining them would make the code
2492 // overly confusing. (This is a true fact! I originally combined
2493 // them, but even I was confused by it!) As much code/comments as
2494 // possible are retained from inline_unsafe_access though to make
2495 // the correspondences clearer. - dl
2496
2497 if (callee()->is_static()) return false; // caller must have the capability!
2498
2499 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2500 decorators |= mo_decorator_for_access_kind(access_kind);
2501
2502 #ifndef PRODUCT
2503 BasicType rtype;
2504 {
2505 ResourceMark rm;
2506 // Check the signatures.
2507 ciSignature* sig = callee()->signature();
2508 rtype = sig->return_type()->basic_type();
2509 switch(kind) {
2510 case LS_get_add:
2511 case LS_get_set: {
2512 // Check the signatures.
2513 #ifdef ASSERT
2514 assert(rtype == type, "get and set must return the expected type");
2515 assert(sig->count() == 3, "get and set has 3 arguments");
2516 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2517 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2518 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2519 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2520 #endif // ASSERT
2521 break;
2522 }
2523 case LS_cmp_swap:
2524 case LS_cmp_swap_weak: {
2525 // Check the signatures.
2526 #ifdef ASSERT
2527 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2528 assert(sig->count() == 4, "CAS has 4 arguments");
2529 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2530 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2531 #endif // ASSERT
2532 break;
2533 }
2534 case LS_cmp_exchange: {
2535 // Check the signatures.
2536 #ifdef ASSERT
2537 assert(rtype == type, "CAS must return the expected type");
2538 assert(sig->count() == 4, "CAS has 4 arguments");
2539 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2540 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2541 #endif // ASSERT
2542 break;
2543 }
2544 default:
2545 ShouldNotReachHere();
2546 }
2547 }
2548 #endif //PRODUCT
2549
2550 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2551
2552 // Get arguments:
2553 Node* receiver = NULL;
2554 Node* base = NULL;
2555 Node* offset = NULL;
2556 Node* oldval = NULL;
2557 Node* newval = NULL;
2558 switch(kind) {
2559 case LS_cmp_swap:
2560 case LS_cmp_swap_weak:
2561 case LS_cmp_exchange: {
2562 const bool two_slot_type = type2size[type] == 2;
2563 receiver = argument(0); // type: oop
2564 base = argument(1); // type: oop
2565 offset = argument(2); // type: long
2566 oldval = argument(4); // type: oop, int, or long
2567 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2568 break;
2569 }
2570 case LS_get_add:
2571 case LS_get_set: {
2572 receiver = argument(0); // type: oop
2573 base = argument(1); // type: oop
2574 offset = argument(2); // type: long
2575 oldval = NULL;
2576 newval = argument(4); // type: oop, int, or long
2577 break;
2578 }
2579 default:
2580 ShouldNotReachHere();
2581 }
2582
2583 // Build field offset expression.
2584 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2585 // to be plain byte offsets, which are also the same as those accepted
2586 // by oopDesc::field_addr.
2587 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2588 // 32-bit machines ignore the high half of long offsets
2589 offset = ConvL2X(offset);
2590 // Save state and restore on bailout
2591 uint old_sp = sp();
2592 SafePointNode* old_map = clone_map();
2593 Node* adr = make_unsafe_address(base, offset,type, false);
2594 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2595
2596 Compile::AliasType* alias_type = C->alias_type(adr_type);
2597 BasicType bt = alias_type->basic_type();
2598 if (bt != T_ILLEGAL &&
2599 (is_reference_type(bt) != (type == T_OBJECT))) {
2600 // Don't intrinsify mismatched object accesses.
2601 set_map(old_map);
2602 set_sp(old_sp);
2603 return false;
2604 }
2605
2606 old_map->destruct(&_gvn);
2607
2608 // For CAS, unlike inline_unsafe_access, there seems no point in
2609 // trying to refine types. Just use the coarse types here.
2610 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2611 const Type *value_type = Type::get_const_basic_type(type);
2612
2613 switch (kind) {
2614 case LS_get_set:
2615 case LS_cmp_exchange: {
2616 if (type == T_OBJECT) {
2617 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2618 if (tjp != NULL) {
2619 value_type = tjp;
2620 }
2621 }
2622 break;
2623 }
2624 case LS_cmp_swap:
2625 case LS_cmp_swap_weak:
2626 case LS_get_add:
2627 break;
2628 default:
2629 ShouldNotReachHere();
2630 }
2631
2632 // Null check receiver.
2633 receiver = null_check(receiver);
2634 if (stopped()) {
2635 return true;
2636 }
2637
2638 int alias_idx = C->get_alias_index(adr_type);
2639
2640 if (is_reference_type(type)) {
2641 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2642
2643 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2644 // could be delayed during Parse (for example, in adjust_map_after_if()).
2645 // Execute transformation here to avoid barrier generation in such case.
2646 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2647 newval = _gvn.makecon(TypePtr::NULL_PTR);
2648
2649 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2650 // Refine the value to a null constant, when it is known to be null
2651 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2652 }
2653 }
2654
2655 Node* result = NULL;
2656 switch (kind) {
2657 case LS_cmp_exchange: {
2658 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2659 oldval, newval, value_type, type, decorators);
2660 break;
2661 }
2662 case LS_cmp_swap_weak:
2663 decorators |= C2_WEAK_CMPXCHG;
2664 case LS_cmp_swap: {
2665 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2666 oldval, newval, value_type, type, decorators);
2667 break;
2668 }
2669 case LS_get_set: {
2670 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2671 newval, value_type, type, decorators);
2672 break;
2673 }
2674 case LS_get_add: {
2675 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2676 newval, value_type, type, decorators);
2677 break;
2678 }
2679 default:
2680 ShouldNotReachHere();
2681 }
2682
2683 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2684 set_result(result);
2685 return true;
2686 }
2687
2688 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2689 // Regardless of form, don't allow previous ld/st to move down,
2690 // then issue acquire, release, or volatile mem_bar.
2691 insert_mem_bar(Op_MemBarCPUOrder);
2692 switch(id) {
2693 case vmIntrinsics::_loadFence:
2694 insert_mem_bar(Op_LoadFence);
2695 return true;
2696 case vmIntrinsics::_storeFence:
2697 insert_mem_bar(Op_StoreFence);
2698 return true;
2699 case vmIntrinsics::_storeStoreFence:
2700 insert_mem_bar(Op_StoreStoreFence);
2701 return true;
2702 case vmIntrinsics::_fullFence:
2703 insert_mem_bar(Op_MemBarVolatile);
2704 return true;
2705 default:
2706 fatal_unexpected_iid(id);
2707 return false;
2708 }
2709 }
2710
2711 bool LibraryCallKit::inline_onspinwait() {
2712 insert_mem_bar(Op_OnSpinWait);
2713 return true;
2714 }
2715
2716 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2717 if (!kls->is_Con()) {
2718 return true;
2719 }
2720 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2721 if (klsptr == NULL) {
2722 return true;
2723 }
2724 ciInstanceKlass* ik = klsptr->instance_klass();
2725 // don't need a guard for a klass that is already initialized
2726 return !ik->is_initialized();
2727 }
2728
2729 //----------------------------inline_unsafe_writeback0-------------------------
2730 // public native void Unsafe.writeback0(long address)
2731 bool LibraryCallKit::inline_unsafe_writeback0() {
2732 if (!Matcher::has_match_rule(Op_CacheWB)) {
2733 return false;
2734 }
2735 #ifndef PRODUCT
2736 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2737 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2738 ciSignature* sig = callee()->signature();
2739 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2740 #endif
2741 null_check_receiver(); // null-check, then ignore
2742 Node *addr = argument(1);
2743 addr = new CastX2PNode(addr);
2744 addr = _gvn.transform(addr);
2745 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2746 flush = _gvn.transform(flush);
2747 set_memory(flush, TypeRawPtr::BOTTOM);
2748 return true;
2749 }
2750
2751 //----------------------------inline_unsafe_writeback0-------------------------
2752 // public native void Unsafe.writeback0(long address)
2753 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2754 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2755 return false;
2756 }
2757 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2758 return false;
2759 }
2760 #ifndef PRODUCT
2761 assert(Matcher::has_match_rule(Op_CacheWB),
2762 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2763 : "found match rule for CacheWBPostSync but not CacheWB"));
2764
2765 #endif
2766 null_check_receiver(); // null-check, then ignore
2767 Node *sync;
2768 if (is_pre) {
2769 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2770 } else {
2771 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2772 }
2773 sync = _gvn.transform(sync);
2774 set_memory(sync, TypeRawPtr::BOTTOM);
2775 return true;
2776 }
2777
2778 //----------------------------inline_unsafe_allocate---------------------------
2779 // public native Object Unsafe.allocateInstance(Class<?> cls);
2780 bool LibraryCallKit::inline_unsafe_allocate() {
2781 if (callee()->is_static()) return false; // caller must have the capability!
2782
2783 null_check_receiver(); // null-check, then ignore
2784 Node* cls = null_check(argument(1));
2785 if (stopped()) return true;
2786
2787 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2788 kls = null_check(kls);
2789 if (stopped()) return true; // argument was like int.class
2790
2791 Node* test = NULL;
2792 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2793 // Note: The argument might still be an illegal value like
2794 // Serializable.class or Object[].class. The runtime will handle it.
2795 // But we must make an explicit check for initialization.
2796 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2797 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2798 // can generate code to load it as unsigned byte.
2799 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2800 Node* bits = intcon(InstanceKlass::fully_initialized);
2801 test = _gvn.transform(new SubINode(inst, bits));
2802 // The 'test' is non-zero if we need to take a slow path.
2803 }
2804
2805 Node* obj = new_instance(kls, test);
2806 set_result(obj);
2807 return true;
2808 }
2809
2810 //------------------------inline_native_time_funcs--------------
2811 // inline code for System.currentTimeMillis() and System.nanoTime()
2812 // these have the same type and signature
2813 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2814 const TypeFunc* tf = OptoRuntime::void_long_Type();
2815 const TypePtr* no_memory_effects = NULL;
2816 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2817 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2818 #ifdef ASSERT
2819 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2820 assert(value_top == top(), "second value must be top");
2821 #endif
2822 set_result(value);
2823 return true;
2824 }
2825
2826 #ifdef JFR_HAVE_INTRINSICS
2827
2828 /**
2829 * if oop->klass != null
2830 * // normal class
2831 * epoch = _epoch_state ? 2 : 1
2832 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2833 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2834 * }
2835 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2836 * else
2837 * // primitive class
2838 * if oop->array_klass != null
2839 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2840 * else
2841 * id = LAST_TYPE_ID + 1 // void class path
2842 * if (!signaled)
2843 * signaled = true
2844 */
2845 bool LibraryCallKit::inline_native_classID() {
2846 Node* cls = argument(0);
2847
2848 IdealKit ideal(this);
2849 #define __ ideal.
2850 IdealVariable result(ideal); __ declarations_done();
2851 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2852 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2853 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2854
2855
2856 __ if_then(kls, BoolTest::ne, null()); {
2857 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2858 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2859
2860 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
2861 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2862 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2863 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2864 mask = _gvn.transform(new OrLNode(mask, epoch));
2865 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2866
2867 float unlikely = PROB_UNLIKELY(0.999);
2868 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
2869 sync_kit(ideal);
2870 make_runtime_call(RC_LEAF,
2871 OptoRuntime::class_id_load_barrier_Type(),
2872 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
2873 "class id load barrier",
2874 TypePtr::BOTTOM,
2875 kls);
2876 ideal.sync_kit(this);
2877 } __ end_if();
2878
2879 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
2880 } __ else_(); {
2881 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2882 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
2883 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2884 __ if_then(array_kls, BoolTest::ne, null()); {
2885 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2886 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2887 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
2888 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
2889 } __ else_(); {
2890 // void class case
2891 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
2892 } __ end_if();
2893
2894 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
2895 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
2896 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
2897 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
2898 } __ end_if();
2899 } __ end_if();
2900
2901 final_sync(ideal);
2902 set_result(ideal.value(result));
2903 #undef __
2904 return true;
2905 }
2906
2907 /*
2908 * The intrinsic is a model of this pseudo-code:
2909 *
2910 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
2911 * jobject h_event_writer = tl->java_event_writer();
2912 * if (h_event_writer == NULL) {
2913 * return NULL;
2914 * }
2915 * oop threadObj = Thread::threadObj();
2916 * oop vthread = java_lang_Thread::vthread(threadObj);
2917 * traceid tid;
2918 * bool excluded;
2919 * if (vthread != threadObj) { // i.e. current thread is virtual
2920 * tid = java_lang_Thread::tid(vthread);
2921 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
2922 * excluded = vthread_epoch_raw & excluded_mask;
2923 * if (!excluded) {
2924 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
2925 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
2926 * if (vthread_epoch != current_epoch) {
2927 * write_checkpoint();
2928 * }
2929 * }
2930 * } else {
2931 * tid = java_lang_Thread::tid(threadObj);
2932 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
2933 * excluded = thread_epoch_raw & excluded_mask;
2934 * }
2935 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
2936 * traceid tid_in_event_writer = getField(event_writer, "threadID");
2937 * if (tid_in_event_writer != tid) {
2938 * setField(event_writer, "threadID", tid);
2939 * setField(event_writer, "excluded", excluded);
2940 * }
2941 * return event_writer
2942 */
2943 bool LibraryCallKit::inline_native_getEventWriter() {
2944 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
2945
2946 // Save input memory and i_o state.
2947 Node* input_memory_state = reset_memory();
2948 set_all_memory(input_memory_state);
2949 Node* input_io_state = i_o();
2950
2951 Node* excluded_mask = _gvn.intcon(32768);
2952 Node* epoch_mask = _gvn.intcon(32767);
2953
2954 // TLS
2955 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2956
2957 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
2958 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2959
2960 // Load the eventwriter jobject handle.
2961 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2962
2963 // Null check the jobject handle.
2964 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
2965 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
2966 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
2967
2968 // False path, jobj is null.
2969 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
2970
2971 // True path, jobj is not null.
2972 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
2973
2974 set_control(jobj_is_not_null);
2975
2976 // Load the threadObj for the CarrierThread.
2977 Node* threadObj = generate_current_thread(tls_ptr);
2978
2979 // Load the vthread.
2980 Node* vthread = generate_virtual_thread(tls_ptr);
2981
2982 // If vthread != threadObj, this is a virtual thread.
2983 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
2984 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
2985 IfNode* iff_vthread_not_equal_threadObj =
2986 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
2987
2988 // False branch, fallback to threadObj.
2989 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
2990 set_control(vthread_equal_threadObj);
2991
2992 // Load the tid field from the vthread object.
2993 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
2994
2995 // Load the raw epoch value from the threadObj.
2996 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
2997 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
2998 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
2999
3000 // Mask off the excluded information from the epoch.
3001 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3002
3003 // True branch, this is a virtual thread.
3004 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3005 set_control(vthread_not_equal_threadObj);
3006
3007 // Load the tid field from the vthread object.
3008 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3009
3010 // Load the raw epoch value from the vthread.
3011 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3012 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3013 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3014
3015 // Mask off the excluded information from the epoch.
3016 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3017
3018 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3019 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3020 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3021 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3022
3023 // False branch, vthread is excluded, no need to write epoch info.
3024 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3025
3026 // True branch, vthread is included, update epoch info.
3027 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3028 set_control(included);
3029
3030 // Get epoch value.
3031 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3032
3033 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3034 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3035 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3036
3037 // Compare the epoch in the vthread to the current epoch generation.
3038 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3039 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3040 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3041
3042 // False path, epoch is equal, checkpoint information is valid.
3043 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3044
3045 // True path, epoch is not equal, write a checkpoint for the vthread.
3046 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3047
3048 set_control(epoch_is_not_equal);
3049
3050 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3051 // The call also updates the native thread local thread id and the vthread with the current epoch.
3052 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3053 OptoRuntime::jfr_write_checkpoint_Type(),
3054 StubRoutines::jfr_write_checkpoint(),
3055 "write_checkpoint", TypePtr::BOTTOM);
3056 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3057
3058 // vthread epoch != current epoch
3059 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3060 record_for_igvn(epoch_compare_rgn);
3061 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3062 record_for_igvn(epoch_compare_mem);
3063 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3064 record_for_igvn(epoch_compare_io);
3065
3066 // Update control and phi nodes.
3067 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3068 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3069 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3070 epoch_compare_mem->init_req(_false_path, input_memory_state);
3071 epoch_compare_io->init_req(_true_path, i_o());
3072 epoch_compare_io->init_req(_false_path, input_io_state);
3073
3074 // excluded != true
3075 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3076 record_for_igvn(exclude_compare_rgn);
3077 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3078 record_for_igvn(exclude_compare_mem);
3079 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3080 record_for_igvn(exclude_compare_io);
3081
3082 // Update control and phi nodes.
3083 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3084 exclude_compare_rgn->init_req(_false_path, excluded);
3085 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3086 exclude_compare_mem->init_req(_false_path, input_memory_state);
3087 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3088 exclude_compare_io->init_req(_false_path, input_io_state);
3089
3090 // vthread != threadObj
3091 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3092 record_for_igvn(vthread_compare_rgn);
3093 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3094 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3095 record_for_igvn(vthread_compare_io);
3096 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3097 record_for_igvn(tid);
3098 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3099 record_for_igvn(exclusion);
3100
3101 // Update control and phi nodes.
3102 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3103 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3104 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3105 vthread_compare_mem->init_req(_false_path, input_memory_state);
3106 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3107 vthread_compare_io->init_req(_false_path, input_io_state);
3108 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3109 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3110 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3111 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3112
3113 // Update branch state.
3114 set_control(_gvn.transform(vthread_compare_rgn));
3115 set_all_memory(_gvn.transform(vthread_compare_mem));
3116 set_i_o(_gvn.transform(vthread_compare_io));
3117
3118 // Load the event writer oop by dereferencing the jobject handle.
3119 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3120 assert(klass_EventWriter->is_loaded(), "invariant");
3121 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3122 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3123 const TypeOopPtr* const xtype = aklass->as_instance_type();
3124 Node* event_writer = access_load(jobj, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3125
3126 // Load the current thread id from the event writer object.
3127 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3128 // Get the field offset to, conditionally, store an updated tid value later.
3129 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3130 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3131 // Get the field offset to, conditionally, store an updated exclusion value later.
3132 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3133 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3134
3135 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3136 record_for_igvn(event_writer_tid_compare_rgn);
3137 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3138 record_for_igvn(event_writer_tid_compare_mem);
3139 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3140 record_for_igvn(event_writer_tid_compare_io);
3141
3142 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3143 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3144 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3145 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3146
3147 // False path, tids are the same.
3148 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3149
3150 // True path, tid is not equal, need to update the tid in the event writer.
3151 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3152 record_for_igvn(tid_is_not_equal);
3153
3154 // Store the exclusion state to the event writer.
3155 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3156
3157 // Store the tid to the event writer.
3158 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3159
3160 // Update control and phi nodes.
3161 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3162 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3163 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3164 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3165 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3166 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3167
3168 // Result of top level CFG, Memory, IO and Value.
3169 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3170 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3171 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3172 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3173
3174 // Result control.
3175 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3176 result_rgn->init_req(_false_path, jobj_is_null);
3177
3178 // Result memory.
3179 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3180 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3181
3182 // Result IO.
3183 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3184 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3185
3186 // Result value.
3187 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3188 result_value->init_req(_false_path, null()); // return NULL
3189
3190 // Set output state.
3191 set_control(_gvn.transform(result_rgn));
3192 set_all_memory(_gvn.transform(result_mem));
3193 set_i_o(_gvn.transform(result_io));
3194 set_result(result_rgn, result_value);
3195 return true;
3196 }
3197
3198 /*
3199 * The intrinsic is a model of this pseudo-code:
3200 *
3201 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3202 * if (carrierThread != thread) { // is virtual thread
3203 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3204 * bool excluded = vthread_epoch_raw & excluded_mask;
3205 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3206 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3207 * if (!excluded) {
3208 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3209 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3210 * }
3211 * Atomic::release_store(&tl->_vthread, true);
3212 * return;
3213 * }
3214 * Atomic::release_store(&tl->_vthread, false);
3215 */
3216 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3217 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3218
3219 Node* input_memory_state = reset_memory();
3220 set_all_memory(input_memory_state);
3221
3222 Node* excluded_mask = _gvn.intcon(32768);
3223 Node* epoch_mask = _gvn.intcon(32767);
3224
3225 Node* const carrierThread = generate_current_thread(jt);
3226 // If thread != carrierThread, this is a virtual thread.
3227 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3228 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3229 IfNode* iff_thread_not_equal_carrierThread =
3230 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3231
3232 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3233
3234 // False branch, is carrierThread.
3235 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3236 // Store release
3237 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3238
3239 set_all_memory(input_memory_state);
3240
3241 // True branch, is virtual thread.
3242 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3243 set_control(thread_not_equal_carrierThread);
3244
3245 // Load the raw epoch value from the vthread.
3246 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3247 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3248 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3249
3250 // Mask off the excluded information from the epoch.
3251 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3252
3253 // Load the tid field from the thread.
3254 Node* tid = load_field_from_object(thread, "tid", "J");
3255
3256 // Store the vthread tid to the jfr thread local.
3257 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3258 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3259
3260 // Branch is_excluded to conditionalize updating the epoch .
3261 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3262 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3263 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3264
3265 // True branch, vthread is excluded, no need to write epoch info.
3266 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3267 set_control(excluded);
3268 Node* vthread_is_excluded = _gvn.intcon(1);
3269
3270 // False branch, vthread is included, update epoch info.
3271 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3272 set_control(included);
3273 Node* vthread_is_included = _gvn.intcon(0);
3274
3275 // Get epoch value.
3276 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3277
3278 // Store the vthread epoch to the jfr thread local.
3279 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3280 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3281
3282 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3283 record_for_igvn(excluded_rgn);
3284 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3285 record_for_igvn(excluded_mem);
3286 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3287 record_for_igvn(exclusion);
3288
3289 // Merge the excluded control and memory.
3290 excluded_rgn->init_req(_true_path, excluded);
3291 excluded_rgn->init_req(_false_path, included);
3292 excluded_mem->init_req(_true_path, tid_memory);
3293 excluded_mem->init_req(_false_path, included_memory);
3294 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3295 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3296
3297 // Set intermediate state.
3298 set_control(_gvn.transform(excluded_rgn));
3299 set_all_memory(excluded_mem);
3300
3301 // Store the vthread exclusion state to the jfr thread local.
3302 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3303 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3304
3305 // Store release
3306 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3307
3308 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3309 record_for_igvn(thread_compare_rgn);
3310 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3311 record_for_igvn(thread_compare_mem);
3312 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3313 record_for_igvn(vthread);
3314
3315 // Merge the thread_compare control and memory.
3316 thread_compare_rgn->init_req(_true_path, control());
3317 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3318 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3319 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3320
3321 // Set output state.
3322 set_control(_gvn.transform(thread_compare_rgn));
3323 set_all_memory(_gvn.transform(thread_compare_mem));
3324 }
3325
3326 #endif // JFR_HAVE_INTRINSICS
3327
3328 //------------------------inline_native_currentCarrierThread------------------
3329 bool LibraryCallKit::inline_native_currentCarrierThread() {
3330 Node* junk = NULL;
3331 set_result(generate_current_thread(junk));
3332 return true;
3333 }
3334
3335 //------------------------inline_native_currentThread------------------
3336 bool LibraryCallKit::inline_native_currentThread() {
3337 Node* junk = NULL;
3338 set_result(generate_virtual_thread(junk));
3339 return true;
3340 }
3341
3342 //------------------------inline_native_setVthread------------------
3343 bool LibraryCallKit::inline_native_setCurrentThread() {
3344 assert(C->method()->changes_current_thread(),
3345 "method changes current Thread but is not annotated ChangesCurrentThread");
3346 Node* arr = argument(1);
3347 Node* thread = _gvn.transform(new ThreadLocalNode());
3348 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3349 Node* thread_obj_handle
3350 = make_load(NULL, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3351 thread_obj_handle = _gvn.transform(thread_obj_handle);
3352 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3353 // Stores of oops to native memory not supported yet by BarrierSetC2::store_at_resolved
3354 // access_store_at(NULL, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3355 store_to_memory(control(), thread_obj_handle, arr, T_OBJECT, adr_type, MemNode::unordered);
3356 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3357 return true;
3358 }
3359
3360 Node* LibraryCallKit::extentLocalCache_helper() {
3361 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3362 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3363
3364 bool xk = etype->klass_is_exact();
3365
3366 Node* thread = _gvn.transform(new ThreadLocalNode());
3367 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::extentLocalCache_offset()));
3368 return _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
3369 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3370 }
3371
3372 //------------------------inline_native_extentLocalCache------------------
3373 bool LibraryCallKit::inline_native_extentLocalCache() {
3374 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3375 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3376 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3377
3378 // Because we create the extentLocal cache lazily we have to make the
3379 // type of the result BotPTR.
3380 bool xk = etype->klass_is_exact();
3381 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3382 Node* cache_obj_handle = extentLocalCache_helper();
3383 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3384
3385 return true;
3386 }
3387
3388 //------------------------inline_native_setExtentLocalCache------------------
3389 bool LibraryCallKit::inline_native_setExtentLocalCache() {
3390 Node* arr = argument(0);
3391 Node* cache_obj_handle = extentLocalCache_helper();
3392
3393 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3394 store_to_memory(control(), cache_obj_handle, arr, T_OBJECT, adr_type,
3395 MemNode::unordered);
3396
3397 return true;
3398 }
3399
3400 //---------------------------load_mirror_from_klass----------------------------
3401 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3402 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3403 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3404 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3405 // mirror = ((OopHandle)mirror)->resolve();
3406 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3407 }
3408
3409 //-----------------------load_klass_from_mirror_common-------------------------
3410 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3411 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3412 // and branch to the given path on the region.
3413 // If never_see_null, take an uncommon trap on null, so we can optimistically
3414 // compile for the non-null case.
3415 // If the region is NULL, force never_see_null = true.
3416 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3417 bool never_see_null,
3418 RegionNode* region,
3419 int null_path,
3420 int offset) {
3421 if (region == NULL) never_see_null = true;
3422 Node* p = basic_plus_adr(mirror, offset);
3423 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3424 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3425 Node* null_ctl = top();
3426 kls = null_check_oop(kls, &null_ctl, never_see_null);
3427 if (region != NULL) {
3428 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3429 region->init_req(null_path, null_ctl);
3430 } else {
3431 assert(null_ctl == top(), "no loose ends");
3432 }
3433 return kls;
3434 }
3435
3436 //--------------------(inline_native_Class_query helpers)---------------------
3437 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3438 // Fall through if (mods & mask) == bits, take the guard otherwise.
3439 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3440 // Branch around if the given klass has the given modifier bit set.
3441 // Like generate_guard, adds a new path onto the region.
3442 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3443 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3444 Node* mask = intcon(modifier_mask);
3445 Node* bits = intcon(modifier_bits);
3446 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3447 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3448 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3449 return generate_fair_guard(bol, region);
3450 }
3451 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3452 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3453 }
3454 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3455 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3456 }
3457
3458 //-------------------------inline_native_Class_query-------------------
3459 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3460 const Type* return_type = TypeInt::BOOL;
3461 Node* prim_return_value = top(); // what happens if it's a primitive class?
3462 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3463 bool expect_prim = false; // most of these guys expect to work on refs
3464
3465 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3466
3467 Node* mirror = argument(0);
3468 Node* obj = top();
3469
3470 switch (id) {
3471 case vmIntrinsics::_isInstance:
3472 // nothing is an instance of a primitive type
3473 prim_return_value = intcon(0);
3474 obj = argument(1);
3475 break;
3476 case vmIntrinsics::_getModifiers:
3477 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3478 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3479 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3480 break;
3481 case vmIntrinsics::_isInterface:
3482 prim_return_value = intcon(0);
3483 break;
3484 case vmIntrinsics::_isArray:
3485 prim_return_value = intcon(0);
3486 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3487 break;
3488 case vmIntrinsics::_isPrimitive:
3489 prim_return_value = intcon(1);
3490 expect_prim = true; // obviously
3491 break;
3492 case vmIntrinsics::_isHidden:
3493 prim_return_value = intcon(0);
3494 break;
3495 case vmIntrinsics::_getSuperclass:
3496 prim_return_value = null();
3497 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3498 break;
3499 case vmIntrinsics::_getClassAccessFlags:
3500 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3501 return_type = TypeInt::INT; // not bool! 6297094
3502 break;
3503 default:
3504 fatal_unexpected_iid(id);
3505 break;
3506 }
3507
3508 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3509 if (mirror_con == NULL) return false; // cannot happen?
3510
3511 #ifndef PRODUCT
3512 if (C->print_intrinsics() || C->print_inlining()) {
3513 ciType* k = mirror_con->java_mirror_type();
3514 if (k) {
3515 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3516 k->print_name();
3517 tty->cr();
3518 }
3519 }
3520 #endif
3521
3522 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3523 RegionNode* region = new RegionNode(PATH_LIMIT);
3524 record_for_igvn(region);
3525 PhiNode* phi = new PhiNode(region, return_type);
3526
3527 // The mirror will never be null of Reflection.getClassAccessFlags, however
3528 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3529 // if it is. See bug 4774291.
3530
3531 // For Reflection.getClassAccessFlags(), the null check occurs in
3532 // the wrong place; see inline_unsafe_access(), above, for a similar
3533 // situation.
3534 mirror = null_check(mirror);
3535 // If mirror or obj is dead, only null-path is taken.
3536 if (stopped()) return true;
3537
3538 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3539
3540 // Now load the mirror's klass metaobject, and null-check it.
3541 // Side-effects region with the control path if the klass is null.
3542 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3543 // If kls is null, we have a primitive mirror.
3544 phi->init_req(_prim_path, prim_return_value);
3545 if (stopped()) { set_result(region, phi); return true; }
3546 bool safe_for_replace = (region->in(_prim_path) == top());
3547
3548 Node* p; // handy temp
3549 Node* null_ctl;
3550
3551 // Now that we have the non-null klass, we can perform the real query.
3552 // For constant classes, the query will constant-fold in LoadNode::Value.
3553 Node* query_value = top();
3554 switch (id) {
3555 case vmIntrinsics::_isInstance:
3556 // nothing is an instance of a primitive type
3557 query_value = gen_instanceof(obj, kls, safe_for_replace);
3558 break;
3559
3560 case vmIntrinsics::_getModifiers:
3561 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3562 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3563 break;
3564
3565 case vmIntrinsics::_isInterface:
3566 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3567 if (generate_interface_guard(kls, region) != NULL)
3568 // A guard was added. If the guard is taken, it was an interface.
3569 phi->add_req(intcon(1));
3570 // If we fall through, it's a plain class.
3571 query_value = intcon(0);
3572 break;
3573
3574 case vmIntrinsics::_isArray:
3575 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3576 if (generate_array_guard(kls, region) != NULL)
3577 // A guard was added. If the guard is taken, it was an array.
3578 phi->add_req(intcon(1));
3579 // If we fall through, it's a plain class.
3580 query_value = intcon(0);
3581 break;
3582
3583 case vmIntrinsics::_isPrimitive:
3584 query_value = intcon(0); // "normal" path produces false
3585 break;
3586
3587 case vmIntrinsics::_isHidden:
3588 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3589 if (generate_hidden_class_guard(kls, region) != NULL)
3590 // A guard was added. If the guard is taken, it was an hidden class.
3591 phi->add_req(intcon(1));
3592 // If we fall through, it's a plain class.
3593 query_value = intcon(0);
3594 break;
3595
3596
3597 case vmIntrinsics::_getSuperclass:
3598 // The rules here are somewhat unfortunate, but we can still do better
3599 // with random logic than with a JNI call.
3600 // Interfaces store null or Object as _super, but must report null.
3601 // Arrays store an intermediate super as _super, but must report Object.
3602 // Other types can report the actual _super.
3603 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3604 if (generate_interface_guard(kls, region) != NULL)
3605 // A guard was added. If the guard is taken, it was an interface.
3606 phi->add_req(null());
3607 if (generate_array_guard(kls, region) != NULL)
3608 // A guard was added. If the guard is taken, it was an array.
3609 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3610 // If we fall through, it's a plain class. Get its _super.
3611 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3612 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3613 null_ctl = top();
3614 kls = null_check_oop(kls, &null_ctl);
3615 if (null_ctl != top()) {
3616 // If the guard is taken, Object.superClass is null (both klass and mirror).
3617 region->add_req(null_ctl);
3618 phi ->add_req(null());
3619 }
3620 if (!stopped()) {
3621 query_value = load_mirror_from_klass(kls);
3622 }
3623 break;
3624
3625 case vmIntrinsics::_getClassAccessFlags:
3626 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3627 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3628 break;
3629
3630 default:
3631 fatal_unexpected_iid(id);
3632 break;
3633 }
3634
3635 // Fall-through is the normal case of a query to a real class.
3636 phi->init_req(1, query_value);
3637 region->init_req(1, control());
3638
3639 C->set_has_split_ifs(true); // Has chance for split-if optimization
3640 set_result(region, phi);
3641 return true;
3642 }
3643
3644 //-------------------------inline_Class_cast-------------------
3645 bool LibraryCallKit::inline_Class_cast() {
3646 Node* mirror = argument(0); // Class
3647 Node* obj = argument(1);
3648 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3649 if (mirror_con == NULL) {
3650 return false; // dead path (mirror->is_top()).
3651 }
3652 if (obj == NULL || obj->is_top()) {
3653 return false; // dead path
3654 }
3655 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3656
3657 // First, see if Class.cast() can be folded statically.
3658 // java_mirror_type() returns non-null for compile-time Class constants.
3659 ciType* tm = mirror_con->java_mirror_type();
3660 if (tm != NULL && tm->is_klass() &&
3661 tp != NULL) {
3662 if (!tp->is_loaded()) {
3663 // Don't use intrinsic when class is not loaded.
3664 return false;
3665 } else {
3666 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass()), tp->as_klass_type());
3667 if (static_res == Compile::SSC_always_true) {
3668 // isInstance() is true - fold the code.
3669 set_result(obj);
3670 return true;
3671 } else if (static_res == Compile::SSC_always_false) {
3672 // Don't use intrinsic, have to throw ClassCastException.
3673 // If the reference is null, the non-intrinsic bytecode will
3674 // be optimized appropriately.
3675 return false;
3676 }
3677 }
3678 }
3679
3680 // Bailout intrinsic and do normal inlining if exception path is frequent.
3681 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3682 return false;
3683 }
3684
3685 // Generate dynamic checks.
3686 // Class.cast() is java implementation of _checkcast bytecode.
3687 // Do checkcast (Parse::do_checkcast()) optimizations here.
3688
3689 mirror = null_check(mirror);
3690 // If mirror is dead, only null-path is taken.
3691 if (stopped()) {
3692 return true;
3693 }
3694
3695 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3696 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3697 RegionNode* region = new RegionNode(PATH_LIMIT);
3698 record_for_igvn(region);
3699
3700 // Now load the mirror's klass metaobject, and null-check it.
3701 // If kls is null, we have a primitive mirror and
3702 // nothing is an instance of a primitive type.
3703 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3704
3705 Node* res = top();
3706 if (!stopped()) {
3707 Node* bad_type_ctrl = top();
3708 // Do checkcast optimizations.
3709 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3710 region->init_req(_bad_type_path, bad_type_ctrl);
3711 }
3712 if (region->in(_prim_path) != top() ||
3713 region->in(_bad_type_path) != top()) {
3714 // Let Interpreter throw ClassCastException.
3715 PreserveJVMState pjvms(this);
3716 set_control(_gvn.transform(region));
3717 uncommon_trap(Deoptimization::Reason_intrinsic,
3718 Deoptimization::Action_maybe_recompile);
3719 }
3720 if (!stopped()) {
3721 set_result(res);
3722 }
3723 return true;
3724 }
3725
3726
3727 //--------------------------inline_native_subtype_check------------------------
3728 // This intrinsic takes the JNI calls out of the heart of
3729 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3730 bool LibraryCallKit::inline_native_subtype_check() {
3731 // Pull both arguments off the stack.
3732 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3733 args[0] = argument(0);
3734 args[1] = argument(1);
3735 Node* klasses[2]; // corresponding Klasses: superk, subk
3736 klasses[0] = klasses[1] = top();
3737
3738 enum {
3739 // A full decision tree on {superc is prim, subc is prim}:
3740 _prim_0_path = 1, // {P,N} => false
3741 // {P,P} & superc!=subc => false
3742 _prim_same_path, // {P,P} & superc==subc => true
3743 _prim_1_path, // {N,P} => false
3744 _ref_subtype_path, // {N,N} & subtype check wins => true
3745 _both_ref_path, // {N,N} & subtype check loses => false
3746 PATH_LIMIT
3747 };
3748
3749 RegionNode* region = new RegionNode(PATH_LIMIT);
3750 Node* phi = new PhiNode(region, TypeInt::BOOL);
3751 record_for_igvn(region);
3752
3753 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3754 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3755 int class_klass_offset = java_lang_Class::klass_offset();
3756
3757 // First null-check both mirrors and load each mirror's klass metaobject.
3758 int which_arg;
3759 for (which_arg = 0; which_arg <= 1; which_arg++) {
3760 Node* arg = args[which_arg];
3761 arg = null_check(arg);
3762 if (stopped()) break;
3763 args[which_arg] = arg;
3764
3765 Node* p = basic_plus_adr(arg, class_klass_offset);
3766 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3767 klasses[which_arg] = _gvn.transform(kls);
3768 }
3769
3770 // Having loaded both klasses, test each for null.
3771 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3772 for (which_arg = 0; which_arg <= 1; which_arg++) {
3773 Node* kls = klasses[which_arg];
3774 Node* null_ctl = top();
3775 kls = null_check_oop(kls, &null_ctl, never_see_null);
3776 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3777 region->init_req(prim_path, null_ctl);
3778 if (stopped()) break;
3779 klasses[which_arg] = kls;
3780 }
3781
3782 if (!stopped()) {
3783 // now we have two reference types, in klasses[0..1]
3784 Node* subk = klasses[1]; // the argument to isAssignableFrom
3785 Node* superk = klasses[0]; // the receiver
3786 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3787 // now we have a successful reference subtype check
3788 region->set_req(_ref_subtype_path, control());
3789 }
3790
3791 // If both operands are primitive (both klasses null), then
3792 // we must return true when they are identical primitives.
3793 // It is convenient to test this after the first null klass check.
3794 set_control(region->in(_prim_0_path)); // go back to first null check
3795 if (!stopped()) {
3796 // Since superc is primitive, make a guard for the superc==subc case.
3797 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3798 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3799 generate_guard(bol_eq, region, PROB_FAIR);
3800 if (region->req() == PATH_LIMIT+1) {
3801 // A guard was added. If the added guard is taken, superc==subc.
3802 region->swap_edges(PATH_LIMIT, _prim_same_path);
3803 region->del_req(PATH_LIMIT);
3804 }
3805 region->set_req(_prim_0_path, control()); // Not equal after all.
3806 }
3807
3808 // these are the only paths that produce 'true':
3809 phi->set_req(_prim_same_path, intcon(1));
3810 phi->set_req(_ref_subtype_path, intcon(1));
3811
3812 // pull together the cases:
3813 assert(region->req() == PATH_LIMIT, "sane region");
3814 for (uint i = 1; i < region->req(); i++) {
3815 Node* ctl = region->in(i);
3816 if (ctl == NULL || ctl == top()) {
3817 region->set_req(i, top());
3818 phi ->set_req(i, top());
3819 } else if (phi->in(i) == NULL) {
3820 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3821 }
3822 }
3823
3824 set_control(_gvn.transform(region));
3825 set_result(_gvn.transform(phi));
3826 return true;
3827 }
3828
3829 //---------------------generate_array_guard_common------------------------
3830 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3831 bool obj_array, bool not_array) {
3832
3833 if (stopped()) {
3834 return NULL;
3835 }
3836
3837 // If obj_array/non_array==false/false:
3838 // Branch around if the given klass is in fact an array (either obj or prim).
3839 // If obj_array/non_array==false/true:
3840 // Branch around if the given klass is not an array klass of any kind.
3841 // If obj_array/non_array==true/true:
3842 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3843 // If obj_array/non_array==true/false:
3844 // Branch around if the kls is an oop array (Object[] or subtype)
3845 //
3846 // Like generate_guard, adds a new path onto the region.
3847 jint layout_con = 0;
3848 Node* layout_val = get_layout_helper(kls, layout_con);
3849 if (layout_val == NULL) {
3850 bool query = (obj_array
3851 ? Klass::layout_helper_is_objArray(layout_con)
3852 : Klass::layout_helper_is_array(layout_con));
3853 if (query == not_array) {
3854 return NULL; // never a branch
3855 } else { // always a branch
3856 Node* always_branch = control();
3857 if (region != NULL)
3858 region->add_req(always_branch);
3859 set_control(top());
3860 return always_branch;
3861 }
3862 }
3863 // Now test the correct condition.
3864 jint nval = (obj_array
3865 ? (jint)(Klass::_lh_array_tag_type_value
3866 << Klass::_lh_array_tag_shift)
3867 : Klass::_lh_neutral_value);
3868 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3869 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3870 // invert the test if we are looking for a non-array
3871 if (not_array) btest = BoolTest(btest).negate();
3872 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3873 return generate_fair_guard(bol, region);
3874 }
3875
3876
3877 //-----------------------inline_native_newArray--------------------------
3878 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3879 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3880 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3881 Node* mirror;
3882 Node* count_val;
3883 if (uninitialized) {
3884 mirror = argument(1);
3885 count_val = argument(2);
3886 } else {
3887 mirror = argument(0);
3888 count_val = argument(1);
3889 }
3890
3891 mirror = null_check(mirror);
3892 // If mirror or obj is dead, only null-path is taken.
3893 if (stopped()) return true;
3894
3895 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3896 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3897 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3898 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3899 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3900
3901 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3902 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3903 result_reg, _slow_path);
3904 Node* normal_ctl = control();
3905 Node* no_array_ctl = result_reg->in(_slow_path);
3906
3907 // Generate code for the slow case. We make a call to newArray().
3908 set_control(no_array_ctl);
3909 if (!stopped()) {
3910 // Either the input type is void.class, or else the
3911 // array klass has not yet been cached. Either the
3912 // ensuing call will throw an exception, or else it
3913 // will cache the array klass for next time.
3914 PreserveJVMState pjvms(this);
3915 CallJavaNode* slow_call = NULL;
3916 if (uninitialized) {
3917 // Generate optimized virtual call (holder class 'Unsafe' is final)
3918 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3919 } else {
3920 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3921 }
3922 Node* slow_result = set_results_for_java_call(slow_call);
3923 // this->control() comes from set_results_for_java_call
3924 result_reg->set_req(_slow_path, control());
3925 result_val->set_req(_slow_path, slow_result);
3926 result_io ->set_req(_slow_path, i_o());
3927 result_mem->set_req(_slow_path, reset_memory());
3928 }
3929
3930 set_control(normal_ctl);
3931 if (!stopped()) {
3932 // Normal case: The array type has been cached in the java.lang.Class.
3933 // The following call works fine even if the array type is polymorphic.
3934 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3935 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3936 result_reg->init_req(_normal_path, control());
3937 result_val->init_req(_normal_path, obj);
3938 result_io ->init_req(_normal_path, i_o());
3939 result_mem->init_req(_normal_path, reset_memory());
3940
3941 if (uninitialized) {
3942 // Mark the allocation so that zeroing is skipped
3943 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3944 alloc->maybe_set_complete(&_gvn);
3945 }
3946 }
3947
3948 // Return the combined state.
3949 set_i_o( _gvn.transform(result_io) );
3950 set_all_memory( _gvn.transform(result_mem));
3951
3952 C->set_has_split_ifs(true); // Has chance for split-if optimization
3953 set_result(result_reg, result_val);
3954 return true;
3955 }
3956
3957 //----------------------inline_native_getLength--------------------------
3958 // public static native int java.lang.reflect.Array.getLength(Object array);
3959 bool LibraryCallKit::inline_native_getLength() {
3960 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3961
3962 Node* array = null_check(argument(0));
3963 // If array is dead, only null-path is taken.
3964 if (stopped()) return true;
3965
3966 // Deoptimize if it is a non-array.
3967 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3968
3969 if (non_array != NULL) {
3970 PreserveJVMState pjvms(this);
3971 set_control(non_array);
3972 uncommon_trap(Deoptimization::Reason_intrinsic,
3973 Deoptimization::Action_maybe_recompile);
3974 }
3975
3976 // If control is dead, only non-array-path is taken.
3977 if (stopped()) return true;
3978
3979 // The works fine even if the array type is polymorphic.
3980 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3981 Node* result = load_array_length(array);
3982
3983 C->set_has_split_ifs(true); // Has chance for split-if optimization
3984 set_result(result);
3985 return true;
3986 }
3987
3988 //------------------------inline_array_copyOf----------------------------
3989 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3990 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3991 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3992 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3993
3994 // Get the arguments.
3995 Node* original = argument(0);
3996 Node* start = is_copyOfRange? argument(1): intcon(0);
3997 Node* end = is_copyOfRange? argument(2): argument(1);
3998 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3999
4000 Node* newcopy = NULL;
4001
4002 // Set the original stack and the reexecute bit for the interpreter to reexecute
4003 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4004 { PreserveReexecuteState preexecs(this);
4005 jvms()->set_should_reexecute(true);
4006
4007 array_type_mirror = null_check(array_type_mirror);
4008 original = null_check(original);
4009
4010 // Check if a null path was taken unconditionally.
4011 if (stopped()) return true;
4012
4013 Node* orig_length = load_array_length(original);
4014
4015 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4016 klass_node = null_check(klass_node);
4017
4018 RegionNode* bailout = new RegionNode(1);
4019 record_for_igvn(bailout);
4020
4021 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4022 // Bail out if that is so.
4023 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4024 if (not_objArray != NULL) {
4025 // Improve the klass node's type from the new optimistic assumption:
4026 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4027 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4028 Node* cast = new CastPPNode(klass_node, akls);
4029 cast->init_req(0, control());
4030 klass_node = _gvn.transform(cast);
4031 }
4032
4033 // Bail out if either start or end is negative.
4034 generate_negative_guard(start, bailout, &start);
4035 generate_negative_guard(end, bailout, &end);
4036
4037 Node* length = end;
4038 if (_gvn.type(start) != TypeInt::ZERO) {
4039 length = _gvn.transform(new SubINode(end, start));
4040 }
4041
4042 // Bail out if length is negative.
4043 // Without this the new_array would throw
4044 // NegativeArraySizeException but IllegalArgumentException is what
4045 // should be thrown
4046 generate_negative_guard(length, bailout, &length);
4047
4048 if (bailout->req() > 1) {
4049 PreserveJVMState pjvms(this);
4050 set_control(_gvn.transform(bailout));
4051 uncommon_trap(Deoptimization::Reason_intrinsic,
4052 Deoptimization::Action_maybe_recompile);
4053 }
4054
4055 if (!stopped()) {
4056 // How many elements will we copy from the original?
4057 // The answer is MinI(orig_length - start, length).
4058 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4059 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4060
4061 // Generate a direct call to the right arraycopy function(s).
4062 // We know the copy is disjoint but we might not know if the
4063 // oop stores need checking.
4064 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4065 // This will fail a store-check if x contains any non-nulls.
4066
4067 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4068 // loads/stores but it is legal only if we're sure the
4069 // Arrays.copyOf would succeed. So we need all input arguments
4070 // to the copyOf to be validated, including that the copy to the
4071 // new array won't trigger an ArrayStoreException. That subtype
4072 // check can be optimized if we know something on the type of
4073 // the input array from type speculation.
4074 if (_gvn.type(klass_node)->singleton()) {
4075 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4076 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4077
4078 int test = C->static_subtype_check(superk, subk);
4079 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4080 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4081 if (t_original->speculative_type() != NULL) {
4082 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4083 }
4084 }
4085 }
4086
4087 bool validated = false;
4088 // Reason_class_check rather than Reason_intrinsic because we
4089 // want to intrinsify even if this traps.
4090 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4091 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4092
4093 if (not_subtype_ctrl != top()) {
4094 PreserveJVMState pjvms(this);
4095 set_control(not_subtype_ctrl);
4096 uncommon_trap(Deoptimization::Reason_class_check,
4097 Deoptimization::Action_make_not_entrant);
4098 assert(stopped(), "Should be stopped");
4099 }
4100 validated = true;
4101 }
4102
4103 if (!stopped()) {
4104 newcopy = new_array(klass_node, length, 0); // no arguments to push
4105
4106 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4107 load_object_klass(original), klass_node);
4108 if (!is_copyOfRange) {
4109 ac->set_copyof(validated);
4110 } else {
4111 ac->set_copyofrange(validated);
4112 }
4113 Node* n = _gvn.transform(ac);
4114 if (n == ac) {
4115 ac->connect_outputs(this);
4116 } else {
4117 assert(validated, "shouldn't transform if all arguments not validated");
4118 set_all_memory(n);
4119 }
4120 }
4121 }
4122 } // original reexecute is set back here
4123
4124 C->set_has_split_ifs(true); // Has chance for split-if optimization
4125 if (!stopped()) {
4126 set_result(newcopy);
4127 }
4128 return true;
4129 }
4130
4131
4132 //----------------------generate_virtual_guard---------------------------
4133 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4134 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4135 RegionNode* slow_region) {
4136 ciMethod* method = callee();
4137 int vtable_index = method->vtable_index();
4138 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4139 "bad index %d", vtable_index);
4140 // Get the Method* out of the appropriate vtable entry.
4141 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4142 vtable_index*vtableEntry::size_in_bytes() +
4143 vtableEntry::method_offset_in_bytes();
4144 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4145 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4146
4147 // Compare the target method with the expected method (e.g., Object.hashCode).
4148 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4149
4150 Node* native_call = makecon(native_call_addr);
4151 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4152 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4153
4154 return generate_slow_guard(test_native, slow_region);
4155 }
4156
4157 //-----------------------generate_method_call----------------------------
4158 // Use generate_method_call to make a slow-call to the real
4159 // method if the fast path fails. An alternative would be to
4160 // use a stub like OptoRuntime::slow_arraycopy_Java.
4161 // This only works for expanding the current library call,
4162 // not another intrinsic. (E.g., don't use this for making an
4163 // arraycopy call inside of the copyOf intrinsic.)
4164 CallJavaNode*
4165 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4166 // When compiling the intrinsic method itself, do not use this technique.
4167 guarantee(callee() != C->method(), "cannot make slow-call to self");
4168
4169 ciMethod* method = callee();
4170 // ensure the JVMS we have will be correct for this call
4171 guarantee(method_id == method->intrinsic_id(), "must match");
4172
4173 const TypeFunc* tf = TypeFunc::make(method);
4174 CallJavaNode* slow_call;
4175 if (is_static) {
4176 assert(!is_virtual, "");
4177 slow_call = new CallStaticJavaNode(C, tf,
4178 SharedRuntime::get_resolve_static_call_stub(), method);
4179 } else if (is_virtual) {
4180 null_check_receiver();
4181 int vtable_index = Method::invalid_vtable_index;
4182 if (UseInlineCaches) {
4183 // Suppress the vtable call
4184 } else {
4185 // hashCode and clone are not a miranda methods,
4186 // so the vtable index is fixed.
4187 // No need to use the linkResolver to get it.
4188 vtable_index = method->vtable_index();
4189 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4190 "bad index %d", vtable_index);
4191 }
4192 slow_call = new CallDynamicJavaNode(tf,
4193 SharedRuntime::get_resolve_virtual_call_stub(),
4194 method, vtable_index);
4195 } else { // neither virtual nor static: opt_virtual
4196 null_check_receiver();
4197 slow_call = new CallStaticJavaNode(C, tf,
4198 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4199 slow_call->set_optimized_virtual(true);
4200 }
4201 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4202 // To be able to issue a direct call (optimized virtual or virtual)
4203 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4204 // about the method being invoked should be attached to the call site to
4205 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4206 slow_call->set_override_symbolic_info(true);
4207 }
4208 set_arguments_for_java_call(slow_call);
4209 set_edges_for_java_call(slow_call);
4210 return slow_call;
4211 }
4212
4213
4214 /**
4215 * Build special case code for calls to hashCode on an object. This call may
4216 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4217 * slightly different code.
4218 */
4219 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4220 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4221 assert(!(is_virtual && is_static), "either virtual, special, or static");
4222
4223 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4224
4225 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4226 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4227 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4228 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4229 Node* obj = NULL;
4230 if (!is_static) {
4231 // Check for hashing null object
4232 obj = null_check_receiver();
4233 if (stopped()) return true; // unconditionally null
4234 result_reg->init_req(_null_path, top());
4235 result_val->init_req(_null_path, top());
4236 } else {
4237 // Do a null check, and return zero if null.
4238 // System.identityHashCode(null) == 0
4239 obj = argument(0);
4240 Node* null_ctl = top();
4241 obj = null_check_oop(obj, &null_ctl);
4242 result_reg->init_req(_null_path, null_ctl);
4243 result_val->init_req(_null_path, _gvn.intcon(0));
4244 }
4245
4246 // Unconditionally null? Then return right away.
4247 if (stopped()) {
4248 set_control( result_reg->in(_null_path));
4249 if (!stopped())
4250 set_result(result_val->in(_null_path));
4251 return true;
4252 }
4253
4254 // We only go to the fast case code if we pass a number of guards. The
4255 // paths which do not pass are accumulated in the slow_region.
4256 RegionNode* slow_region = new RegionNode(1);
4257 record_for_igvn(slow_region);
4258
4259 // If this is a virtual call, we generate a funny guard. We pull out
4260 // the vtable entry corresponding to hashCode() from the target object.
4261 // If the target method which we are calling happens to be the native
4262 // Object hashCode() method, we pass the guard. We do not need this
4263 // guard for non-virtual calls -- the caller is known to be the native
4264 // Object hashCode().
4265 if (is_virtual) {
4266 // After null check, get the object's klass.
4267 Node* obj_klass = load_object_klass(obj);
4268 generate_virtual_guard(obj_klass, slow_region);
4269 }
4270
4271 // Get the header out of the object, use LoadMarkNode when available
4272 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4273 // The control of the load must be NULL. Otherwise, the load can move before
4274 // the null check after castPP removal.
4275 Node* no_ctrl = NULL;
4276 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4277
4278 // Test the header to see if it is unlocked.
4279 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4280 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4281 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4282 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4283 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4284
4285 generate_slow_guard(test_unlocked, slow_region);
4286
4287 // Get the hash value and check to see that it has been properly assigned.
4288 // We depend on hash_mask being at most 32 bits and avoid the use of
4289 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4290 // vm: see markWord.hpp.
4291 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4292 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4293 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4294 // This hack lets the hash bits live anywhere in the mark object now, as long
4295 // as the shift drops the relevant bits into the low 32 bits. Note that
4296 // Java spec says that HashCode is an int so there's no point in capturing
4297 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4298 hshifted_header = ConvX2I(hshifted_header);
4299 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4300
4301 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4302 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4303 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4304
4305 generate_slow_guard(test_assigned, slow_region);
4306
4307 Node* init_mem = reset_memory();
4308 // fill in the rest of the null path:
4309 result_io ->init_req(_null_path, i_o());
4310 result_mem->init_req(_null_path, init_mem);
4311
4312 result_val->init_req(_fast_path, hash_val);
4313 result_reg->init_req(_fast_path, control());
4314 result_io ->init_req(_fast_path, i_o());
4315 result_mem->init_req(_fast_path, init_mem);
4316
4317 // Generate code for the slow case. We make a call to hashCode().
4318 set_control(_gvn.transform(slow_region));
4319 if (!stopped()) {
4320 // No need for PreserveJVMState, because we're using up the present state.
4321 set_all_memory(init_mem);
4322 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4323 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4324 Node* slow_result = set_results_for_java_call(slow_call);
4325 // this->control() comes from set_results_for_java_call
4326 result_reg->init_req(_slow_path, control());
4327 result_val->init_req(_slow_path, slow_result);
4328 result_io ->set_req(_slow_path, i_o());
4329 result_mem ->set_req(_slow_path, reset_memory());
4330 }
4331
4332 // Return the combined state.
4333 set_i_o( _gvn.transform(result_io) );
4334 set_all_memory( _gvn.transform(result_mem));
4335
4336 set_result(result_reg, result_val);
4337 return true;
4338 }
4339
4340 //---------------------------inline_native_getClass----------------------------
4341 // public final native Class<?> java.lang.Object.getClass();
4342 //
4343 // Build special case code for calls to getClass on an object.
4344 bool LibraryCallKit::inline_native_getClass() {
4345 Node* obj = null_check_receiver();
4346 if (stopped()) return true;
4347 set_result(load_mirror_from_klass(load_object_klass(obj)));
4348 return true;
4349 }
4350
4351 //-----------------inline_native_Reflection_getCallerClass---------------------
4352 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4353 //
4354 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4355 //
4356 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4357 // in that it must skip particular security frames and checks for
4358 // caller sensitive methods.
4359 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4360 #ifndef PRODUCT
4361 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4362 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4363 }
4364 #endif
4365
4366 if (!jvms()->has_method()) {
4367 #ifndef PRODUCT
4368 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4369 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4370 }
4371 #endif
4372 return false;
4373 }
4374
4375 // Walk back up the JVM state to find the caller at the required
4376 // depth.
4377 JVMState* caller_jvms = jvms();
4378
4379 // Cf. JVM_GetCallerClass
4380 // NOTE: Start the loop at depth 1 because the current JVM state does
4381 // not include the Reflection.getCallerClass() frame.
4382 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4383 ciMethod* m = caller_jvms->method();
4384 switch (n) {
4385 case 0:
4386 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4387 break;
4388 case 1:
4389 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4390 if (!m->caller_sensitive()) {
4391 #ifndef PRODUCT
4392 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4393 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4394 }
4395 #endif
4396 return false; // bail-out; let JVM_GetCallerClass do the work
4397 }
4398 break;
4399 default:
4400 if (!m->is_ignored_by_security_stack_walk()) {
4401 // We have reached the desired frame; return the holder class.
4402 // Acquire method holder as java.lang.Class and push as constant.
4403 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4404 ciInstance* caller_mirror = caller_klass->java_mirror();
4405 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4406
4407 #ifndef PRODUCT
4408 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4409 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());
4410 tty->print_cr(" JVM state at this point:");
4411 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4412 ciMethod* m = jvms()->of_depth(i)->method();
4413 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4414 }
4415 }
4416 #endif
4417 return true;
4418 }
4419 break;
4420 }
4421 }
4422
4423 #ifndef PRODUCT
4424 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4425 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4426 tty->print_cr(" JVM state at this point:");
4427 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4428 ciMethod* m = jvms()->of_depth(i)->method();
4429 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4430 }
4431 }
4432 #endif
4433
4434 return false; // bail-out; let JVM_GetCallerClass do the work
4435 }
4436
4437 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4438 Node* arg = argument(0);
4439 Node* result = NULL;
4440
4441 switch (id) {
4442 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4443 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4444 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4445 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4446
4447 case vmIntrinsics::_doubleToLongBits: {
4448 // two paths (plus control) merge in a wood
4449 RegionNode *r = new RegionNode(3);
4450 Node *phi = new PhiNode(r, TypeLong::LONG);
4451
4452 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4453 // Build the boolean node
4454 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4455
4456 // Branch either way.
4457 // NaN case is less traveled, which makes all the difference.
4458 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4459 Node *opt_isnan = _gvn.transform(ifisnan);
4460 assert( opt_isnan->is_If(), "Expect an IfNode");
4461 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4462 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4463
4464 set_control(iftrue);
4465
4466 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4467 Node *slow_result = longcon(nan_bits); // return NaN
4468 phi->init_req(1, _gvn.transform( slow_result ));
4469 r->init_req(1, iftrue);
4470
4471 // Else fall through
4472 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4473 set_control(iffalse);
4474
4475 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4476 r->init_req(2, iffalse);
4477
4478 // Post merge
4479 set_control(_gvn.transform(r));
4480 record_for_igvn(r);
4481
4482 C->set_has_split_ifs(true); // Has chance for split-if optimization
4483 result = phi;
4484 assert(result->bottom_type()->isa_long(), "must be");
4485 break;
4486 }
4487
4488 case vmIntrinsics::_floatToIntBits: {
4489 // two paths (plus control) merge in a wood
4490 RegionNode *r = new RegionNode(3);
4491 Node *phi = new PhiNode(r, TypeInt::INT);
4492
4493 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4494 // Build the boolean node
4495 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4496
4497 // Branch either way.
4498 // NaN case is less traveled, which makes all the difference.
4499 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4500 Node *opt_isnan = _gvn.transform(ifisnan);
4501 assert( opt_isnan->is_If(), "Expect an IfNode");
4502 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4503 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4504
4505 set_control(iftrue);
4506
4507 static const jint nan_bits = 0x7fc00000;
4508 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4509 phi->init_req(1, _gvn.transform( slow_result ));
4510 r->init_req(1, iftrue);
4511
4512 // Else fall through
4513 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4514 set_control(iffalse);
4515
4516 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4517 r->init_req(2, iffalse);
4518
4519 // Post merge
4520 set_control(_gvn.transform(r));
4521 record_for_igvn(r);
4522
4523 C->set_has_split_ifs(true); // Has chance for split-if optimization
4524 result = phi;
4525 assert(result->bottom_type()->isa_int(), "must be");
4526 break;
4527 }
4528
4529 default:
4530 fatal_unexpected_iid(id);
4531 break;
4532 }
4533 set_result(_gvn.transform(result));
4534 return true;
4535 }
4536
4537 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4538 Node* arg = argument(0);
4539 Node* result = NULL;
4540
4541 switch (id) {
4542 case vmIntrinsics::_floatIsInfinite:
4543 result = new IsInfiniteFNode(arg);
4544 break;
4545 case vmIntrinsics::_doubleIsInfinite:
4546 result = new IsInfiniteDNode(arg);
4547 break;
4548 default:
4549 fatal_unexpected_iid(id);
4550 break;
4551 }
4552 set_result(_gvn.transform(result));
4553 return true;
4554 }
4555
4556 //----------------------inline_unsafe_copyMemory-------------------------
4557 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4558
4559 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4560 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4561 const Type* base_t = gvn.type(base);
4562
4563 bool in_native = (base_t == TypePtr::NULL_PTR);
4564 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4565 bool is_mixed = !in_heap && !in_native;
4566
4567 if (is_mixed) {
4568 return true; // mixed accesses can touch both on-heap and off-heap memory
4569 }
4570 if (in_heap) {
4571 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4572 if (!is_prim_array) {
4573 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4574 // there's not enough type information available to determine proper memory slice for it.
4575 return true;
4576 }
4577 }
4578 return false;
4579 }
4580
4581 bool LibraryCallKit::inline_unsafe_copyMemory() {
4582 if (callee()->is_static()) return false; // caller must have the capability!
4583 null_check_receiver(); // null-check receiver
4584 if (stopped()) return true;
4585
4586 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4587
4588 Node* src_base = argument(1); // type: oop
4589 Node* src_off = ConvL2X(argument(2)); // type: long
4590 Node* dst_base = argument(4); // type: oop
4591 Node* dst_off = ConvL2X(argument(5)); // type: long
4592 Node* size = ConvL2X(argument(7)); // type: long
4593
4594 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4595 "fieldOffset must be byte-scaled");
4596
4597 Node* src_addr = make_unsafe_address(src_base, src_off);
4598 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4599
4600 Node* thread = _gvn.transform(new ThreadLocalNode());
4601 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4602 BasicType doing_unsafe_access_bt = T_BYTE;
4603 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4604
4605 // update volatile field
4606 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4607
4608 int flags = RC_LEAF | RC_NO_FP;
4609
4610 const TypePtr* dst_type = TypePtr::BOTTOM;
4611
4612 // Adjust memory effects of the runtime call based on input values.
4613 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4614 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4615 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4616
4617 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4618 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4619 flags |= RC_NARROW_MEM; // narrow in memory
4620 }
4621 }
4622
4623 // Call it. Note that the length argument is not scaled.
4624 make_runtime_call(flags,
4625 OptoRuntime::fast_arraycopy_Type(),
4626 StubRoutines::unsafe_arraycopy(),
4627 "unsafe_arraycopy",
4628 dst_type,
4629 src_addr, dst_addr, size XTOP);
4630
4631 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4632
4633 return true;
4634 }
4635
4636 #undef XTOP
4637
4638 //------------------------clone_coping-----------------------------------
4639 // Helper function for inline_native_clone.
4640 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4641 assert(obj_size != NULL, "");
4642 Node* raw_obj = alloc_obj->in(1);
4643 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4644
4645 AllocateNode* alloc = NULL;
4646 if (ReduceBulkZeroing) {
4647 // We will be completely responsible for initializing this object -
4648 // mark Initialize node as complete.
4649 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4650 // The object was just allocated - there should be no any stores!
4651 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4652 // Mark as complete_with_arraycopy so that on AllocateNode
4653 // expansion, we know this AllocateNode is initialized by an array
4654 // copy and a StoreStore barrier exists after the array copy.
4655 alloc->initialization()->set_complete_with_arraycopy();
4656 }
4657
4658 Node* size = _gvn.transform(obj_size);
4659 access_clone(obj, alloc_obj, size, is_array);
4660
4661 // Do not let reads from the cloned object float above the arraycopy.
4662 if (alloc != NULL) {
4663 // Do not let stores that initialize this object be reordered with
4664 // a subsequent store that would make this object accessible by
4665 // other threads.
4666 // Record what AllocateNode this StoreStore protects so that
4667 // escape analysis can go from the MemBarStoreStoreNode to the
4668 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4669 // based on the escape status of the AllocateNode.
4670 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4671 } else {
4672 insert_mem_bar(Op_MemBarCPUOrder);
4673 }
4674 }
4675
4676 //------------------------inline_native_clone----------------------------
4677 // protected native Object java.lang.Object.clone();
4678 //
4679 // Here are the simple edge cases:
4680 // null receiver => normal trap
4681 // virtual and clone was overridden => slow path to out-of-line clone
4682 // not cloneable or finalizer => slow path to out-of-line Object.clone
4683 //
4684 // The general case has two steps, allocation and copying.
4685 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4686 //
4687 // Copying also has two cases, oop arrays and everything else.
4688 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4689 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4690 //
4691 // These steps fold up nicely if and when the cloned object's klass
4692 // can be sharply typed as an object array, a type array, or an instance.
4693 //
4694 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4695 PhiNode* result_val;
4696
4697 // Set the reexecute bit for the interpreter to reexecute
4698 // the bytecode that invokes Object.clone if deoptimization happens.
4699 { PreserveReexecuteState preexecs(this);
4700 jvms()->set_should_reexecute(true);
4701
4702 Node* obj = null_check_receiver();
4703 if (stopped()) return true;
4704
4705 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4706
4707 // If we are going to clone an instance, we need its exact type to
4708 // know the number and types of fields to convert the clone to
4709 // loads/stores. Maybe a speculative type can help us.
4710 if (!obj_type->klass_is_exact() &&
4711 obj_type->speculative_type() != NULL &&
4712 obj_type->speculative_type()->is_instance_klass()) {
4713 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4714 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4715 !spec_ik->has_injected_fields()) {
4716 if (!obj_type->isa_instptr() ||
4717 obj_type->is_instptr()->instance_klass()->has_subklass()) {
4718 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4719 }
4720 }
4721 }
4722
4723 // Conservatively insert a memory barrier on all memory slices.
4724 // Do not let writes into the original float below the clone.
4725 insert_mem_bar(Op_MemBarCPUOrder);
4726
4727 // paths into result_reg:
4728 enum {
4729 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4730 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4731 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4732 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4733 PATH_LIMIT
4734 };
4735 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4736 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4737 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4738 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4739 record_for_igvn(result_reg);
4740
4741 Node* obj_klass = load_object_klass(obj);
4742 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4743 if (array_ctl != NULL) {
4744 // It's an array.
4745 PreserveJVMState pjvms(this);
4746 set_control(array_ctl);
4747 Node* obj_length = load_array_length(obj);
4748 Node* obj_size = NULL;
4749 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4750
4751 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4752 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4753 // If it is an oop array, it requires very special treatment,
4754 // because gc barriers are required when accessing the array.
4755 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4756 if (is_obja != NULL) {
4757 PreserveJVMState pjvms2(this);
4758 set_control(is_obja);
4759 // Generate a direct call to the right arraycopy function(s).
4760 // Clones are always tightly coupled.
4761 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4762 ac->set_clone_oop_array();
4763 Node* n = _gvn.transform(ac);
4764 assert(n == ac, "cannot disappear");
4765 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4766
4767 result_reg->init_req(_objArray_path, control());
4768 result_val->init_req(_objArray_path, alloc_obj);
4769 result_i_o ->set_req(_objArray_path, i_o());
4770 result_mem ->set_req(_objArray_path, reset_memory());
4771 }
4772 }
4773 // Otherwise, there are no barriers to worry about.
4774 // (We can dispense with card marks if we know the allocation
4775 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4776 // causes the non-eden paths to take compensating steps to
4777 // simulate a fresh allocation, so that no further
4778 // card marks are required in compiled code to initialize
4779 // the object.)
4780
4781 if (!stopped()) {
4782 copy_to_clone(obj, alloc_obj, obj_size, true);
4783
4784 // Present the results of the copy.
4785 result_reg->init_req(_array_path, control());
4786 result_val->init_req(_array_path, alloc_obj);
4787 result_i_o ->set_req(_array_path, i_o());
4788 result_mem ->set_req(_array_path, reset_memory());
4789 }
4790 }
4791
4792 // We only go to the instance fast case code if we pass a number of guards.
4793 // The paths which do not pass are accumulated in the slow_region.
4794 RegionNode* slow_region = new RegionNode(1);
4795 record_for_igvn(slow_region);
4796 if (!stopped()) {
4797 // It's an instance (we did array above). Make the slow-path tests.
4798 // If this is a virtual call, we generate a funny guard. We grab
4799 // the vtable entry corresponding to clone() from the target object.
4800 // If the target method which we are calling happens to be the
4801 // Object clone() method, we pass the guard. We do not need this
4802 // guard for non-virtual calls; the caller is known to be the native
4803 // Object clone().
4804 if (is_virtual) {
4805 generate_virtual_guard(obj_klass, slow_region);
4806 }
4807
4808 // The object must be easily cloneable and must not have a finalizer.
4809 // Both of these conditions may be checked in a single test.
4810 // We could optimize the test further, but we don't care.
4811 generate_access_flags_guard(obj_klass,
4812 // Test both conditions:
4813 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4814 // Must be cloneable but not finalizer:
4815 JVM_ACC_IS_CLONEABLE_FAST,
4816 slow_region);
4817 }
4818
4819 if (!stopped()) {
4820 // It's an instance, and it passed the slow-path tests.
4821 PreserveJVMState pjvms(this);
4822 Node* obj_size = NULL;
4823 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4824 // is reexecuted if deoptimization occurs and there could be problems when merging
4825 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4826 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4827
4828 copy_to_clone(obj, alloc_obj, obj_size, false);
4829
4830 // Present the results of the slow call.
4831 result_reg->init_req(_instance_path, control());
4832 result_val->init_req(_instance_path, alloc_obj);
4833 result_i_o ->set_req(_instance_path, i_o());
4834 result_mem ->set_req(_instance_path, reset_memory());
4835 }
4836
4837 // Generate code for the slow case. We make a call to clone().
4838 set_control(_gvn.transform(slow_region));
4839 if (!stopped()) {
4840 PreserveJVMState pjvms(this);
4841 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4842 // We need to deoptimize on exception (see comment above)
4843 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4844 // this->control() comes from set_results_for_java_call
4845 result_reg->init_req(_slow_path, control());
4846 result_val->init_req(_slow_path, slow_result);
4847 result_i_o ->set_req(_slow_path, i_o());
4848 result_mem ->set_req(_slow_path, reset_memory());
4849 }
4850
4851 // Return the combined state.
4852 set_control( _gvn.transform(result_reg));
4853 set_i_o( _gvn.transform(result_i_o));
4854 set_all_memory( _gvn.transform(result_mem));
4855 } // original reexecute is set back here
4856
4857 set_result(_gvn.transform(result_val));
4858 return true;
4859 }
4860
4861 // If we have a tightly coupled allocation, the arraycopy may take care
4862 // of the array initialization. If one of the guards we insert between
4863 // the allocation and the arraycopy causes a deoptimization, an
4864 // uninitialized array will escape the compiled method. To prevent that
4865 // we set the JVM state for uncommon traps between the allocation and
4866 // the arraycopy to the state before the allocation so, in case of
4867 // deoptimization, we'll reexecute the allocation and the
4868 // initialization.
4869 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4870 if (alloc != NULL) {
4871 ciMethod* trap_method = alloc->jvms()->method();
4872 int trap_bci = alloc->jvms()->bci();
4873
4874 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4875 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4876 // Make sure there's no store between the allocation and the
4877 // arraycopy otherwise visible side effects could be rexecuted
4878 // in case of deoptimization and cause incorrect execution.
4879 bool no_interfering_store = true;
4880 Node* mem = alloc->in(TypeFunc::Memory);
4881 if (mem->is_MergeMem()) {
4882 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4883 Node* n = mms.memory();
4884 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4885 assert(n->is_Store(), "what else?");
4886 no_interfering_store = false;
4887 break;
4888 }
4889 }
4890 } else {
4891 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4892 Node* n = mms.memory();
4893 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4894 assert(n->is_Store(), "what else?");
4895 no_interfering_store = false;
4896 break;
4897 }
4898 }
4899 }
4900
4901 if (no_interfering_store) {
4902 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4903 uint size = alloc->req();
4904 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4905 old_jvms->set_map(sfpt);
4906 for (uint i = 0; i < size; i++) {
4907 sfpt->init_req(i, alloc->in(i));
4908 }
4909 // re-push array length for deoptimization
4910 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4911 old_jvms->set_sp(old_jvms->sp()+1);
4912 old_jvms->set_monoff(old_jvms->monoff()+1);
4913 old_jvms->set_scloff(old_jvms->scloff()+1);
4914 old_jvms->set_endoff(old_jvms->endoff()+1);
4915 old_jvms->set_should_reexecute(true);
4916
4917 sfpt->set_i_o(map()->i_o());
4918 sfpt->set_memory(map()->memory());
4919 sfpt->set_control(map()->control());
4920
4921 JVMState* saved_jvms = jvms();
4922 saved_reexecute_sp = _reexecute_sp;
4923
4924 set_jvms(sfpt->jvms());
4925 _reexecute_sp = jvms()->sp();
4926
4927 return saved_jvms;
4928 }
4929 }
4930 }
4931 return NULL;
4932 }
4933
4934 // In case of a deoptimization, we restart execution at the
4935 // allocation, allocating a new array. We would leave an uninitialized
4936 // array in the heap that GCs wouldn't expect. Move the allocation
4937 // after the traps so we don't allocate the array if we
4938 // deoptimize. This is possible because tightly_coupled_allocation()
4939 // guarantees there's no observer of the allocated array at this point
4940 // and the control flow is simple enough.
4941 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4942 int saved_reexecute_sp, uint new_idx) {
4943 if (saved_jvms != NULL && !stopped()) {
4944 assert(alloc != NULL, "only with a tightly coupled allocation");
4945 // restore JVM state to the state at the arraycopy
4946 saved_jvms->map()->set_control(map()->control());
4947 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4948 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4949 // If we've improved the types of some nodes (null check) while
4950 // emitting the guards, propagate them to the current state
4951 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4952 set_jvms(saved_jvms);
4953 _reexecute_sp = saved_reexecute_sp;
4954
4955 // Remove the allocation from above the guards
4956 CallProjections callprojs;
4957 alloc->extract_projections(&callprojs, true);
4958 InitializeNode* init = alloc->initialization();
4959 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4960 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4961 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4962
4963 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4964 // the allocation (i.e. is only valid if the allocation succeeds):
4965 // 1) replace CastIINode with AllocateArrayNode's length here
4966 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4967 //
4968 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4969 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4970 Node* init_control = init->proj_out(TypeFunc::Control);
4971 Node* alloc_length = alloc->Ideal_length();
4972 #ifdef ASSERT
4973 Node* prev_cast = NULL;
4974 #endif
4975 for (uint i = 0; i < init_control->outcnt(); i++) {
4976 Node* init_out = init_control->raw_out(i);
4977 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4978 #ifdef ASSERT
4979 if (prev_cast == NULL) {
4980 prev_cast = init_out;
4981 } else {
4982 if (prev_cast->cmp(*init_out) == false) {
4983 prev_cast->dump();
4984 init_out->dump();
4985 assert(false, "not equal CastIINode");
4986 }
4987 }
4988 #endif
4989 C->gvn_replace_by(init_out, alloc_length);
4990 }
4991 }
4992 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4993
4994 // move the allocation here (after the guards)
4995 _gvn.hash_delete(alloc);
4996 alloc->set_req(TypeFunc::Control, control());
4997 alloc->set_req(TypeFunc::I_O, i_o());
4998 Node *mem = reset_memory();
4999 set_all_memory(mem);
5000 alloc->set_req(TypeFunc::Memory, mem);
5001 set_control(init->proj_out_or_null(TypeFunc::Control));
5002 set_i_o(callprojs.fallthrough_ioproj);
5003
5004 // Update memory as done in GraphKit::set_output_for_allocation()
5005 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5006 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5007 if (ary_type->isa_aryptr() && length_type != NULL) {
5008 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5009 }
5010 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5011 int elemidx = C->get_alias_index(telemref);
5012 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5013 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5014
5015 Node* allocx = _gvn.transform(alloc);
5016 assert(allocx == alloc, "where has the allocation gone?");
5017 assert(dest->is_CheckCastPP(), "not an allocation result?");
5018
5019 _gvn.hash_delete(dest);
5020 dest->set_req(0, control());
5021 Node* destx = _gvn.transform(dest);
5022 assert(destx == dest, "where has the allocation result gone?");
5023
5024 array_ideal_length(alloc, ary_type, true);
5025 }
5026 }
5027
5028
5029 //------------------------------inline_arraycopy-----------------------
5030 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5031 // Object dest, int destPos,
5032 // int length);
5033 bool LibraryCallKit::inline_arraycopy() {
5034 // Get the arguments.
5035 Node* src = argument(0); // type: oop
5036 Node* src_offset = argument(1); // type: int
5037 Node* dest = argument(2); // type: oop
5038 Node* dest_offset = argument(3); // type: int
5039 Node* length = argument(4); // type: int
5040
5041 uint new_idx = C->unique();
5042
5043 // Check for allocation before we add nodes that would confuse
5044 // tightly_coupled_allocation()
5045 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5046
5047 int saved_reexecute_sp = -1;
5048 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5049 // See arraycopy_restore_alloc_state() comment
5050 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5051 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5052 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5053 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5054
5055 // The following tests must be performed
5056 // (1) src and dest are arrays.
5057 // (2) src and dest arrays must have elements of the same BasicType
5058 // (3) src and dest must not be null.
5059 // (4) src_offset must not be negative.
5060 // (5) dest_offset must not be negative.
5061 // (6) length must not be negative.
5062 // (7) src_offset + length must not exceed length of src.
5063 // (8) dest_offset + length must not exceed length of dest.
5064 // (9) each element of an oop array must be assignable
5065
5066 // (3) src and dest must not be null.
5067 // always do this here because we need the JVM state for uncommon traps
5068 Node* null_ctl = top();
5069 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5070 assert(null_ctl->is_top(), "no null control here");
5071 dest = null_check(dest, T_ARRAY);
5072
5073 if (!can_emit_guards) {
5074 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5075 // guards but the arraycopy node could still take advantage of a
5076 // tightly allocated allocation. tightly_coupled_allocation() is
5077 // called again to make sure it takes the null check above into
5078 // account: the null check is mandatory and if it caused an
5079 // uncommon trap to be emitted then the allocation can't be
5080 // considered tightly coupled in this context.
5081 alloc = tightly_coupled_allocation(dest);
5082 }
5083
5084 bool validated = false;
5085
5086 const Type* src_type = _gvn.type(src);
5087 const Type* dest_type = _gvn.type(dest);
5088 const TypeAryPtr* top_src = src_type->isa_aryptr();
5089 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5090
5091 // Do we have the type of src?
5092 bool has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5093 // Do we have the type of dest?
5094 bool has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5095 // Is the type for src from speculation?
5096 bool src_spec = false;
5097 // Is the type for dest from speculation?
5098 bool dest_spec = false;
5099
5100 if ((!has_src || !has_dest) && can_emit_guards) {
5101 // We don't have sufficient type information, let's see if
5102 // speculative types can help. We need to have types for both src
5103 // and dest so that it pays off.
5104
5105 // Do we already have or could we have type information for src
5106 bool could_have_src = has_src;
5107 // Do we already have or could we have type information for dest
5108 bool could_have_dest = has_dest;
5109
5110 ciKlass* src_k = NULL;
5111 if (!has_src) {
5112 src_k = src_type->speculative_type_not_null();
5113 if (src_k != NULL && src_k->is_array_klass()) {
5114 could_have_src = true;
5115 }
5116 }
5117
5118 ciKlass* dest_k = NULL;
5119 if (!has_dest) {
5120 dest_k = dest_type->speculative_type_not_null();
5121 if (dest_k != NULL && dest_k->is_array_klass()) {
5122 could_have_dest = true;
5123 }
5124 }
5125
5126 if (could_have_src && could_have_dest) {
5127 // This is going to pay off so emit the required guards
5128 if (!has_src) {
5129 src = maybe_cast_profiled_obj(src, src_k, true);
5130 src_type = _gvn.type(src);
5131 top_src = src_type->isa_aryptr();
5132 has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5133 src_spec = true;
5134 }
5135 if (!has_dest) {
5136 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5137 dest_type = _gvn.type(dest);
5138 top_dest = dest_type->isa_aryptr();
5139 has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5140 dest_spec = true;
5141 }
5142 }
5143 }
5144
5145 if (has_src && has_dest && can_emit_guards) {
5146 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5147 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5148 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5149 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5150
5151 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5152 // If both arrays are object arrays then having the exact types
5153 // for both will remove the need for a subtype check at runtime
5154 // before the call and may make it possible to pick a faster copy
5155 // routine (without a subtype check on every element)
5156 // Do we have the exact type of src?
5157 bool could_have_src = src_spec;
5158 // Do we have the exact type of dest?
5159 bool could_have_dest = dest_spec;
5160 ciKlass* src_k = NULL;
5161 ciKlass* dest_k = NULL;
5162 if (!src_spec) {
5163 src_k = src_type->speculative_type_not_null();
5164 if (src_k != NULL && src_k->is_array_klass()) {
5165 could_have_src = true;
5166 }
5167 }
5168 if (!dest_spec) {
5169 dest_k = dest_type->speculative_type_not_null();
5170 if (dest_k != NULL && dest_k->is_array_klass()) {
5171 could_have_dest = true;
5172 }
5173 }
5174 if (could_have_src && could_have_dest) {
5175 // If we can have both exact types, emit the missing guards
5176 if (could_have_src && !src_spec) {
5177 src = maybe_cast_profiled_obj(src, src_k, true);
5178 }
5179 if (could_have_dest && !dest_spec) {
5180 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5181 }
5182 }
5183 }
5184 }
5185
5186 ciMethod* trap_method = method();
5187 int trap_bci = bci();
5188 if (saved_jvms != NULL) {
5189 trap_method = alloc->jvms()->method();
5190 trap_bci = alloc->jvms()->bci();
5191 }
5192
5193 bool negative_length_guard_generated = false;
5194
5195 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5196 can_emit_guards &&
5197 !src->is_top() && !dest->is_top()) {
5198 // validate arguments: enables transformation the ArrayCopyNode
5199 validated = true;
5200
5201 RegionNode* slow_region = new RegionNode(1);
5202 record_for_igvn(slow_region);
5203
5204 // (1) src and dest are arrays.
5205 generate_non_array_guard(load_object_klass(src), slow_region);
5206 generate_non_array_guard(load_object_klass(dest), slow_region);
5207
5208 // (2) src and dest arrays must have elements of the same BasicType
5209 // done at macro expansion or at Ideal transformation time
5210
5211 // (4) src_offset must not be negative.
5212 generate_negative_guard(src_offset, slow_region);
5213
5214 // (5) dest_offset must not be negative.
5215 generate_negative_guard(dest_offset, slow_region);
5216
5217 // (7) src_offset + length must not exceed length of src.
5218 generate_limit_guard(src_offset, length,
5219 load_array_length(src),
5220 slow_region);
5221
5222 // (8) dest_offset + length must not exceed length of dest.
5223 generate_limit_guard(dest_offset, length,
5224 load_array_length(dest),
5225 slow_region);
5226
5227 // (6) length must not be negative.
5228 // This is also checked in generate_arraycopy() during macro expansion, but
5229 // we also have to check it here for the case where the ArrayCopyNode will
5230 // be eliminated by Escape Analysis.
5231 if (EliminateAllocations) {
5232 generate_negative_guard(length, slow_region);
5233 negative_length_guard_generated = true;
5234 }
5235
5236 // (9) each element of an oop array must be assignable
5237 Node* dest_klass = load_object_klass(dest);
5238 if (src != dest) {
5239 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5240
5241 if (not_subtype_ctrl != top()) {
5242 PreserveJVMState pjvms(this);
5243 set_control(not_subtype_ctrl);
5244 uncommon_trap(Deoptimization::Reason_intrinsic,
5245 Deoptimization::Action_make_not_entrant);
5246 assert(stopped(), "Should be stopped");
5247 }
5248 }
5249 {
5250 PreserveJVMState pjvms(this);
5251 set_control(_gvn.transform(slow_region));
5252 uncommon_trap(Deoptimization::Reason_intrinsic,
5253 Deoptimization::Action_make_not_entrant);
5254 assert(stopped(), "Should be stopped");
5255 }
5256
5257 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5258 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5259 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5260 }
5261
5262 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5263
5264 if (stopped()) {
5265 return true;
5266 }
5267
5268 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5269 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5270 // so the compiler has a chance to eliminate them: during macro expansion,
5271 // we have to set their control (CastPP nodes are eliminated).
5272 load_object_klass(src), load_object_klass(dest),
5273 load_array_length(src), load_array_length(dest));
5274
5275 ac->set_arraycopy(validated);
5276
5277 Node* n = _gvn.transform(ac);
5278 if (n == ac) {
5279 ac->connect_outputs(this);
5280 } else {
5281 assert(validated, "shouldn't transform if all arguments not validated");
5282 set_all_memory(n);
5283 }
5284 clear_upper_avx();
5285
5286
5287 return true;
5288 }
5289
5290
5291 // Helper function which determines if an arraycopy immediately follows
5292 // an allocation, with no intervening tests or other escapes for the object.
5293 AllocateArrayNode*
5294 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5295 if (stopped()) return NULL; // no fast path
5296 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5297
5298 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5299 if (alloc == NULL) return NULL;
5300
5301 Node* rawmem = memory(Compile::AliasIdxRaw);
5302 // Is the allocation's memory state untouched?
5303 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5304 // Bail out if there have been raw-memory effects since the allocation.
5305 // (Example: There might have been a call or safepoint.)
5306 return NULL;
5307 }
5308 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5309 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5310 return NULL;
5311 }
5312
5313 // There must be no unexpected observers of this allocation.
5314 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5315 Node* obs = ptr->fast_out(i);
5316 if (obs != this->map()) {
5317 return NULL;
5318 }
5319 }
5320
5321 // This arraycopy must unconditionally follow the allocation of the ptr.
5322 Node* alloc_ctl = ptr->in(0);
5323 Node* ctl = control();
5324 while (ctl != alloc_ctl) {
5325 // There may be guards which feed into the slow_region.
5326 // Any other control flow means that we might not get a chance
5327 // to finish initializing the allocated object.
5328 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5329 IfNode* iff = ctl->in(0)->as_If();
5330 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5331 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5332 // One more try: Various low-level checks bottom out in
5333 // uncommon traps. If the debug-info of the trap omits
5334 // any reference to the allocation, as we've already
5335 // observed, then there can be no objection to the trap.
5336 bool found_trap = false;
5337 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5338 Node* obs = not_ctl->fast_out(j);
5339 if (obs->in(0) == not_ctl && obs->is_Call() &&
5340 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5341 found_trap = true; break;
5342 }
5343 }
5344 if (found_trap) {
5345 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5346 continue;
5347 }
5348 }
5349 return NULL;
5350 }
5351
5352 // If we get this far, we have an allocation which immediately
5353 // precedes the arraycopy, and we can take over zeroing the new object.
5354 // The arraycopy will finish the initialization, and provide
5355 // a new control state to which we will anchor the destination pointer.
5356
5357 return alloc;
5358 }
5359
5360 //-------------inline_encodeISOArray-----------------------------------
5361 // encode char[] to byte[] in ISO_8859_1 or ASCII
5362 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5363 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5364 // no receiver since it is static method
5365 Node *src = argument(0);
5366 Node *src_offset = argument(1);
5367 Node *dst = argument(2);
5368 Node *dst_offset = argument(3);
5369 Node *length = argument(4);
5370
5371 src = must_be_not_null(src, true);
5372 dst = must_be_not_null(dst, true);
5373
5374 const Type* src_type = src->Value(&_gvn);
5375 const Type* dst_type = dst->Value(&_gvn);
5376 const TypeAryPtr* top_src = src_type->isa_aryptr();
5377 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5378 if (top_src == NULL || top_src->elem() == Type::BOTTOM ||
5379 top_dest == NULL || top_dest->elem() == Type::BOTTOM) {
5380 // failed array check
5381 return false;
5382 }
5383
5384 // Figure out the size and type of the elements we will be copying.
5385 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5386 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
5387 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5388 return false;
5389 }
5390
5391 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5392 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5393 // 'src_start' points to src array + scaled offset
5394 // 'dst_start' points to dst array + scaled offset
5395
5396 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5397 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5398 enc = _gvn.transform(enc);
5399 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5400 set_memory(res_mem, mtype);
5401 set_result(enc);
5402 clear_upper_avx();
5403
5404 return true;
5405 }
5406
5407 //-------------inline_multiplyToLen-----------------------------------
5408 bool LibraryCallKit::inline_multiplyToLen() {
5409 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5410
5411 address stubAddr = StubRoutines::multiplyToLen();
5412 if (stubAddr == NULL) {
5413 return false; // Intrinsic's stub is not implemented on this platform
5414 }
5415 const char* stubName = "multiplyToLen";
5416
5417 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5418
5419 // no receiver because it is a static method
5420 Node* x = argument(0);
5421 Node* xlen = argument(1);
5422 Node* y = argument(2);
5423 Node* ylen = argument(3);
5424 Node* z = argument(4);
5425
5426 x = must_be_not_null(x, true);
5427 y = must_be_not_null(y, true);
5428
5429 const Type* x_type = x->Value(&_gvn);
5430 const Type* y_type = y->Value(&_gvn);
5431 const TypeAryPtr* top_x = x_type->isa_aryptr();
5432 const TypeAryPtr* top_y = y_type->isa_aryptr();
5433 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5434 top_y == NULL || top_y->elem() == Type::BOTTOM) {
5435 // failed array check
5436 return false;
5437 }
5438
5439 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5440 BasicType y_elem = y_type->isa_aryptr()->elem()->array_element_basic_type();
5441 if (x_elem != T_INT || y_elem != T_INT) {
5442 return false;
5443 }
5444
5445 // Set the original stack and the reexecute bit for the interpreter to reexecute
5446 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5447 // on the return from z array allocation in runtime.
5448 { PreserveReexecuteState preexecs(this);
5449 jvms()->set_should_reexecute(true);
5450
5451 Node* x_start = array_element_address(x, intcon(0), x_elem);
5452 Node* y_start = array_element_address(y, intcon(0), y_elem);
5453 // 'x_start' points to x array + scaled xlen
5454 // 'y_start' points to y array + scaled ylen
5455
5456 // Allocate the result array
5457 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5458 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5459 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5460
5461 IdealKit ideal(this);
5462
5463 #define __ ideal.
5464 Node* one = __ ConI(1);
5465 Node* zero = __ ConI(0);
5466 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5467 __ set(need_alloc, zero);
5468 __ set(z_alloc, z);
5469 __ if_then(z, BoolTest::eq, null()); {
5470 __ increment (need_alloc, one);
5471 } __ else_(); {
5472 // Update graphKit memory and control from IdealKit.
5473 sync_kit(ideal);
5474 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5475 cast->init_req(0, control());
5476 _gvn.set_type(cast, cast->bottom_type());
5477 C->record_for_igvn(cast);
5478
5479 Node* zlen_arg = load_array_length(cast);
5480 // Update IdealKit memory and control from graphKit.
5481 __ sync_kit(this);
5482 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5483 __ increment (need_alloc, one);
5484 } __ end_if();
5485 } __ end_if();
5486
5487 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5488 // Update graphKit memory and control from IdealKit.
5489 sync_kit(ideal);
5490 Node * narr = new_array(klass_node, zlen, 1);
5491 // Update IdealKit memory and control from graphKit.
5492 __ sync_kit(this);
5493 __ set(z_alloc, narr);
5494 } __ end_if();
5495
5496 sync_kit(ideal);
5497 z = __ value(z_alloc);
5498 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5499 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5500 // Final sync IdealKit and GraphKit.
5501 final_sync(ideal);
5502 #undef __
5503
5504 Node* z_start = array_element_address(z, intcon(0), T_INT);
5505
5506 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5507 OptoRuntime::multiplyToLen_Type(),
5508 stubAddr, stubName, TypePtr::BOTTOM,
5509 x_start, xlen, y_start, ylen, z_start, zlen);
5510 } // original reexecute is set back here
5511
5512 C->set_has_split_ifs(true); // Has chance for split-if optimization
5513 set_result(z);
5514 return true;
5515 }
5516
5517 //-------------inline_squareToLen------------------------------------
5518 bool LibraryCallKit::inline_squareToLen() {
5519 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5520
5521 address stubAddr = StubRoutines::squareToLen();
5522 if (stubAddr == NULL) {
5523 return false; // Intrinsic's stub is not implemented on this platform
5524 }
5525 const char* stubName = "squareToLen";
5526
5527 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5528
5529 Node* x = argument(0);
5530 Node* len = argument(1);
5531 Node* z = argument(2);
5532 Node* zlen = argument(3);
5533
5534 x = must_be_not_null(x, true);
5535 z = must_be_not_null(z, true);
5536
5537 const Type* x_type = x->Value(&_gvn);
5538 const Type* z_type = z->Value(&_gvn);
5539 const TypeAryPtr* top_x = x_type->isa_aryptr();
5540 const TypeAryPtr* top_z = z_type->isa_aryptr();
5541 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5542 top_z == NULL || top_z->elem() == Type::BOTTOM) {
5543 // failed array check
5544 return false;
5545 }
5546
5547 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5548 BasicType z_elem = z_type->isa_aryptr()->elem()->array_element_basic_type();
5549 if (x_elem != T_INT || z_elem != T_INT) {
5550 return false;
5551 }
5552
5553
5554 Node* x_start = array_element_address(x, intcon(0), x_elem);
5555 Node* z_start = array_element_address(z, intcon(0), z_elem);
5556
5557 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5558 OptoRuntime::squareToLen_Type(),
5559 stubAddr, stubName, TypePtr::BOTTOM,
5560 x_start, len, z_start, zlen);
5561
5562 set_result(z);
5563 return true;
5564 }
5565
5566 //-------------inline_mulAdd------------------------------------------
5567 bool LibraryCallKit::inline_mulAdd() {
5568 assert(UseMulAddIntrinsic, "not implemented on this platform");
5569
5570 address stubAddr = StubRoutines::mulAdd();
5571 if (stubAddr == NULL) {
5572 return false; // Intrinsic's stub is not implemented on this platform
5573 }
5574 const char* stubName = "mulAdd";
5575
5576 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5577
5578 Node* out = argument(0);
5579 Node* in = argument(1);
5580 Node* offset = argument(2);
5581 Node* len = argument(3);
5582 Node* k = argument(4);
5583
5584 out = must_be_not_null(out, true);
5585
5586 const Type* out_type = out->Value(&_gvn);
5587 const Type* in_type = in->Value(&_gvn);
5588 const TypeAryPtr* top_out = out_type->isa_aryptr();
5589 const TypeAryPtr* top_in = in_type->isa_aryptr();
5590 if (top_out == NULL || top_out->elem() == Type::BOTTOM ||
5591 top_in == NULL || top_in->elem() == Type::BOTTOM) {
5592 // failed array check
5593 return false;
5594 }
5595
5596 BasicType out_elem = out_type->isa_aryptr()->elem()->array_element_basic_type();
5597 BasicType in_elem = in_type->isa_aryptr()->elem()->array_element_basic_type();
5598 if (out_elem != T_INT || in_elem != T_INT) {
5599 return false;
5600 }
5601
5602 Node* outlen = load_array_length(out);
5603 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5604 Node* out_start = array_element_address(out, intcon(0), out_elem);
5605 Node* in_start = array_element_address(in, intcon(0), in_elem);
5606
5607 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5608 OptoRuntime::mulAdd_Type(),
5609 stubAddr, stubName, TypePtr::BOTTOM,
5610 out_start,in_start, new_offset, len, k);
5611 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5612 set_result(result);
5613 return true;
5614 }
5615
5616 //-------------inline_montgomeryMultiply-----------------------------------
5617 bool LibraryCallKit::inline_montgomeryMultiply() {
5618 address stubAddr = StubRoutines::montgomeryMultiply();
5619 if (stubAddr == NULL) {
5620 return false; // Intrinsic's stub is not implemented on this platform
5621 }
5622
5623 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5624 const char* stubName = "montgomery_multiply";
5625
5626 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5627
5628 Node* a = argument(0);
5629 Node* b = argument(1);
5630 Node* n = argument(2);
5631 Node* len = argument(3);
5632 Node* inv = argument(4);
5633 Node* m = argument(6);
5634
5635 const Type* a_type = a->Value(&_gvn);
5636 const TypeAryPtr* top_a = a_type->isa_aryptr();
5637 const Type* b_type = b->Value(&_gvn);
5638 const TypeAryPtr* top_b = b_type->isa_aryptr();
5639 const Type* n_type = a->Value(&_gvn);
5640 const TypeAryPtr* top_n = n_type->isa_aryptr();
5641 const Type* m_type = a->Value(&_gvn);
5642 const TypeAryPtr* top_m = m_type->isa_aryptr();
5643 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5644 top_b == NULL || top_b->elem() == Type::BOTTOM ||
5645 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5646 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5647 // failed array check
5648 return false;
5649 }
5650
5651 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5652 BasicType b_elem = b_type->isa_aryptr()->elem()->array_element_basic_type();
5653 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5654 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5655 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5656 return false;
5657 }
5658
5659 // Make the call
5660 {
5661 Node* a_start = array_element_address(a, intcon(0), a_elem);
5662 Node* b_start = array_element_address(b, intcon(0), b_elem);
5663 Node* n_start = array_element_address(n, intcon(0), n_elem);
5664 Node* m_start = array_element_address(m, intcon(0), m_elem);
5665
5666 Node* call = make_runtime_call(RC_LEAF,
5667 OptoRuntime::montgomeryMultiply_Type(),
5668 stubAddr, stubName, TypePtr::BOTTOM,
5669 a_start, b_start, n_start, len, inv, top(),
5670 m_start);
5671 set_result(m);
5672 }
5673
5674 return true;
5675 }
5676
5677 bool LibraryCallKit::inline_montgomerySquare() {
5678 address stubAddr = StubRoutines::montgomerySquare();
5679 if (stubAddr == NULL) {
5680 return false; // Intrinsic's stub is not implemented on this platform
5681 }
5682
5683 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5684 const char* stubName = "montgomery_square";
5685
5686 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5687
5688 Node* a = argument(0);
5689 Node* n = argument(1);
5690 Node* len = argument(2);
5691 Node* inv = argument(3);
5692 Node* m = argument(5);
5693
5694 const Type* a_type = a->Value(&_gvn);
5695 const TypeAryPtr* top_a = a_type->isa_aryptr();
5696 const Type* n_type = a->Value(&_gvn);
5697 const TypeAryPtr* top_n = n_type->isa_aryptr();
5698 const Type* m_type = a->Value(&_gvn);
5699 const TypeAryPtr* top_m = m_type->isa_aryptr();
5700 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5701 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5702 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5703 // failed array check
5704 return false;
5705 }
5706
5707 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5708 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5709 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5710 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5711 return false;
5712 }
5713
5714 // Make the call
5715 {
5716 Node* a_start = array_element_address(a, intcon(0), a_elem);
5717 Node* n_start = array_element_address(n, intcon(0), n_elem);
5718 Node* m_start = array_element_address(m, intcon(0), m_elem);
5719
5720 Node* call = make_runtime_call(RC_LEAF,
5721 OptoRuntime::montgomerySquare_Type(),
5722 stubAddr, stubName, TypePtr::BOTTOM,
5723 a_start, n_start, len, inv, top(),
5724 m_start);
5725 set_result(m);
5726 }
5727
5728 return true;
5729 }
5730
5731 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5732 address stubAddr = NULL;
5733 const char* stubName = NULL;
5734
5735 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5736 if (stubAddr == NULL) {
5737 return false; // Intrinsic's stub is not implemented on this platform
5738 }
5739
5740 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5741
5742 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5743
5744 Node* newArr = argument(0);
5745 Node* oldArr = argument(1);
5746 Node* newIdx = argument(2);
5747 Node* shiftCount = argument(3);
5748 Node* numIter = argument(4);
5749
5750 const Type* newArr_type = newArr->Value(&_gvn);
5751 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5752 const Type* oldArr_type = oldArr->Value(&_gvn);
5753 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5754 if (top_newArr == NULL || top_newArr->elem() == Type::BOTTOM || top_oldArr == NULL
5755 || top_oldArr->elem() == Type::BOTTOM) {
5756 return false;
5757 }
5758
5759 BasicType newArr_elem = newArr_type->isa_aryptr()->elem()->array_element_basic_type();
5760 BasicType oldArr_elem = oldArr_type->isa_aryptr()->elem()->array_element_basic_type();
5761 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5762 return false;
5763 }
5764
5765 // Make the call
5766 {
5767 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5768 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5769
5770 Node* call = make_runtime_call(RC_LEAF,
5771 OptoRuntime::bigIntegerShift_Type(),
5772 stubAddr,
5773 stubName,
5774 TypePtr::BOTTOM,
5775 newArr_start,
5776 oldArr_start,
5777 newIdx,
5778 shiftCount,
5779 numIter);
5780 }
5781
5782 return true;
5783 }
5784
5785 //-------------inline_vectorizedMismatch------------------------------
5786 bool LibraryCallKit::inline_vectorizedMismatch() {
5787 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5788
5789 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5790 Node* obja = argument(0); // Object
5791 Node* aoffset = argument(1); // long
5792 Node* objb = argument(3); // Object
5793 Node* boffset = argument(4); // long
5794 Node* length = argument(6); // int
5795 Node* scale = argument(7); // int
5796
5797 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5798 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5799 if (obja_t == NULL || obja_t->elem() == Type::BOTTOM ||
5800 objb_t == NULL || objb_t->elem() == Type::BOTTOM ||
5801 scale == top()) {
5802 return false; // failed input validation
5803 }
5804
5805 Node* obja_adr = make_unsafe_address(obja, aoffset);
5806 Node* objb_adr = make_unsafe_address(objb, boffset);
5807
5808 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5809 //
5810 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5811 // if (length <= inline_limit) {
5812 // inline_path:
5813 // vmask = VectorMaskGen length
5814 // vload1 = LoadVectorMasked obja, vmask
5815 // vload2 = LoadVectorMasked objb, vmask
5816 // result1 = VectorCmpMasked vload1, vload2, vmask
5817 // } else {
5818 // call_stub_path:
5819 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5820 // }
5821 // exit_block:
5822 // return Phi(result1, result2);
5823 //
5824 enum { inline_path = 1, // input is small enough to process it all at once
5825 stub_path = 2, // input is too large; call into the VM
5826 PATH_LIMIT = 3
5827 };
5828
5829 Node* exit_block = new RegionNode(PATH_LIMIT);
5830 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5831 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5832
5833 Node* call_stub_path = control();
5834
5835 BasicType elem_bt = T_ILLEGAL;
5836
5837 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5838 if (scale_t->is_con()) {
5839 switch (scale_t->get_con()) {
5840 case 0: elem_bt = T_BYTE; break;
5841 case 1: elem_bt = T_SHORT; break;
5842 case 2: elem_bt = T_INT; break;
5843 case 3: elem_bt = T_LONG; break;
5844
5845 default: elem_bt = T_ILLEGAL; break; // not supported
5846 }
5847 }
5848
5849 int inline_limit = 0;
5850 bool do_partial_inline = false;
5851
5852 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5853 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5854 do_partial_inline = inline_limit >= 16;
5855 }
5856
5857 if (do_partial_inline) {
5858 assert(elem_bt != T_ILLEGAL, "sanity");
5859
5860 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5861 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5862 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5863
5864 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5865 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5866 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5867
5868 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5869
5870 if (!stopped()) {
5871 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5872
5873 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5874 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5875 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5876 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5877
5878 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
5879 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5880 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5881 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5882
5883 exit_block->init_req(inline_path, control());
5884 memory_phi->init_req(inline_path, map()->memory());
5885 result_phi->init_req(inline_path, result);
5886
5887 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5888 clear_upper_avx();
5889 }
5890 }
5891 }
5892
5893 if (call_stub_path != NULL) {
5894 set_control(call_stub_path);
5895
5896 Node* call = make_runtime_call(RC_LEAF,
5897 OptoRuntime::vectorizedMismatch_Type(),
5898 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5899 obja_adr, objb_adr, length, scale);
5900
5901 exit_block->init_req(stub_path, control());
5902 memory_phi->init_req(stub_path, map()->memory());
5903 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5904 }
5905
5906 exit_block = _gvn.transform(exit_block);
5907 memory_phi = _gvn.transform(memory_phi);
5908 result_phi = _gvn.transform(result_phi);
5909
5910 set_control(exit_block);
5911 set_all_memory(memory_phi);
5912 set_result(result_phi);
5913
5914 return true;
5915 }
5916
5917 /**
5918 * Calculate CRC32 for byte.
5919 * int java.util.zip.CRC32.update(int crc, int b)
5920 */
5921 bool LibraryCallKit::inline_updateCRC32() {
5922 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5923 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5924 // no receiver since it is static method
5925 Node* crc = argument(0); // type: int
5926 Node* b = argument(1); // type: int
5927
5928 /*
5929 * int c = ~ crc;
5930 * b = timesXtoThe32[(b ^ c) & 0xFF];
5931 * b = b ^ (c >>> 8);
5932 * crc = ~b;
5933 */
5934
5935 Node* M1 = intcon(-1);
5936 crc = _gvn.transform(new XorINode(crc, M1));
5937 Node* result = _gvn.transform(new XorINode(crc, b));
5938 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5939
5940 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5941 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5942 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5943 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5944
5945 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5946 result = _gvn.transform(new XorINode(crc, result));
5947 result = _gvn.transform(new XorINode(result, M1));
5948 set_result(result);
5949 return true;
5950 }
5951
5952 /**
5953 * Calculate CRC32 for byte[] array.
5954 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5955 */
5956 bool LibraryCallKit::inline_updateBytesCRC32() {
5957 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5958 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5959 // no receiver since it is static method
5960 Node* crc = argument(0); // type: int
5961 Node* src = argument(1); // type: oop
5962 Node* offset = argument(2); // type: int
5963 Node* length = argument(3); // type: int
5964
5965 const Type* src_type = src->Value(&_gvn);
5966 const TypeAryPtr* top_src = src_type->isa_aryptr();
5967 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
5968 // failed array check
5969 return false;
5970 }
5971
5972 // Figure out the size and type of the elements we will be copying.
5973 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5974 if (src_elem != T_BYTE) {
5975 return false;
5976 }
5977
5978 // 'src_start' points to src array + scaled offset
5979 src = must_be_not_null(src, true);
5980 Node* src_start = array_element_address(src, offset, src_elem);
5981
5982 // We assume that range check is done by caller.
5983 // TODO: generate range check (offset+length < src.length) in debug VM.
5984
5985 // Call the stub.
5986 address stubAddr = StubRoutines::updateBytesCRC32();
5987 const char *stubName = "updateBytesCRC32";
5988
5989 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5990 stubAddr, stubName, TypePtr::BOTTOM,
5991 crc, src_start, length);
5992 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5993 set_result(result);
5994 return true;
5995 }
5996
5997 /**
5998 * Calculate CRC32 for ByteBuffer.
5999 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6000 */
6001 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6002 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6003 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6004 // no receiver since it is static method
6005 Node* crc = argument(0); // type: int
6006 Node* src = argument(1); // type: long
6007 Node* offset = argument(3); // type: int
6008 Node* length = argument(4); // type: int
6009
6010 src = ConvL2X(src); // adjust Java long to machine word
6011 Node* base = _gvn.transform(new CastX2PNode(src));
6012 offset = ConvI2X(offset);
6013
6014 // 'src_start' points to src array + scaled offset
6015 Node* src_start = basic_plus_adr(top(), base, offset);
6016
6017 // Call the stub.
6018 address stubAddr = StubRoutines::updateBytesCRC32();
6019 const char *stubName = "updateBytesCRC32";
6020
6021 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6022 stubAddr, stubName, TypePtr::BOTTOM,
6023 crc, src_start, length);
6024 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6025 set_result(result);
6026 return true;
6027 }
6028
6029 //------------------------------get_table_from_crc32c_class-----------------------
6030 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6031 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6032 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
6033
6034 return table;
6035 }
6036
6037 //------------------------------inline_updateBytesCRC32C-----------------------
6038 //
6039 // Calculate CRC32C for byte[] array.
6040 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6041 //
6042 bool LibraryCallKit::inline_updateBytesCRC32C() {
6043 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6044 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6045 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6046 // no receiver since it is a static method
6047 Node* crc = argument(0); // type: int
6048 Node* src = argument(1); // type: oop
6049 Node* offset = argument(2); // type: int
6050 Node* end = argument(3); // type: int
6051
6052 Node* length = _gvn.transform(new SubINode(end, offset));
6053
6054 const Type* src_type = src->Value(&_gvn);
6055 const TypeAryPtr* top_src = src_type->isa_aryptr();
6056 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6057 // failed array check
6058 return false;
6059 }
6060
6061 // Figure out the size and type of the elements we will be copying.
6062 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6063 if (src_elem != T_BYTE) {
6064 return false;
6065 }
6066
6067 // 'src_start' points to src array + scaled offset
6068 src = must_be_not_null(src, true);
6069 Node* src_start = array_element_address(src, offset, src_elem);
6070
6071 // static final int[] byteTable in class CRC32C
6072 Node* table = get_table_from_crc32c_class(callee()->holder());
6073 table = must_be_not_null(table, true);
6074 Node* table_start = array_element_address(table, intcon(0), T_INT);
6075
6076 // We assume that range check is done by caller.
6077 // TODO: generate range check (offset+length < src.length) in debug VM.
6078
6079 // Call the stub.
6080 address stubAddr = StubRoutines::updateBytesCRC32C();
6081 const char *stubName = "updateBytesCRC32C";
6082
6083 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6084 stubAddr, stubName, TypePtr::BOTTOM,
6085 crc, src_start, length, table_start);
6086 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6087 set_result(result);
6088 return true;
6089 }
6090
6091 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6092 //
6093 // Calculate CRC32C for DirectByteBuffer.
6094 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6095 //
6096 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6097 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6098 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6099 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6100 // no receiver since it is a static method
6101 Node* crc = argument(0); // type: int
6102 Node* src = argument(1); // type: long
6103 Node* offset = argument(3); // type: int
6104 Node* end = argument(4); // type: int
6105
6106 Node* length = _gvn.transform(new SubINode(end, offset));
6107
6108 src = ConvL2X(src); // adjust Java long to machine word
6109 Node* base = _gvn.transform(new CastX2PNode(src));
6110 offset = ConvI2X(offset);
6111
6112 // 'src_start' points to src array + scaled offset
6113 Node* src_start = basic_plus_adr(top(), base, offset);
6114
6115 // static final int[] byteTable in class CRC32C
6116 Node* table = get_table_from_crc32c_class(callee()->holder());
6117 table = must_be_not_null(table, true);
6118 Node* table_start = array_element_address(table, intcon(0), T_INT);
6119
6120 // Call the stub.
6121 address stubAddr = StubRoutines::updateBytesCRC32C();
6122 const char *stubName = "updateBytesCRC32C";
6123
6124 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6125 stubAddr, stubName, TypePtr::BOTTOM,
6126 crc, src_start, length, table_start);
6127 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6128 set_result(result);
6129 return true;
6130 }
6131
6132 //------------------------------inline_updateBytesAdler32----------------------
6133 //
6134 // Calculate Adler32 checksum for byte[] array.
6135 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6136 //
6137 bool LibraryCallKit::inline_updateBytesAdler32() {
6138 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6139 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6140 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6141 // no receiver since it is static method
6142 Node* crc = argument(0); // type: int
6143 Node* src = argument(1); // type: oop
6144 Node* offset = argument(2); // type: int
6145 Node* length = argument(3); // type: int
6146
6147 const Type* src_type = src->Value(&_gvn);
6148 const TypeAryPtr* top_src = src_type->isa_aryptr();
6149 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6150 // failed array check
6151 return false;
6152 }
6153
6154 // Figure out the size and type of the elements we will be copying.
6155 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6156 if (src_elem != T_BYTE) {
6157 return false;
6158 }
6159
6160 // 'src_start' points to src array + scaled offset
6161 Node* src_start = array_element_address(src, offset, src_elem);
6162
6163 // We assume that range check is done by caller.
6164 // TODO: generate range check (offset+length < src.length) in debug VM.
6165
6166 // Call the stub.
6167 address stubAddr = StubRoutines::updateBytesAdler32();
6168 const char *stubName = "updateBytesAdler32";
6169
6170 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6171 stubAddr, stubName, TypePtr::BOTTOM,
6172 crc, src_start, length);
6173 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6174 set_result(result);
6175 return true;
6176 }
6177
6178 //------------------------------inline_updateByteBufferAdler32---------------
6179 //
6180 // Calculate Adler32 checksum for DirectByteBuffer.
6181 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6182 //
6183 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6184 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6185 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6186 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6187 // no receiver since it is static method
6188 Node* crc = argument(0); // type: int
6189 Node* src = argument(1); // type: long
6190 Node* offset = argument(3); // type: int
6191 Node* length = argument(4); // type: int
6192
6193 src = ConvL2X(src); // adjust Java long to machine word
6194 Node* base = _gvn.transform(new CastX2PNode(src));
6195 offset = ConvI2X(offset);
6196
6197 // 'src_start' points to src array + scaled offset
6198 Node* src_start = basic_plus_adr(top(), base, offset);
6199
6200 // Call the stub.
6201 address stubAddr = StubRoutines::updateBytesAdler32();
6202 const char *stubName = "updateBytesAdler32";
6203
6204 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6205 stubAddr, stubName, TypePtr::BOTTOM,
6206 crc, src_start, length);
6207
6208 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6209 set_result(result);
6210 return true;
6211 }
6212
6213 //----------------------------inline_reference_get----------------------------
6214 // public T java.lang.ref.Reference.get();
6215 bool LibraryCallKit::inline_reference_get() {
6216 const int referent_offset = java_lang_ref_Reference::referent_offset();
6217
6218 // Get the argument:
6219 Node* reference_obj = null_check_receiver();
6220 if (stopped()) return true;
6221
6222 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6223 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6224 decorators, /*is_static*/ false, NULL);
6225 if (result == NULL) return false;
6226
6227 // Add memory barrier to prevent commoning reads from this field
6228 // across safepoint since GC can change its value.
6229 insert_mem_bar(Op_MemBarCPUOrder);
6230
6231 set_result(result);
6232 return true;
6233 }
6234
6235 //----------------------------inline_reference_refersTo0----------------------------
6236 // bool java.lang.ref.Reference.refersTo0();
6237 // bool java.lang.ref.PhantomReference.refersTo0();
6238 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6239 // Get arguments:
6240 Node* reference_obj = null_check_receiver();
6241 Node* other_obj = argument(1);
6242 if (stopped()) return true;
6243
6244 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6245 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6246 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6247 decorators, /*is_static*/ false, NULL);
6248 if (referent == NULL) return false;
6249
6250 // Add memory barrier to prevent commoning reads from this field
6251 // across safepoint since GC can change its value.
6252 insert_mem_bar(Op_MemBarCPUOrder);
6253
6254 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6255 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6256 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6257
6258 RegionNode* region = new RegionNode(3);
6259 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6260
6261 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6262 region->init_req(1, if_true);
6263 phi->init_req(1, intcon(1));
6264
6265 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6266 region->init_req(2, if_false);
6267 phi->init_req(2, intcon(0));
6268
6269 set_control(_gvn.transform(region));
6270 record_for_igvn(region);
6271 set_result(_gvn.transform(phi));
6272 return true;
6273 }
6274
6275
6276 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6277 DecoratorSet decorators, bool is_static,
6278 ciInstanceKlass* fromKls) {
6279 if (fromKls == NULL) {
6280 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6281 assert(tinst != NULL, "obj is null");
6282 assert(tinst->is_loaded(), "obj is not loaded");
6283 fromKls = tinst->instance_klass();
6284 } else {
6285 assert(is_static, "only for static field access");
6286 }
6287 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6288 ciSymbol::make(fieldTypeString),
6289 is_static);
6290
6291 assert(field != NULL, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6292 if (field == NULL) return (Node *) NULL;
6293
6294 if (is_static) {
6295 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6296 fromObj = makecon(tip);
6297 }
6298
6299 // Next code copied from Parse::do_get_xxx():
6300
6301 // Compute address and memory type.
6302 int offset = field->offset_in_bytes();
6303 bool is_vol = field->is_volatile();
6304 ciType* field_klass = field->type();
6305 assert(field_klass->is_loaded(), "should be loaded");
6306 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6307 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6308 BasicType bt = field->layout_type();
6309
6310 // Build the resultant type of the load
6311 const Type *type;
6312 if (bt == T_OBJECT) {
6313 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6314 } else {
6315 type = Type::get_const_basic_type(bt);
6316 }
6317
6318 if (is_vol) {
6319 decorators |= MO_SEQ_CST;
6320 }
6321
6322 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6323 }
6324
6325 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6326 bool is_exact /* true */, bool is_static /* false */,
6327 ciInstanceKlass * fromKls /* NULL */) {
6328 if (fromKls == NULL) {
6329 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6330 assert(tinst != NULL, "obj is null");
6331 assert(tinst->is_loaded(), "obj is not loaded");
6332 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6333 fromKls = tinst->instance_klass();
6334 }
6335 else {
6336 assert(is_static, "only for static field access");
6337 }
6338 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6339 ciSymbol::make(fieldTypeString),
6340 is_static);
6341
6342 assert(field != NULL, "undefined field");
6343 assert(!field->is_volatile(), "not defined for volatile fields");
6344
6345 if (is_static) {
6346 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6347 fromObj = makecon(tip);
6348 }
6349
6350 // Next code copied from Parse::do_get_xxx():
6351
6352 // Compute address and memory type.
6353 int offset = field->offset_in_bytes();
6354 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6355
6356 return adr;
6357 }
6358
6359 //------------------------------inline_aescrypt_Block-----------------------
6360 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6361 address stubAddr = NULL;
6362 const char *stubName;
6363 assert(UseAES, "need AES instruction support");
6364
6365 switch(id) {
6366 case vmIntrinsics::_aescrypt_encryptBlock:
6367 stubAddr = StubRoutines::aescrypt_encryptBlock();
6368 stubName = "aescrypt_encryptBlock";
6369 break;
6370 case vmIntrinsics::_aescrypt_decryptBlock:
6371 stubAddr = StubRoutines::aescrypt_decryptBlock();
6372 stubName = "aescrypt_decryptBlock";
6373 break;
6374 default:
6375 break;
6376 }
6377 if (stubAddr == NULL) return false;
6378
6379 Node* aescrypt_object = argument(0);
6380 Node* src = argument(1);
6381 Node* src_offset = argument(2);
6382 Node* dest = argument(3);
6383 Node* dest_offset = argument(4);
6384
6385 src = must_be_not_null(src, true);
6386 dest = must_be_not_null(dest, true);
6387
6388 // (1) src and dest are arrays.
6389 const Type* src_type = src->Value(&_gvn);
6390 const Type* dest_type = dest->Value(&_gvn);
6391 const TypeAryPtr* top_src = src_type->isa_aryptr();
6392 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6393 assert (top_src != NULL && top_src->elem() != Type::BOTTOM && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6394
6395 // for the quick and dirty code we will skip all the checks.
6396 // we are just trying to get the call to be generated.
6397 Node* src_start = src;
6398 Node* dest_start = dest;
6399 if (src_offset != NULL || dest_offset != NULL) {
6400 assert(src_offset != NULL && dest_offset != NULL, "");
6401 src_start = array_element_address(src, src_offset, T_BYTE);
6402 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6403 }
6404
6405 // now need to get the start of its expanded key array
6406 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6407 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6408 if (k_start == NULL) return false;
6409
6410 // Call the stub.
6411 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6412 stubAddr, stubName, TypePtr::BOTTOM,
6413 src_start, dest_start, k_start);
6414
6415 return true;
6416 }
6417
6418 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6419 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6420 address stubAddr = NULL;
6421 const char *stubName = NULL;
6422
6423 assert(UseAES, "need AES instruction support");
6424
6425 switch(id) {
6426 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6427 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6428 stubName = "cipherBlockChaining_encryptAESCrypt";
6429 break;
6430 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6431 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6432 stubName = "cipherBlockChaining_decryptAESCrypt";
6433 break;
6434 default:
6435 break;
6436 }
6437 if (stubAddr == NULL) return false;
6438
6439 Node* cipherBlockChaining_object = argument(0);
6440 Node* src = argument(1);
6441 Node* src_offset = argument(2);
6442 Node* len = argument(3);
6443 Node* dest = argument(4);
6444 Node* dest_offset = argument(5);
6445
6446 src = must_be_not_null(src, false);
6447 dest = must_be_not_null(dest, false);
6448
6449 // (1) src and dest are arrays.
6450 const Type* src_type = src->Value(&_gvn);
6451 const Type* dest_type = dest->Value(&_gvn);
6452 const TypeAryPtr* top_src = src_type->isa_aryptr();
6453 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6454 assert (top_src != NULL && top_src->elem() != Type::BOTTOM
6455 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6456
6457 // checks are the responsibility of the caller
6458 Node* src_start = src;
6459 Node* dest_start = dest;
6460 if (src_offset != NULL || dest_offset != NULL) {
6461 assert(src_offset != NULL && dest_offset != NULL, "");
6462 src_start = array_element_address(src, src_offset, T_BYTE);
6463 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6464 }
6465
6466 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6467 // (because of the predicated logic executed earlier).
6468 // so we cast it here safely.
6469 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6470
6471 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6472 if (embeddedCipherObj == NULL) return false;
6473
6474 // cast it to what we know it will be at runtime
6475 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6476 assert(tinst != NULL, "CBC obj is null");
6477 assert(tinst->is_loaded(), "CBC obj is not loaded");
6478 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6479 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6480
6481 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6482 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6483 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6484 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6485 aescrypt_object = _gvn.transform(aescrypt_object);
6486
6487 // we need to get the start of the aescrypt_object's expanded key array
6488 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6489 if (k_start == NULL) return false;
6490
6491 // similarly, get the start address of the r vector
6492 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6493 if (objRvec == NULL) return false;
6494 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6495
6496 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6497 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6498 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6499 stubAddr, stubName, TypePtr::BOTTOM,
6500 src_start, dest_start, k_start, r_start, len);
6501
6502 // return cipher length (int)
6503 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6504 set_result(retvalue);
6505 return true;
6506 }
6507
6508 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6509 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6510 address stubAddr = NULL;
6511 const char *stubName = NULL;
6512
6513 assert(UseAES, "need AES instruction support");
6514
6515 switch (id) {
6516 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6517 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6518 stubName = "electronicCodeBook_encryptAESCrypt";
6519 break;
6520 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6521 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6522 stubName = "electronicCodeBook_decryptAESCrypt";
6523 break;
6524 default:
6525 break;
6526 }
6527
6528 if (stubAddr == NULL) return false;
6529
6530 Node* electronicCodeBook_object = argument(0);
6531 Node* src = argument(1);
6532 Node* src_offset = argument(2);
6533 Node* len = argument(3);
6534 Node* dest = argument(4);
6535 Node* dest_offset = argument(5);
6536
6537 // (1) src and dest are arrays.
6538 const Type* src_type = src->Value(&_gvn);
6539 const Type* dest_type = dest->Value(&_gvn);
6540 const TypeAryPtr* top_src = src_type->isa_aryptr();
6541 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6542 assert(top_src != NULL && top_src->elem() != Type::BOTTOM
6543 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6544
6545 // checks are the responsibility of the caller
6546 Node* src_start = src;
6547 Node* dest_start = dest;
6548 if (src_offset != NULL || dest_offset != NULL) {
6549 assert(src_offset != NULL && dest_offset != NULL, "");
6550 src_start = array_element_address(src, src_offset, T_BYTE);
6551 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6552 }
6553
6554 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6555 // (because of the predicated logic executed earlier).
6556 // so we cast it here safely.
6557 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6558
6559 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6560 if (embeddedCipherObj == NULL) return false;
6561
6562 // cast it to what we know it will be at runtime
6563 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6564 assert(tinst != NULL, "ECB obj is null");
6565 assert(tinst->is_loaded(), "ECB obj is not loaded");
6566 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6567 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6568
6569 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6570 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6571 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6572 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6573 aescrypt_object = _gvn.transform(aescrypt_object);
6574
6575 // we need to get the start of the aescrypt_object's expanded key array
6576 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6577 if (k_start == NULL) return false;
6578
6579 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6580 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6581 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6582 stubAddr, stubName, TypePtr::BOTTOM,
6583 src_start, dest_start, k_start, len);
6584
6585 // return cipher length (int)
6586 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6587 set_result(retvalue);
6588 return true;
6589 }
6590
6591 //------------------------------inline_counterMode_AESCrypt-----------------------
6592 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6593 assert(UseAES, "need AES instruction support");
6594 if (!UseAESCTRIntrinsics) return false;
6595
6596 address stubAddr = NULL;
6597 const char *stubName = NULL;
6598 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6599 stubAddr = StubRoutines::counterMode_AESCrypt();
6600 stubName = "counterMode_AESCrypt";
6601 }
6602 if (stubAddr == NULL) return false;
6603
6604 Node* counterMode_object = argument(0);
6605 Node* src = argument(1);
6606 Node* src_offset = argument(2);
6607 Node* len = argument(3);
6608 Node* dest = argument(4);
6609 Node* dest_offset = argument(5);
6610
6611 // (1) src and dest are arrays.
6612 const Type* src_type = src->Value(&_gvn);
6613 const Type* dest_type = dest->Value(&_gvn);
6614 const TypeAryPtr* top_src = src_type->isa_aryptr();
6615 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6616 assert(top_src != NULL && top_src->elem() != Type::BOTTOM &&
6617 top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6618
6619 // checks are the responsibility of the caller
6620 Node* src_start = src;
6621 Node* dest_start = dest;
6622 if (src_offset != NULL || dest_offset != NULL) {
6623 assert(src_offset != NULL && dest_offset != NULL, "");
6624 src_start = array_element_address(src, src_offset, T_BYTE);
6625 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6626 }
6627
6628 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6629 // (because of the predicated logic executed earlier).
6630 // so we cast it here safely.
6631 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6632 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6633 if (embeddedCipherObj == NULL) return false;
6634 // cast it to what we know it will be at runtime
6635 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6636 assert(tinst != NULL, "CTR obj is null");
6637 assert(tinst->is_loaded(), "CTR obj is not loaded");
6638 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6639 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6640 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6641 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6642 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6643 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6644 aescrypt_object = _gvn.transform(aescrypt_object);
6645 // we need to get the start of the aescrypt_object's expanded key array
6646 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6647 if (k_start == NULL) return false;
6648 // similarly, get the start address of the r vector
6649 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6650 if (obj_counter == NULL) return false;
6651 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6652
6653 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6654 if (saved_encCounter == NULL) return false;
6655 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6656 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6657
6658 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6659 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6660 OptoRuntime::counterMode_aescrypt_Type(),
6661 stubAddr, stubName, TypePtr::BOTTOM,
6662 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6663
6664 // return cipher length (int)
6665 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6666 set_result(retvalue);
6667 return true;
6668 }
6669
6670 //------------------------------get_key_start_from_aescrypt_object-----------------------
6671 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6672 #if defined(PPC64) || defined(S390)
6673 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6674 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6675 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6676 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6677 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6678 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6679 if (objSessionK == NULL) {
6680 return (Node *) NULL;
6681 }
6682 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6683 #else
6684 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6685 #endif // PPC64
6686 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6687 if (objAESCryptKey == NULL) return (Node *) NULL;
6688
6689 // now have the array, need to get the start address of the K array
6690 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6691 return k_start;
6692 }
6693
6694 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6695 // Return node representing slow path of predicate check.
6696 // the pseudo code we want to emulate with this predicate is:
6697 // for encryption:
6698 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6699 // for decryption:
6700 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6701 // note cipher==plain is more conservative than the original java code but that's OK
6702 //
6703 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6704 // The receiver was checked for NULL already.
6705 Node* objCBC = argument(0);
6706
6707 Node* src = argument(1);
6708 Node* dest = argument(4);
6709
6710 // Load embeddedCipher field of CipherBlockChaining object.
6711 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6712
6713 // get AESCrypt klass for instanceOf check
6714 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6715 // will have same classloader as CipherBlockChaining object
6716 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6717 assert(tinst != NULL, "CBCobj is null");
6718 assert(tinst->is_loaded(), "CBCobj is not loaded");
6719
6720 // we want to do an instanceof comparison against the AESCrypt class
6721 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6722 if (!klass_AESCrypt->is_loaded()) {
6723 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6724 Node* ctrl = control();
6725 set_control(top()); // no regular fast path
6726 return ctrl;
6727 }
6728
6729 src = must_be_not_null(src, true);
6730 dest = must_be_not_null(dest, true);
6731
6732 // Resolve oops to stable for CmpP below.
6733 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6734
6735 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6736 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6737 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6738
6739 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6740
6741 // for encryption, we are done
6742 if (!decrypting)
6743 return instof_false; // even if it is NULL
6744
6745 // for decryption, we need to add a further check to avoid
6746 // taking the intrinsic path when cipher and plain are the same
6747 // see the original java code for why.
6748 RegionNode* region = new RegionNode(3);
6749 region->init_req(1, instof_false);
6750
6751 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6752 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6753 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6754 region->init_req(2, src_dest_conjoint);
6755
6756 record_for_igvn(region);
6757 return _gvn.transform(region);
6758 }
6759
6760 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6761 // Return node representing slow path of predicate check.
6762 // the pseudo code we want to emulate with this predicate is:
6763 // for encryption:
6764 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6765 // for decryption:
6766 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6767 // note cipher==plain is more conservative than the original java code but that's OK
6768 //
6769 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6770 // The receiver was checked for NULL already.
6771 Node* objECB = argument(0);
6772
6773 // Load embeddedCipher field of ElectronicCodeBook object.
6774 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6775
6776 // get AESCrypt klass for instanceOf check
6777 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6778 // will have same classloader as ElectronicCodeBook object
6779 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6780 assert(tinst != NULL, "ECBobj is null");
6781 assert(tinst->is_loaded(), "ECBobj is not loaded");
6782
6783 // we want to do an instanceof comparison against the AESCrypt class
6784 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6785 if (!klass_AESCrypt->is_loaded()) {
6786 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6787 Node* ctrl = control();
6788 set_control(top()); // no regular fast path
6789 return ctrl;
6790 }
6791 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6792
6793 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6794 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6795 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6796
6797 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6798
6799 // for encryption, we are done
6800 if (!decrypting)
6801 return instof_false; // even if it is NULL
6802
6803 // for decryption, we need to add a further check to avoid
6804 // taking the intrinsic path when cipher and plain are the same
6805 // see the original java code for why.
6806 RegionNode* region = new RegionNode(3);
6807 region->init_req(1, instof_false);
6808 Node* src = argument(1);
6809 Node* dest = argument(4);
6810 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6811 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6812 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6813 region->init_req(2, src_dest_conjoint);
6814
6815 record_for_igvn(region);
6816 return _gvn.transform(region);
6817 }
6818
6819 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6820 // Return node representing slow path of predicate check.
6821 // the pseudo code we want to emulate with this predicate is:
6822 // for encryption:
6823 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6824 // for decryption:
6825 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6826 // note cipher==plain is more conservative than the original java code but that's OK
6827 //
6828
6829 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6830 // The receiver was checked for NULL already.
6831 Node* objCTR = argument(0);
6832
6833 // Load embeddedCipher field of CipherBlockChaining object.
6834 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6835
6836 // get AESCrypt klass for instanceOf check
6837 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6838 // will have same classloader as CipherBlockChaining object
6839 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6840 assert(tinst != NULL, "CTRobj is null");
6841 assert(tinst->is_loaded(), "CTRobj is not loaded");
6842
6843 // we want to do an instanceof comparison against the AESCrypt class
6844 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6845 if (!klass_AESCrypt->is_loaded()) {
6846 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6847 Node* ctrl = control();
6848 set_control(top()); // no regular fast path
6849 return ctrl;
6850 }
6851
6852 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6853 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6854 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6855 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6856 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6857
6858 return instof_false; // even if it is NULL
6859 }
6860
6861 //------------------------------inline_ghash_processBlocks
6862 bool LibraryCallKit::inline_ghash_processBlocks() {
6863 address stubAddr;
6864 const char *stubName;
6865 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6866
6867 stubAddr = StubRoutines::ghash_processBlocks();
6868 stubName = "ghash_processBlocks";
6869
6870 Node* data = argument(0);
6871 Node* offset = argument(1);
6872 Node* len = argument(2);
6873 Node* state = argument(3);
6874 Node* subkeyH = argument(4);
6875
6876 state = must_be_not_null(state, true);
6877 subkeyH = must_be_not_null(subkeyH, true);
6878 data = must_be_not_null(data, true);
6879
6880 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6881 assert(state_start, "state is NULL");
6882 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6883 assert(subkeyH_start, "subkeyH is NULL");
6884 Node* data_start = array_element_address(data, offset, T_BYTE);
6885 assert(data_start, "data is NULL");
6886
6887 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6888 OptoRuntime::ghash_processBlocks_Type(),
6889 stubAddr, stubName, TypePtr::BOTTOM,
6890 state_start, subkeyH_start, data_start, len);
6891 return true;
6892 }
6893
6894 bool LibraryCallKit::inline_base64_encodeBlock() {
6895 address stubAddr;
6896 const char *stubName;
6897 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6898 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6899 stubAddr = StubRoutines::base64_encodeBlock();
6900 stubName = "encodeBlock";
6901
6902 if (!stubAddr) return false;
6903 Node* base64obj = argument(0);
6904 Node* src = argument(1);
6905 Node* offset = argument(2);
6906 Node* len = argument(3);
6907 Node* dest = argument(4);
6908 Node* dp = argument(5);
6909 Node* isURL = argument(6);
6910
6911 src = must_be_not_null(src, true);
6912 dest = must_be_not_null(dest, true);
6913
6914 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6915 assert(src_start, "source array is NULL");
6916 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6917 assert(dest_start, "destination array is NULL");
6918
6919 Node* base64 = make_runtime_call(RC_LEAF,
6920 OptoRuntime::base64_encodeBlock_Type(),
6921 stubAddr, stubName, TypePtr::BOTTOM,
6922 src_start, offset, len, dest_start, dp, isURL);
6923 return true;
6924 }
6925
6926 bool LibraryCallKit::inline_base64_decodeBlock() {
6927 address stubAddr;
6928 const char *stubName;
6929 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6930 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6931 stubAddr = StubRoutines::base64_decodeBlock();
6932 stubName = "decodeBlock";
6933
6934 if (!stubAddr) return false;
6935 Node* base64obj = argument(0);
6936 Node* src = argument(1);
6937 Node* src_offset = argument(2);
6938 Node* len = argument(3);
6939 Node* dest = argument(4);
6940 Node* dest_offset = argument(5);
6941 Node* isURL = argument(6);
6942 Node* isMIME = argument(7);
6943
6944 src = must_be_not_null(src, true);
6945 dest = must_be_not_null(dest, true);
6946
6947 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6948 assert(src_start, "source array is NULL");
6949 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6950 assert(dest_start, "destination array is NULL");
6951
6952 Node* call = make_runtime_call(RC_LEAF,
6953 OptoRuntime::base64_decodeBlock_Type(),
6954 stubAddr, stubName, TypePtr::BOTTOM,
6955 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6956 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6957 set_result(result);
6958 return true;
6959 }
6960
6961 //------------------------------inline_digestBase_implCompress-----------------------
6962 //
6963 // Calculate MD5 for single-block byte[] array.
6964 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6965 //
6966 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6967 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6968 //
6969 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6970 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6971 //
6972 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6973 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6974 //
6975 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6976 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6977 //
6978 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6979 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6980
6981 Node* digestBase_obj = argument(0);
6982 Node* src = argument(1); // type oop
6983 Node* ofs = argument(2); // type int
6984
6985 const Type* src_type = src->Value(&_gvn);
6986 const TypeAryPtr* top_src = src_type->isa_aryptr();
6987 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6988 // failed array check
6989 return false;
6990 }
6991 // Figure out the size and type of the elements we will be copying.
6992 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6993 if (src_elem != T_BYTE) {
6994 return false;
6995 }
6996 // 'src_start' points to src array + offset
6997 src = must_be_not_null(src, true);
6998 Node* src_start = array_element_address(src, ofs, src_elem);
6999 Node* state = NULL;
7000 Node* digest_length = NULL;
7001 address stubAddr;
7002 const char *stubName;
7003
7004 switch(id) {
7005 case vmIntrinsics::_md5_implCompress:
7006 assert(UseMD5Intrinsics, "need MD5 instruction support");
7007 state = get_state_from_digest_object(digestBase_obj, T_INT);
7008 stubAddr = StubRoutines::md5_implCompress();
7009 stubName = "md5_implCompress";
7010 break;
7011 case vmIntrinsics::_sha_implCompress:
7012 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7013 state = get_state_from_digest_object(digestBase_obj, T_INT);
7014 stubAddr = StubRoutines::sha1_implCompress();
7015 stubName = "sha1_implCompress";
7016 break;
7017 case vmIntrinsics::_sha2_implCompress:
7018 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7019 state = get_state_from_digest_object(digestBase_obj, T_INT);
7020 stubAddr = StubRoutines::sha256_implCompress();
7021 stubName = "sha256_implCompress";
7022 break;
7023 case vmIntrinsics::_sha5_implCompress:
7024 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7025 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7026 stubAddr = StubRoutines::sha512_implCompress();
7027 stubName = "sha512_implCompress";
7028 break;
7029 case vmIntrinsics::_sha3_implCompress:
7030 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7031 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7032 stubAddr = StubRoutines::sha3_implCompress();
7033 stubName = "sha3_implCompress";
7034 digest_length = get_digest_length_from_digest_object(digestBase_obj);
7035 if (digest_length == NULL) return false;
7036 break;
7037 default:
7038 fatal_unexpected_iid(id);
7039 return false;
7040 }
7041 if (state == NULL) return false;
7042
7043 assert(stubAddr != NULL, "Stub is generated");
7044 if (stubAddr == NULL) return false;
7045
7046 // Call the stub.
7047 Node* call;
7048 if (digest_length == NULL) {
7049 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7050 stubAddr, stubName, TypePtr::BOTTOM,
7051 src_start, state);
7052 } else {
7053 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7054 stubAddr, stubName, TypePtr::BOTTOM,
7055 src_start, state, digest_length);
7056 }
7057
7058 return true;
7059 }
7060
7061 //------------------------------inline_digestBase_implCompressMB-----------------------
7062 //
7063 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7064 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7065 //
7066 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7067 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7068 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7069 assert((uint)predicate < 5, "sanity");
7070 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7071
7072 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7073 Node* src = argument(1); // byte[] array
7074 Node* ofs = argument(2); // type int
7075 Node* limit = argument(3); // type int
7076
7077 const Type* src_type = src->Value(&_gvn);
7078 const TypeAryPtr* top_src = src_type->isa_aryptr();
7079 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7080 // failed array check
7081 return false;
7082 }
7083 // Figure out the size and type of the elements we will be copying.
7084 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7085 if (src_elem != T_BYTE) {
7086 return false;
7087 }
7088 // 'src_start' points to src array + offset
7089 src = must_be_not_null(src, false);
7090 Node* src_start = array_element_address(src, ofs, src_elem);
7091
7092 const char* klass_digestBase_name = NULL;
7093 const char* stub_name = NULL;
7094 address stub_addr = NULL;
7095 BasicType elem_type = T_INT;
7096
7097 switch (predicate) {
7098 case 0:
7099 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7100 klass_digestBase_name = "sun/security/provider/MD5";
7101 stub_name = "md5_implCompressMB";
7102 stub_addr = StubRoutines::md5_implCompressMB();
7103 }
7104 break;
7105 case 1:
7106 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7107 klass_digestBase_name = "sun/security/provider/SHA";
7108 stub_name = "sha1_implCompressMB";
7109 stub_addr = StubRoutines::sha1_implCompressMB();
7110 }
7111 break;
7112 case 2:
7113 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7114 klass_digestBase_name = "sun/security/provider/SHA2";
7115 stub_name = "sha256_implCompressMB";
7116 stub_addr = StubRoutines::sha256_implCompressMB();
7117 }
7118 break;
7119 case 3:
7120 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7121 klass_digestBase_name = "sun/security/provider/SHA5";
7122 stub_name = "sha512_implCompressMB";
7123 stub_addr = StubRoutines::sha512_implCompressMB();
7124 elem_type = T_LONG;
7125 }
7126 break;
7127 case 4:
7128 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7129 klass_digestBase_name = "sun/security/provider/SHA3";
7130 stub_name = "sha3_implCompressMB";
7131 stub_addr = StubRoutines::sha3_implCompressMB();
7132 elem_type = T_BYTE;
7133 }
7134 break;
7135 default:
7136 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7137 }
7138 if (klass_digestBase_name != NULL) {
7139 assert(stub_addr != NULL, "Stub is generated");
7140 if (stub_addr == NULL) return false;
7141
7142 // get DigestBase klass to lookup for SHA klass
7143 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7144 assert(tinst != NULL, "digestBase_obj is not instance???");
7145 assert(tinst->is_loaded(), "DigestBase is not loaded");
7146
7147 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7148 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7149 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7150 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7151 }
7152 return false;
7153 }
7154
7155 //------------------------------inline_digestBase_implCompressMB-----------------------
7156 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7157 BasicType elem_type, address stubAddr, const char *stubName,
7158 Node* src_start, Node* ofs, Node* limit) {
7159 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7160 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7161 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7162 digest_obj = _gvn.transform(digest_obj);
7163
7164 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7165 if (state == NULL) return false;
7166
7167 Node* digest_length = NULL;
7168 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7169 digest_length = get_digest_length_from_digest_object(digest_obj);
7170 if (digest_length == NULL) return false;
7171 }
7172
7173 // Call the stub.
7174 Node* call;
7175 if (digest_length == NULL) {
7176 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7177 OptoRuntime::digestBase_implCompressMB_Type(false),
7178 stubAddr, stubName, TypePtr::BOTTOM,
7179 src_start, state, ofs, limit);
7180 } else {
7181 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7182 OptoRuntime::digestBase_implCompressMB_Type(true),
7183 stubAddr, stubName, TypePtr::BOTTOM,
7184 src_start, state, digest_length, ofs, limit);
7185 }
7186
7187 // return ofs (int)
7188 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7189 set_result(result);
7190
7191 return true;
7192 }
7193
7194 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7195 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7196 assert(UseAES, "need AES instruction support");
7197 address stubAddr = NULL;
7198 const char *stubName = NULL;
7199 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7200 stubName = "galoisCounterMode_AESCrypt";
7201
7202 if (stubAddr == NULL) return false;
7203
7204 Node* in = argument(0);
7205 Node* inOfs = argument(1);
7206 Node* len = argument(2);
7207 Node* ct = argument(3);
7208 Node* ctOfs = argument(4);
7209 Node* out = argument(5);
7210 Node* outOfs = argument(6);
7211 Node* gctr_object = argument(7);
7212 Node* ghash_object = argument(8);
7213
7214 // (1) in, ct and out are arrays.
7215 const Type* in_type = in->Value(&_gvn);
7216 const Type* ct_type = ct->Value(&_gvn);
7217 const Type* out_type = out->Value(&_gvn);
7218 const TypeAryPtr* top_in = in_type->isa_aryptr();
7219 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7220 const TypeAryPtr* top_out = out_type->isa_aryptr();
7221 assert(top_in != NULL && top_in->elem() != Type::BOTTOM &&
7222 top_ct != NULL && top_ct->elem() != Type::BOTTOM &&
7223 top_out != NULL && top_out->elem() != Type::BOTTOM, "args are strange");
7224
7225 // checks are the responsibility of the caller
7226 Node* in_start = in;
7227 Node* ct_start = ct;
7228 Node* out_start = out;
7229 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7230 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7231 in_start = array_element_address(in, inOfs, T_BYTE);
7232 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7233 out_start = array_element_address(out, outOfs, T_BYTE);
7234 }
7235
7236 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7237 // (because of the predicated logic executed earlier).
7238 // so we cast it here safely.
7239 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7240 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7241 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7242 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7243 Node* state = load_field_from_object(ghash_object, "state", "[J");
7244
7245 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7246 return false;
7247 }
7248 // cast it to what we know it will be at runtime
7249 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7250 assert(tinst != NULL, "GCTR obj is null");
7251 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7252 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7253 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7254 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7255 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7256 const TypeOopPtr* xtype = aklass->as_instance_type();
7257 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7258 aescrypt_object = _gvn.transform(aescrypt_object);
7259 // we need to get the start of the aescrypt_object's expanded key array
7260 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7261 if (k_start == NULL) return false;
7262 // similarly, get the start address of the r vector
7263 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7264 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7265 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7266
7267
7268 // Call the stub, passing params
7269 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7270 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7271 stubAddr, stubName, TypePtr::BOTTOM,
7272 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7273
7274 // return cipher length (int)
7275 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7276 set_result(retvalue);
7277
7278 return true;
7279 }
7280
7281 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7282 // Return node representing slow path of predicate check.
7283 // the pseudo code we want to emulate with this predicate is:
7284 // for encryption:
7285 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7286 // for decryption:
7287 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7288 // note cipher==plain is more conservative than the original java code but that's OK
7289 //
7290
7291 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7292 // The receiver was checked for NULL already.
7293 Node* objGCTR = argument(7);
7294 // Load embeddedCipher field of GCTR object.
7295 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7296 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7297
7298 // get AESCrypt klass for instanceOf check
7299 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7300 // will have same classloader as CipherBlockChaining object
7301 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7302 assert(tinst != NULL, "GCTR obj is null");
7303 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7304
7305 // we want to do an instanceof comparison against the AESCrypt class
7306 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7307 if (!klass_AESCrypt->is_loaded()) {
7308 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7309 Node* ctrl = control();
7310 set_control(top()); // no regular fast path
7311 return ctrl;
7312 }
7313
7314 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7315 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7316 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7317 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7318 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7319
7320 return instof_false; // even if it is NULL
7321 }
7322
7323 //------------------------------get_state_from_digest_object-----------------------
7324 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7325 const char* state_type;
7326 switch (elem_type) {
7327 case T_BYTE: state_type = "[B"; break;
7328 case T_INT: state_type = "[I"; break;
7329 case T_LONG: state_type = "[J"; break;
7330 default: ShouldNotReachHere();
7331 }
7332 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7333 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7334 if (digest_state == NULL) return (Node *) NULL;
7335
7336 // now have the array, need to get the start address of the state array
7337 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7338 return state;
7339 }
7340
7341 //------------------------------get_digest_length_from_sha3_object----------------------------------
7342 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
7343 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
7344 assert (digest_length != NULL, "sanity");
7345 return digest_length;
7346 }
7347
7348 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7349 // Return node representing slow path of predicate check.
7350 // the pseudo code we want to emulate with this predicate is:
7351 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7352 //
7353 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7354 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7355 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7356 assert((uint)predicate < 5, "sanity");
7357
7358 // The receiver was checked for NULL already.
7359 Node* digestBaseObj = argument(0);
7360
7361 // get DigestBase klass for instanceOf check
7362 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7363 assert(tinst != NULL, "digestBaseObj is null");
7364 assert(tinst->is_loaded(), "DigestBase is not loaded");
7365
7366 const char* klass_name = NULL;
7367 switch (predicate) {
7368 case 0:
7369 if (UseMD5Intrinsics) {
7370 // we want to do an instanceof comparison against the MD5 class
7371 klass_name = "sun/security/provider/MD5";
7372 }
7373 break;
7374 case 1:
7375 if (UseSHA1Intrinsics) {
7376 // we want to do an instanceof comparison against the SHA class
7377 klass_name = "sun/security/provider/SHA";
7378 }
7379 break;
7380 case 2:
7381 if (UseSHA256Intrinsics) {
7382 // we want to do an instanceof comparison against the SHA2 class
7383 klass_name = "sun/security/provider/SHA2";
7384 }
7385 break;
7386 case 3:
7387 if (UseSHA512Intrinsics) {
7388 // we want to do an instanceof comparison against the SHA5 class
7389 klass_name = "sun/security/provider/SHA5";
7390 }
7391 break;
7392 case 4:
7393 if (UseSHA3Intrinsics) {
7394 // we want to do an instanceof comparison against the SHA3 class
7395 klass_name = "sun/security/provider/SHA3";
7396 }
7397 break;
7398 default:
7399 fatal("unknown SHA intrinsic predicate: %d", predicate);
7400 }
7401
7402 ciKlass* klass = NULL;
7403 if (klass_name != NULL) {
7404 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7405 }
7406 if ((klass == NULL) || !klass->is_loaded()) {
7407 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7408 Node* ctrl = control();
7409 set_control(top()); // no intrinsic path
7410 return ctrl;
7411 }
7412 ciInstanceKlass* instklass = klass->as_instance_klass();
7413
7414 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7415 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7416 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7417 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7418
7419 return instof_false; // even if it is NULL
7420 }
7421
7422 bool LibraryCallKit::inline_continuation_do_yield() {
7423 address call_addr = StubRoutines::cont_doYield();
7424 const TypeFunc* tf = OptoRuntime::continuation_doYield_Type();
7425 Node* call = make_runtime_call(RC_NO_LEAF, tf, call_addr, "doYield", TypeRawPtr::BOTTOM);
7426 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7427 set_result(result);
7428 return true;
7429 }
7430
7431 //-------------inline_fma-----------------------------------
7432 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7433 Node *a = NULL;
7434 Node *b = NULL;
7435 Node *c = NULL;
7436 Node* result = NULL;
7437 switch (id) {
7438 case vmIntrinsics::_fmaD:
7439 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7440 // no receiver since it is static method
7441 a = round_double_node(argument(0));
7442 b = round_double_node(argument(2));
7443 c = round_double_node(argument(4));
7444 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7445 break;
7446 case vmIntrinsics::_fmaF:
7447 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7448 a = argument(0);
7449 b = argument(1);
7450 c = argument(2);
7451 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7452 break;
7453 default:
7454 fatal_unexpected_iid(id); break;
7455 }
7456 set_result(result);
7457 return true;
7458 }
7459
7460 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7461 // argument(0) is receiver
7462 Node* codePoint = argument(1);
7463 Node* n = NULL;
7464
7465 switch (id) {
7466 case vmIntrinsics::_isDigit :
7467 n = new DigitNode(control(), codePoint);
7468 break;
7469 case vmIntrinsics::_isLowerCase :
7470 n = new LowerCaseNode(control(), codePoint);
7471 break;
7472 case vmIntrinsics::_isUpperCase :
7473 n = new UpperCaseNode(control(), codePoint);
7474 break;
7475 case vmIntrinsics::_isWhitespace :
7476 n = new WhitespaceNode(control(), codePoint);
7477 break;
7478 default:
7479 fatal_unexpected_iid(id);
7480 }
7481
7482 set_result(_gvn.transform(n));
7483 return true;
7484 }
7485
7486 //------------------------------inline_fp_min_max------------------------------
7487 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7488 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7489
7490 // The intrinsic should be used only when the API branches aren't predictable,
7491 // the last one performing the most important comparison. The following heuristic
7492 // uses the branch statistics to eventually bail out if necessary.
7493
7494 ciMethodData *md = callee()->method_data();
7495
7496 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7497 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7498
7499 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7500 // Bail out if the call-site didn't contribute enough to the statistics.
7501 return false;
7502 }
7503
7504 uint taken = 0, not_taken = 0;
7505
7506 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7507 if (p->is_BranchData()) {
7508 taken = ((ciBranchData*)p)->taken();
7509 not_taken = ((ciBranchData*)p)->not_taken();
7510 }
7511 }
7512
7513 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7514 balance = balance < 0 ? -balance : balance;
7515 if ( balance > 0.2 ) {
7516 // Bail out if the most important branch is predictable enough.
7517 return false;
7518 }
7519 }
7520 */
7521
7522 Node *a = NULL;
7523 Node *b = NULL;
7524 Node *n = NULL;
7525 switch (id) {
7526 case vmIntrinsics::_maxF:
7527 case vmIntrinsics::_minF:
7528 case vmIntrinsics::_maxF_strict:
7529 case vmIntrinsics::_minF_strict:
7530 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7531 a = argument(0);
7532 b = argument(1);
7533 break;
7534 case vmIntrinsics::_maxD:
7535 case vmIntrinsics::_minD:
7536 case vmIntrinsics::_maxD_strict:
7537 case vmIntrinsics::_minD_strict:
7538 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7539 a = round_double_node(argument(0));
7540 b = round_double_node(argument(2));
7541 break;
7542 default:
7543 fatal_unexpected_iid(id);
7544 break;
7545 }
7546 switch (id) {
7547 case vmIntrinsics::_maxF:
7548 case vmIntrinsics::_maxF_strict:
7549 n = new MaxFNode(a, b);
7550 break;
7551 case vmIntrinsics::_minF:
7552 case vmIntrinsics::_minF_strict:
7553 n = new MinFNode(a, b);
7554 break;
7555 case vmIntrinsics::_maxD:
7556 case vmIntrinsics::_maxD_strict:
7557 n = new MaxDNode(a, b);
7558 break;
7559 case vmIntrinsics::_minD:
7560 case vmIntrinsics::_minD_strict:
7561 n = new MinDNode(a, b);
7562 break;
7563 default:
7564 fatal_unexpected_iid(id);
7565 break;
7566 }
7567 set_result(_gvn.transform(n));
7568 return true;
7569 }
7570
7571 bool LibraryCallKit::inline_profileBoolean() {
7572 Node* counts = argument(1);
7573 const TypeAryPtr* ary = NULL;
7574 ciArray* aobj = NULL;
7575 if (counts->is_Con()
7576 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7577 && (aobj = ary->const_oop()->as_array()) != NULL
7578 && (aobj->length() == 2)) {
7579 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7580 jint false_cnt = aobj->element_value(0).as_int();
7581 jint true_cnt = aobj->element_value(1).as_int();
7582
7583 if (C->log() != NULL) {
7584 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7585 false_cnt, true_cnt);
7586 }
7587
7588 if (false_cnt + true_cnt == 0) {
7589 // According to profile, never executed.
7590 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7591 Deoptimization::Action_reinterpret);
7592 return true;
7593 }
7594
7595 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7596 // is a number of each value occurrences.
7597 Node* result = argument(0);
7598 if (false_cnt == 0 || true_cnt == 0) {
7599 // According to profile, one value has been never seen.
7600 int expected_val = (false_cnt == 0) ? 1 : 0;
7601
7602 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7603 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7604
7605 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7606 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7607 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7608
7609 { // Slow path: uncommon trap for never seen value and then reexecute
7610 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7611 // the value has been seen at least once.
7612 PreserveJVMState pjvms(this);
7613 PreserveReexecuteState preexecs(this);
7614 jvms()->set_should_reexecute(true);
7615
7616 set_control(slow_path);
7617 set_i_o(i_o());
7618
7619 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7620 Deoptimization::Action_reinterpret);
7621 }
7622 // The guard for never seen value enables sharpening of the result and
7623 // returning a constant. It allows to eliminate branches on the same value
7624 // later on.
7625 set_control(fast_path);
7626 result = intcon(expected_val);
7627 }
7628 // Stop profiling.
7629 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7630 // By replacing method body with profile data (represented as ProfileBooleanNode
7631 // on IR level) we effectively disable profiling.
7632 // It enables full speed execution once optimized code is generated.
7633 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7634 C->record_for_igvn(profile);
7635 set_result(profile);
7636 return true;
7637 } else {
7638 // Continue profiling.
7639 // Profile data isn't available at the moment. So, execute method's bytecode version.
7640 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7641 // is compiled and counters aren't available since corresponding MethodHandle
7642 // isn't a compile-time constant.
7643 return false;
7644 }
7645 }
7646
7647 bool LibraryCallKit::inline_isCompileConstant() {
7648 Node* n = argument(0);
7649 set_result(n->is_Con() ? intcon(1) : intcon(0));
7650 return true;
7651 }
7652
7653 //------------------------------- inline_getObjectSize/sizeOf --------------------------------------
7654 //
7655 // Calculate the runtime size of the object/array.
7656 //
7657 // long java.lang.Runtime.sizeOf(Object obj)
7658 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7659 //
7660 bool LibraryCallKit::inline_sizeOf() {
7661 Node* obj = argument(0);
7662 return inline_sizeOf_impl(obj);
7663 }
7664
7665 bool LibraryCallKit::inline_getObjectSize() {
7666 Node* obj = argument(3);
7667 return inline_sizeOf_impl(obj);
7668 }
7669
7670 bool LibraryCallKit::inline_sizeOf_impl(Node* obj) {
7671 Node* klass_node = load_object_klass(obj);
7672
7673 jint layout_con = Klass::_lh_neutral_value;
7674 Node* layout_val = get_layout_helper(klass_node, layout_con);
7675 int layout_is_con = (layout_val == NULL);
7676
7677 if (layout_is_con) {
7678 // Layout helper is constant, can figure out things at compile time.
7679
7680 if (Klass::layout_helper_is_instance(layout_con)) {
7681 // Instance case: layout_con contains the size itself.
7682 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7683 set_result(size);
7684 } else {
7685 // Array case: size is round(header + element_size*arraylength).
7686 // Since arraylength is different for every array instance, we have to
7687 // compute the whole thing at runtime.
7688
7689 Node* arr_length = load_array_length(obj);
7690
7691 int round_mask = MinObjAlignmentInBytes - 1;
7692 int hsize = Klass::layout_helper_header_size(layout_con);
7693 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7694
7695 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7696 round_mask = 0; // strength-reduce it if it goes away completely
7697 }
7698 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7699 Node* header_size = intcon(hsize + round_mask);
7700
7701 Node* lengthx = ConvI2X(arr_length);
7702 Node* headerx = ConvI2X(header_size);
7703
7704 Node* abody = lengthx;
7705 if (eshift != 0) {
7706 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7707 }
7708 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7709 if (round_mask != 0) {
7710 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7711 }
7712 size = ConvX2L(size);
7713 set_result(size);
7714 }
7715 } else {
7716 // Layout helper is not constant, need to test for array-ness at runtime.
7717
7718 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7719 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7720 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7721 record_for_igvn(result_reg);
7722
7723 Node* array_ctl = generate_array_guard(klass_node, NULL);
7724 if (array_ctl != NULL) {
7725 // Array case: size is round(header + element_size*arraylength).
7726 // Since arraylength is different for every array instance, we have to
7727 // compute the whole thing at runtime.
7728
7729 PreserveJVMState pjvms(this);
7730 set_control(array_ctl);
7731 Node* arr_length = load_array_length(obj);
7732
7733 int round_mask = MinObjAlignmentInBytes - 1;
7734 Node* mask = intcon(round_mask);
7735
7736 Node* hss = intcon(Klass::_lh_header_size_shift);
7737 Node* hsm = intcon(Klass::_lh_header_size_mask);
7738 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7739 header_size = _gvn.transform(new AndINode(header_size, hsm));
7740 header_size = _gvn.transform(new AddINode(header_size, mask));
7741
7742 // There is no need to mask or shift this value.
7743 // The semantics of LShiftINode include an implicit mask to 0x1F.
7744 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7745 Node* elem_shift = layout_val;
7746
7747 Node* lengthx = ConvI2X(arr_length);
7748 Node* headerx = ConvI2X(header_size);
7749
7750 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7751 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7752 if (round_mask != 0) {
7753 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7754 }
7755 size = ConvX2L(size);
7756
7757 result_reg->init_req(_array_path, control());
7758 result_val->init_req(_array_path, size);
7759 }
7760
7761 if (!stopped()) {
7762 // Instance case: the layout helper gives us instance size almost directly,
7763 // but we need to mask out the _lh_instance_slow_path_bit.
7764 Node* size = ConvI2X(layout_val);
7765 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7766 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7767 size = _gvn.transform(new AndXNode(size, mask));
7768 size = ConvX2L(size);
7769
7770 result_reg->init_req(_instance_path, control());
7771 result_val->init_req(_instance_path, size);
7772 }
7773
7774 set_result(result_reg, result_val);
7775 }
7776
7777 return true;
7778 }
7779
7780 //------------------------------- inline_blackhole --------------------------------------
7781 //
7782 // Make sure all arguments to this node are alive.
7783 // This matches methods that were requested to be blackholed through compile commands.
7784 //
7785 bool LibraryCallKit::inline_blackhole() {
7786 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7787 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7788 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7789
7790 // Bind call arguments as blackhole arguments to keep them alive
7791 Node* bh = insert_mem_bar(Op_Blackhole);
7792 uint nargs = callee()->arg_size();
7793 for (uint i = 0; i < nargs; i++) {
7794 bh->add_req(argument(i));
7795 }
7796
7797 return true;
7798 }
7799
7800 bool LibraryCallKit::inline_timestamp(bool serial) {
7801 insert_mem_bar(Op_MemBarCPUOrder);
7802 Node* node = serial ?
7803 _gvn.transform(new TimestampSerialNode(control())) :
7804 _gvn.transform(new TimestampNode(control()));
7805 set_result(node);
7806 insert_mem_bar(Op_MemBarCPUOrder);
7807 return true;
7808 }
7809
7810 bool LibraryCallKit::inline_addressOf() {
7811 Node* obj = argument(0);
7812 Node* raw_val = _gvn.transform(new CastP2XNode(NULL, obj));
7813 Node* long_val = ConvX2L(raw_val);
7814 set_result(long_val);
7815 return true;
7816 }