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.freeze();
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.freeze();
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::_scopedValueCache: return inline_native_scopedValueCache();
476 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
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:
518 case vmIntrinsics::_floatToFloat16:
519 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
520
521 case vmIntrinsics::_floatIsFinite:
522 case vmIntrinsics::_floatIsInfinite:
523 case vmIntrinsics::_doubleIsFinite:
524 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
525
526 case vmIntrinsics::_numberOfLeadingZeros_i:
527 case vmIntrinsics::_numberOfLeadingZeros_l:
528 case vmIntrinsics::_numberOfTrailingZeros_i:
529 case vmIntrinsics::_numberOfTrailingZeros_l:
530 case vmIntrinsics::_bitCount_i:
531 case vmIntrinsics::_bitCount_l:
532 case vmIntrinsics::_reverse_i:
533 case vmIntrinsics::_reverse_l:
534 case vmIntrinsics::_reverseBytes_i:
535 case vmIntrinsics::_reverseBytes_l:
536 case vmIntrinsics::_reverseBytes_s:
537 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
538
539 case vmIntrinsics::_compress_i:
540 case vmIntrinsics::_compress_l:
541 case vmIntrinsics::_expand_i:
542 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
543
544 case vmIntrinsics::_compareUnsigned_i:
545 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
546
547 case vmIntrinsics::_divideUnsigned_i:
548 case vmIntrinsics::_divideUnsigned_l:
549 case vmIntrinsics::_remainderUnsigned_i:
550 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
551
552 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
553
554 case vmIntrinsics::_Reference_get: return inline_reference_get();
555 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
556 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
557
558 case vmIntrinsics::_Class_cast: return inline_Class_cast();
559
560 case vmIntrinsics::_aescrypt_encryptBlock:
561 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
562
563 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
564 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
565 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
566
567 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
568 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
569 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
570
571 case vmIntrinsics::_counterMode_AESCrypt:
572 return inline_counterMode_AESCrypt(intrinsic_id());
573
574 case vmIntrinsics::_galoisCounterMode_AESCrypt:
575 return inline_galoisCounterMode_AESCrypt();
576
577 case vmIntrinsics::_md5_implCompress:
578 case vmIntrinsics::_sha_implCompress:
579 case vmIntrinsics::_sha2_implCompress:
580 case vmIntrinsics::_sha5_implCompress:
581 case vmIntrinsics::_sha3_implCompress:
582 return inline_digestBase_implCompress(intrinsic_id());
583
584 case vmIntrinsics::_digestBase_implCompressMB:
585 return inline_digestBase_implCompressMB(predicate);
586
587 case vmIntrinsics::_multiplyToLen:
588 return inline_multiplyToLen();
589
590 case vmIntrinsics::_squareToLen:
591 return inline_squareToLen();
592
593 case vmIntrinsics::_mulAdd:
594 return inline_mulAdd();
595
596 case vmIntrinsics::_montgomeryMultiply:
597 return inline_montgomeryMultiply();
598 case vmIntrinsics::_montgomerySquare:
599 return inline_montgomerySquare();
600
601 case vmIntrinsics::_bigIntegerRightShiftWorker:
602 return inline_bigIntegerShift(true);
603 case vmIntrinsics::_bigIntegerLeftShiftWorker:
604 return inline_bigIntegerShift(false);
605
606 case vmIntrinsics::_vectorizedMismatch:
607 return inline_vectorizedMismatch();
608
609 case vmIntrinsics::_ghash_processBlocks:
610 return inline_ghash_processBlocks();
611 case vmIntrinsics::_chacha20Block:
612 return inline_chacha20Block();
613 case vmIntrinsics::_base64_encodeBlock:
614 return inline_base64_encodeBlock();
615 case vmIntrinsics::_base64_decodeBlock:
616 return inline_base64_decodeBlock();
617 case vmIntrinsics::_poly1305_processBlocks:
618 return inline_poly1305_processBlocks();
619
620 case vmIntrinsics::_encodeISOArray:
621 case vmIntrinsics::_encodeByteISOArray:
622 return inline_encodeISOArray(false);
623 case vmIntrinsics::_encodeAsciiArray:
624 return inline_encodeISOArray(true);
625
626 case vmIntrinsics::_updateCRC32:
627 return inline_updateCRC32();
628 case vmIntrinsics::_updateBytesCRC32:
629 return inline_updateBytesCRC32();
630 case vmIntrinsics::_updateByteBufferCRC32:
631 return inline_updateByteBufferCRC32();
632
633 case vmIntrinsics::_updateBytesCRC32C:
634 return inline_updateBytesCRC32C();
635 case vmIntrinsics::_updateDirectByteBufferCRC32C:
636 return inline_updateDirectByteBufferCRC32C();
637
638 case vmIntrinsics::_updateBytesAdler32:
639 return inline_updateBytesAdler32();
640 case vmIntrinsics::_updateByteBufferAdler32:
641 return inline_updateByteBufferAdler32();
642
643 case vmIntrinsics::_profileBoolean:
644 return inline_profileBoolean();
645 case vmIntrinsics::_isCompileConstant:
646 return inline_isCompileConstant();
647
648 case vmIntrinsics::_countPositives:
649 return inline_countPositives();
650
651 case vmIntrinsics::_fmaD:
652 case vmIntrinsics::_fmaF:
653 return inline_fma(intrinsic_id());
654
655 case vmIntrinsics::_isDigit:
656 case vmIntrinsics::_isLowerCase:
657 case vmIntrinsics::_isUpperCase:
658 case vmIntrinsics::_isWhitespace:
659 return inline_character_compare(intrinsic_id());
660
661 case vmIntrinsics::_min:
662 case vmIntrinsics::_max:
663 case vmIntrinsics::_min_strict:
664 case vmIntrinsics::_max_strict:
665 return inline_min_max(intrinsic_id());
666
667 case vmIntrinsics::_maxF:
668 case vmIntrinsics::_minF:
669 case vmIntrinsics::_maxD:
670 case vmIntrinsics::_minD:
671 case vmIntrinsics::_maxF_strict:
672 case vmIntrinsics::_minF_strict:
673 case vmIntrinsics::_maxD_strict:
674 case vmIntrinsics::_minD_strict:
675 return inline_fp_min_max(intrinsic_id());
676
677 case vmIntrinsics::_VectorUnaryOp:
678 return inline_vector_nary_operation(1);
679 case vmIntrinsics::_VectorBinaryOp:
680 return inline_vector_nary_operation(2);
681 case vmIntrinsics::_VectorTernaryOp:
682 return inline_vector_nary_operation(3);
683 case vmIntrinsics::_VectorFromBitsCoerced:
684 return inline_vector_frombits_coerced();
685 case vmIntrinsics::_VectorShuffleIota:
686 return inline_vector_shuffle_iota();
687 case vmIntrinsics::_VectorMaskOp:
688 return inline_vector_mask_operation();
689 case vmIntrinsics::_VectorShuffleToVector:
690 return inline_vector_shuffle_to_vector();
691 case vmIntrinsics::_VectorLoadOp:
692 return inline_vector_mem_operation(/*is_store=*/false);
693 case vmIntrinsics::_VectorLoadMaskedOp:
694 return inline_vector_mem_masked_operation(/*is_store*/false);
695 case vmIntrinsics::_VectorStoreOp:
696 return inline_vector_mem_operation(/*is_store=*/true);
697 case vmIntrinsics::_VectorStoreMaskedOp:
698 return inline_vector_mem_masked_operation(/*is_store=*/true);
699 case vmIntrinsics::_VectorGatherOp:
700 return inline_vector_gather_scatter(/*is_scatter*/ false);
701 case vmIntrinsics::_VectorScatterOp:
702 return inline_vector_gather_scatter(/*is_scatter*/ true);
703 case vmIntrinsics::_VectorReductionCoerced:
704 return inline_vector_reduction();
705 case vmIntrinsics::_VectorTest:
706 return inline_vector_test();
707 case vmIntrinsics::_VectorBlend:
708 return inline_vector_blend();
709 case vmIntrinsics::_VectorRearrange:
710 return inline_vector_rearrange();
711 case vmIntrinsics::_VectorCompare:
712 return inline_vector_compare();
713 case vmIntrinsics::_VectorBroadcastInt:
714 return inline_vector_broadcast_int();
715 case vmIntrinsics::_VectorConvert:
716 return inline_vector_convert();
717 case vmIntrinsics::_VectorInsert:
718 return inline_vector_insert();
719 case vmIntrinsics::_VectorExtract:
720 return inline_vector_extract();
721 case vmIntrinsics::_VectorCompressExpand:
722 return inline_vector_compress_expand();
723 case vmIntrinsics::_IndexVector:
724 return inline_index_vector();
725
726 case vmIntrinsics::_getObjectSize:
727 return inline_getObjectSize();
728
729 case vmIntrinsics::_blackhole:
730 return inline_blackhole();
731
732 default:
733 // If you get here, it may be that someone has added a new intrinsic
734 // to the list in vmIntrinsics.hpp without implementing it here.
735 #ifndef PRODUCT
736 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
737 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
738 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
739 }
740 #endif
741 return false;
742 }
743 }
744
745 Node* LibraryCallKit::try_to_predicate(int predicate) {
746 if (!jvms()->has_method()) {
747 // Root JVMState has a null method.
748 assert(map()->memory()->Opcode() == Op_Parm, "");
749 // Insert the memory aliasing node
750 set_all_memory(reset_memory());
751 }
752 assert(merged_memory(), "");
753
754 switch (intrinsic_id()) {
755 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
756 return inline_cipherBlockChaining_AESCrypt_predicate(false);
757 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
758 return inline_cipherBlockChaining_AESCrypt_predicate(true);
759 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
760 return inline_electronicCodeBook_AESCrypt_predicate(false);
761 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
762 return inline_electronicCodeBook_AESCrypt_predicate(true);
763 case vmIntrinsics::_counterMode_AESCrypt:
764 return inline_counterMode_AESCrypt_predicate();
765 case vmIntrinsics::_digestBase_implCompressMB:
766 return inline_digestBase_implCompressMB_predicate(predicate);
767 case vmIntrinsics::_galoisCounterMode_AESCrypt:
768 return inline_galoisCounterMode_AESCrypt_predicate();
769
770 default:
771 // If you get here, it may be that someone has added a new intrinsic
772 // to the list in vmIntrinsics.hpp without implementing it here.
773 #ifndef PRODUCT
774 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
775 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
776 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
777 }
778 #endif
779 Node* slow_ctl = control();
780 set_control(top()); // No fast path intrinsic
781 return slow_ctl;
782 }
783 }
784
785 //------------------------------set_result-------------------------------
786 // Helper function for finishing intrinsics.
787 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
788 record_for_igvn(region);
789 set_control(_gvn.transform(region));
790 set_result( _gvn.transform(value));
791 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
792 }
793
794 //------------------------------generate_guard---------------------------
795 // Helper function for generating guarded fast-slow graph structures.
796 // The given 'test', if true, guards a slow path. If the test fails
797 // then a fast path can be taken. (We generally hope it fails.)
798 // In all cases, GraphKit::control() is updated to the fast path.
799 // The returned value represents the control for the slow path.
800 // The return value is never 'top'; it is either a valid control
801 // or NULL if it is obvious that the slow path can never be taken.
802 // Also, if region and the slow control are not NULL, the slow edge
803 // is appended to the region.
804 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
805 if (stopped()) {
806 // Already short circuited.
807 return NULL;
808 }
809
810 // Build an if node and its projections.
811 // If test is true we take the slow path, which we assume is uncommon.
812 if (_gvn.type(test) == TypeInt::ZERO) {
813 // The slow branch is never taken. No need to build this guard.
814 return NULL;
815 }
816
817 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
818
819 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
820 if (if_slow == top()) {
821 // The slow branch is never taken. No need to build this guard.
822 return NULL;
823 }
824
825 if (region != NULL)
826 region->add_req(if_slow);
827
828 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
829 set_control(if_fast);
830
831 return if_slow;
832 }
833
834 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
835 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
836 }
837 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
838 return generate_guard(test, region, PROB_FAIR);
839 }
840
841 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
842 Node* *pos_index) {
843 if (stopped())
844 return NULL; // already stopped
845 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
846 return NULL; // index is already adequately typed
847 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
848 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
849 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
850 if (is_neg != NULL && pos_index != NULL) {
851 // Emulate effect of Parse::adjust_map_after_if.
852 Node* ccast = new CastIINode(index, TypeInt::POS);
853 ccast->set_req(0, control());
854 (*pos_index) = _gvn.transform(ccast);
855 }
856 return is_neg;
857 }
858
859 // Make sure that 'position' is a valid limit index, in [0..length].
860 // There are two equivalent plans for checking this:
861 // A. (offset + copyLength) unsigned<= arrayLength
862 // B. offset <= (arrayLength - copyLength)
863 // We require that all of the values above, except for the sum and
864 // difference, are already known to be non-negative.
865 // Plan A is robust in the face of overflow, if offset and copyLength
866 // are both hugely positive.
867 //
868 // Plan B is less direct and intuitive, but it does not overflow at
869 // all, since the difference of two non-negatives is always
870 // representable. Whenever Java methods must perform the equivalent
871 // check they generally use Plan B instead of Plan A.
872 // For the moment we use Plan A.
873 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
874 Node* subseq_length,
875 Node* array_length,
876 RegionNode* region) {
877 if (stopped())
878 return NULL; // already stopped
879 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
880 if (zero_offset && subseq_length->eqv_uncast(array_length))
881 return NULL; // common case of whole-array copy
882 Node* last = subseq_length;
883 if (!zero_offset) // last += offset
884 last = _gvn.transform(new AddINode(last, offset));
885 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
886 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
887 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
888 return is_over;
889 }
890
891 // Emit range checks for the given String.value byte array
892 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
893 if (stopped()) {
894 return; // already stopped
895 }
896 RegionNode* bailout = new RegionNode(1);
897 record_for_igvn(bailout);
898 if (char_count) {
899 // Convert char count to byte count
900 count = _gvn.transform(new LShiftINode(count, intcon(1)));
901 }
902
903 // Offset and count must not be negative
904 generate_negative_guard(offset, bailout);
905 generate_negative_guard(count, bailout);
906 // Offset + count must not exceed length of array
907 generate_limit_guard(offset, count, load_array_length(array), bailout);
908
909 if (bailout->req() > 1) {
910 PreserveJVMState pjvms(this);
911 set_control(_gvn.transform(bailout));
912 uncommon_trap(Deoptimization::Reason_intrinsic,
913 Deoptimization::Action_maybe_recompile);
914 }
915 }
916
917 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
918 bool is_immutable) {
919 ciKlass* thread_klass = env()->Thread_klass();
920 const Type* thread_type
921 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
922
923 Node* thread = _gvn.transform(new ThreadLocalNode());
924 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
925 tls_output = thread;
926
927 Node* thread_obj_handle
928 = (is_immutable
929 ? LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
930 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
931 : make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
932 thread_obj_handle = _gvn.transform(thread_obj_handle);
933
934 DecoratorSet decorators = IN_NATIVE;
935 if (is_immutable) {
936 decorators |= C2_IMMUTABLE_MEMORY;
937 }
938 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
939 }
940
941 //--------------------------generate_current_thread--------------------
942 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
943 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
944 /*is_immutable*/false);
945 }
946
947 //--------------------------generate_virtual_thread--------------------
948 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
949 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
950 !C->method()->changes_current_thread());
951 }
952
953 //------------------------------make_string_method_node------------------------
954 // Helper method for String intrinsic functions. This version is called with
955 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
956 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
957 // containing the lengths of str1 and str2.
958 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
959 Node* result = NULL;
960 switch (opcode) {
961 case Op_StrIndexOf:
962 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
963 str1_start, cnt1, str2_start, cnt2, ae);
964 break;
965 case Op_StrComp:
966 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
967 str1_start, cnt1, str2_start, cnt2, ae);
968 break;
969 case Op_StrEquals:
970 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
971 // Use the constant length if there is one because optimized match rule may exist.
972 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
973 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
974 break;
975 default:
976 ShouldNotReachHere();
977 return NULL;
978 }
979
980 // All these intrinsics have checks.
981 C->set_has_split_ifs(true); // Has chance for split-if optimization
982 clear_upper_avx();
983
984 return _gvn.transform(result);
985 }
986
987 //------------------------------inline_string_compareTo------------------------
988 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
989 Node* arg1 = argument(0);
990 Node* arg2 = argument(1);
991
992 arg1 = must_be_not_null(arg1, true);
993 arg2 = must_be_not_null(arg2, true);
994
995 // Get start addr and length of first argument
996 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
997 Node* arg1_cnt = load_array_length(arg1);
998
999 // Get start addr and length of second argument
1000 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1001 Node* arg2_cnt = load_array_length(arg2);
1002
1003 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1004 set_result(result);
1005 return true;
1006 }
1007
1008 //------------------------------inline_string_equals------------------------
1009 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1010 Node* arg1 = argument(0);
1011 Node* arg2 = argument(1);
1012
1013 // paths (plus control) merge
1014 RegionNode* region = new RegionNode(3);
1015 Node* phi = new PhiNode(region, TypeInt::BOOL);
1016
1017 if (!stopped()) {
1018
1019 arg1 = must_be_not_null(arg1, true);
1020 arg2 = must_be_not_null(arg2, true);
1021
1022 // Get start addr and length of first argument
1023 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1024 Node* arg1_cnt = load_array_length(arg1);
1025
1026 // Get start addr and length of second argument
1027 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1028 Node* arg2_cnt = load_array_length(arg2);
1029
1030 // Check for arg1_cnt != arg2_cnt
1031 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1032 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1033 Node* if_ne = generate_slow_guard(bol, NULL);
1034 if (if_ne != NULL) {
1035 phi->init_req(2, intcon(0));
1036 region->init_req(2, if_ne);
1037 }
1038
1039 // Check for count == 0 is done by assembler code for StrEquals.
1040
1041 if (!stopped()) {
1042 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1043 phi->init_req(1, equals);
1044 region->init_req(1, control());
1045 }
1046 }
1047
1048 // post merge
1049 set_control(_gvn.transform(region));
1050 record_for_igvn(region);
1051
1052 set_result(_gvn.transform(phi));
1053 return true;
1054 }
1055
1056 //------------------------------inline_array_equals----------------------------
1057 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1058 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1059 Node* arg1 = argument(0);
1060 Node* arg2 = argument(1);
1061
1062 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1063 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1064 clear_upper_avx();
1065
1066 return true;
1067 }
1068
1069 //------------------------------inline_countPositives------------------------------
1070 bool LibraryCallKit::inline_countPositives() {
1071 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1072 return false;
1073 }
1074
1075 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1076 // no receiver since it is static method
1077 Node* ba = argument(0);
1078 Node* offset = argument(1);
1079 Node* len = argument(2);
1080
1081 ba = must_be_not_null(ba, true);
1082
1083 // Range checks
1084 generate_string_range_check(ba, offset, len, false);
1085 if (stopped()) {
1086 return true;
1087 }
1088 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1089 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1090 set_result(_gvn.transform(result));
1091 return true;
1092 }
1093
1094 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1095 Node* index = argument(0);
1096 Node* length = bt == T_INT ? argument(1) : argument(2);
1097 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1098 return false;
1099 }
1100
1101 // check that length is positive
1102 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1103 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1104
1105 {
1106 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1107 uncommon_trap(Deoptimization::Reason_intrinsic,
1108 Deoptimization::Action_make_not_entrant);
1109 }
1110
1111 if (stopped()) {
1112 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1113 return true;
1114 }
1115
1116 // length is now known positive, add a cast node to make this explicit
1117 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1118 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1119 casted_length = _gvn.transform(casted_length);
1120 replace_in_map(length, casted_length);
1121 length = casted_length;
1122
1123 // Use an unsigned comparison for the range check itself
1124 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1125 BoolTest::mask btest = BoolTest::lt;
1126 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1127 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1128 _gvn.set_type(rc, rc->Value(&_gvn));
1129 if (!rc_bool->is_Con()) {
1130 record_for_igvn(rc);
1131 }
1132 set_control(_gvn.transform(new IfTrueNode(rc)));
1133 {
1134 PreserveJVMState pjvms(this);
1135 set_control(_gvn.transform(new IfFalseNode(rc)));
1136 uncommon_trap(Deoptimization::Reason_range_check,
1137 Deoptimization::Action_make_not_entrant);
1138 }
1139
1140 if (stopped()) {
1141 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1142 return true;
1143 }
1144
1145 // index is now known to be >= 0 and < length, cast it
1146 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1147 result = _gvn.transform(result);
1148 set_result(result);
1149 replace_in_map(index, result);
1150 return true;
1151 }
1152
1153 //------------------------------inline_string_indexOf------------------------
1154 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1155 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1156 return false;
1157 }
1158 Node* src = argument(0);
1159 Node* tgt = argument(1);
1160
1161 // Make the merge point
1162 RegionNode* result_rgn = new RegionNode(4);
1163 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1164
1165 src = must_be_not_null(src, true);
1166 tgt = must_be_not_null(tgt, true);
1167
1168 // Get start addr and length of source string
1169 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1170 Node* src_count = load_array_length(src);
1171
1172 // Get start addr and length of substring
1173 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1174 Node* tgt_count = load_array_length(tgt);
1175
1176 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1177 // Divide src size by 2 if String is UTF16 encoded
1178 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1179 }
1180 if (ae == StrIntrinsicNode::UU) {
1181 // Divide substring size by 2 if String is UTF16 encoded
1182 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1183 }
1184
1185 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1186 if (result != NULL) {
1187 result_phi->init_req(3, result);
1188 result_rgn->init_req(3, control());
1189 }
1190 set_control(_gvn.transform(result_rgn));
1191 record_for_igvn(result_rgn);
1192 set_result(_gvn.transform(result_phi));
1193
1194 return true;
1195 }
1196
1197 //-----------------------------inline_string_indexOf-----------------------
1198 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1199 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1200 return false;
1201 }
1202 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1203 return false;
1204 }
1205 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1206 Node* src = argument(0); // byte[]
1207 Node* src_count = argument(1); // char count
1208 Node* tgt = argument(2); // byte[]
1209 Node* tgt_count = argument(3); // char count
1210 Node* from_index = argument(4); // char index
1211
1212 src = must_be_not_null(src, true);
1213 tgt = must_be_not_null(tgt, true);
1214
1215 // Multiply byte array index by 2 if String is UTF16 encoded
1216 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1217 src_count = _gvn.transform(new SubINode(src_count, from_index));
1218 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1219 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1220
1221 // Range checks
1222 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1223 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1224 if (stopped()) {
1225 return true;
1226 }
1227
1228 RegionNode* region = new RegionNode(5);
1229 Node* phi = new PhiNode(region, TypeInt::INT);
1230
1231 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1232 if (result != NULL) {
1233 // The result is index relative to from_index if substring was found, -1 otherwise.
1234 // Generate code which will fold into cmove.
1235 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1236 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1237
1238 Node* if_lt = generate_slow_guard(bol, NULL);
1239 if (if_lt != NULL) {
1240 // result == -1
1241 phi->init_req(3, result);
1242 region->init_req(3, if_lt);
1243 }
1244 if (!stopped()) {
1245 result = _gvn.transform(new AddINode(result, from_index));
1246 phi->init_req(4, result);
1247 region->init_req(4, control());
1248 }
1249 }
1250
1251 set_control(_gvn.transform(region));
1252 record_for_igvn(region);
1253 set_result(_gvn.transform(phi));
1254 clear_upper_avx();
1255
1256 return true;
1257 }
1258
1259 // Create StrIndexOfNode with fast path checks
1260 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1261 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1262 // Check for substr count > string count
1263 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1264 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1265 Node* if_gt = generate_slow_guard(bol, NULL);
1266 if (if_gt != NULL) {
1267 phi->init_req(1, intcon(-1));
1268 region->init_req(1, if_gt);
1269 }
1270 if (!stopped()) {
1271 // Check for substr count == 0
1272 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1273 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1274 Node* if_zero = generate_slow_guard(bol, NULL);
1275 if (if_zero != NULL) {
1276 phi->init_req(2, intcon(0));
1277 region->init_req(2, if_zero);
1278 }
1279 }
1280 if (!stopped()) {
1281 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1282 }
1283 return NULL;
1284 }
1285
1286 //-----------------------------inline_string_indexOfChar-----------------------
1287 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1288 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1289 return false;
1290 }
1291 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1292 return false;
1293 }
1294 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1295 Node* src = argument(0); // byte[]
1296 Node* tgt = argument(1); // tgt is int ch
1297 Node* from_index = argument(2);
1298 Node* max = argument(3);
1299
1300 src = must_be_not_null(src, true);
1301
1302 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1303 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1304 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1305
1306 // Range checks
1307 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1308 if (stopped()) {
1309 return true;
1310 }
1311
1312 RegionNode* region = new RegionNode(3);
1313 Node* phi = new PhiNode(region, TypeInt::INT);
1314
1315 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1316 C->set_has_split_ifs(true); // Has chance for split-if optimization
1317 _gvn.transform(result);
1318
1319 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1320 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1321
1322 Node* if_lt = generate_slow_guard(bol, NULL);
1323 if (if_lt != NULL) {
1324 // result == -1
1325 phi->init_req(2, result);
1326 region->init_req(2, if_lt);
1327 }
1328 if (!stopped()) {
1329 result = _gvn.transform(new AddINode(result, from_index));
1330 phi->init_req(1, result);
1331 region->init_req(1, control());
1332 }
1333 set_control(_gvn.transform(region));
1334 record_for_igvn(region);
1335 set_result(_gvn.transform(phi));
1336
1337 return true;
1338 }
1339 //---------------------------inline_string_copy---------------------
1340 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1341 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1342 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1343 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1344 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1345 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1346 bool LibraryCallKit::inline_string_copy(bool compress) {
1347 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1348 return false;
1349 }
1350 int nargs = 5; // 2 oops, 3 ints
1351 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1352
1353 Node* src = argument(0);
1354 Node* src_offset = argument(1);
1355 Node* dst = argument(2);
1356 Node* dst_offset = argument(3);
1357 Node* length = argument(4);
1358
1359 // Check for allocation before we add nodes that would confuse
1360 // tightly_coupled_allocation()
1361 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1362
1363 // Figure out the size and type of the elements we will be copying.
1364 const Type* src_type = src->Value(&_gvn);
1365 const Type* dst_type = dst->Value(&_gvn);
1366 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
1367 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
1368 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1369 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1370 "Unsupported array types for inline_string_copy");
1371
1372 src = must_be_not_null(src, true);
1373 dst = must_be_not_null(dst, true);
1374
1375 // Convert char[] offsets to byte[] offsets
1376 bool convert_src = (compress && src_elem == T_BYTE);
1377 bool convert_dst = (!compress && dst_elem == T_BYTE);
1378 if (convert_src) {
1379 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1380 } else if (convert_dst) {
1381 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1382 }
1383
1384 // Range checks
1385 generate_string_range_check(src, src_offset, length, convert_src);
1386 generate_string_range_check(dst, dst_offset, length, convert_dst);
1387 if (stopped()) {
1388 return true;
1389 }
1390
1391 Node* src_start = array_element_address(src, src_offset, src_elem);
1392 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1393 // 'src_start' points to src array + scaled offset
1394 // 'dst_start' points to dst array + scaled offset
1395 Node* count = NULL;
1396 if (compress) {
1397 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1398 } else {
1399 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1400 }
1401
1402 if (alloc != NULL) {
1403 if (alloc->maybe_set_complete(&_gvn)) {
1404 // "You break it, you buy it."
1405 InitializeNode* init = alloc->initialization();
1406 assert(init->is_complete(), "we just did this");
1407 init->set_complete_with_arraycopy();
1408 assert(dst->is_CheckCastPP(), "sanity");
1409 assert(dst->in(0)->in(0) == init, "dest pinned");
1410 }
1411 // Do not let stores that initialize this object be reordered with
1412 // a subsequent store that would make this object accessible by
1413 // other threads.
1414 // Record what AllocateNode this StoreStore protects so that
1415 // escape analysis can go from the MemBarStoreStoreNode to the
1416 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1417 // based on the escape status of the AllocateNode.
1418 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1419 }
1420 if (compress) {
1421 set_result(_gvn.transform(count));
1422 }
1423 clear_upper_avx();
1424
1425 return true;
1426 }
1427
1428 #ifdef _LP64
1429 #define XTOP ,top() /*additional argument*/
1430 #else //_LP64
1431 #define XTOP /*no additional argument*/
1432 #endif //_LP64
1433
1434 //------------------------inline_string_toBytesU--------------------------
1435 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1436 bool LibraryCallKit::inline_string_toBytesU() {
1437 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1438 return false;
1439 }
1440 // Get the arguments.
1441 Node* value = argument(0);
1442 Node* offset = argument(1);
1443 Node* length = argument(2);
1444
1445 Node* newcopy = NULL;
1446
1447 // Set the original stack and the reexecute bit for the interpreter to reexecute
1448 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1449 { PreserveReexecuteState preexecs(this);
1450 jvms()->set_should_reexecute(true);
1451
1452 // Check if a null path was taken unconditionally.
1453 value = null_check(value);
1454
1455 RegionNode* bailout = new RegionNode(1);
1456 record_for_igvn(bailout);
1457
1458 // Range checks
1459 generate_negative_guard(offset, bailout);
1460 generate_negative_guard(length, bailout);
1461 generate_limit_guard(offset, length, load_array_length(value), bailout);
1462 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1463 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1464
1465 if (bailout->req() > 1) {
1466 PreserveJVMState pjvms(this);
1467 set_control(_gvn.transform(bailout));
1468 uncommon_trap(Deoptimization::Reason_intrinsic,
1469 Deoptimization::Action_maybe_recompile);
1470 }
1471 if (stopped()) {
1472 return true;
1473 }
1474
1475 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1476 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1477 newcopy = new_array(klass_node, size, 0); // no arguments to push
1478 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1479 guarantee(alloc != NULL, "created above");
1480
1481 // Calculate starting addresses.
1482 Node* src_start = array_element_address(value, offset, T_CHAR);
1483 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1484
1485 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1486 const TypeInt* toffset = gvn().type(offset)->is_int();
1487 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1488
1489 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1490 const char* copyfunc_name = "arraycopy";
1491 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1492 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1493 OptoRuntime::fast_arraycopy_Type(),
1494 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1495 src_start, dst_start, ConvI2X(length) XTOP);
1496 // Do not let reads from the cloned object float above the arraycopy.
1497 if (alloc->maybe_set_complete(&_gvn)) {
1498 // "You break it, you buy it."
1499 InitializeNode* init = alloc->initialization();
1500 assert(init->is_complete(), "we just did this");
1501 init->set_complete_with_arraycopy();
1502 assert(newcopy->is_CheckCastPP(), "sanity");
1503 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1504 }
1505 // Do not let stores that initialize this object be reordered with
1506 // a subsequent store that would make this object accessible by
1507 // other threads.
1508 // Record what AllocateNode this StoreStore protects so that
1509 // escape analysis can go from the MemBarStoreStoreNode to the
1510 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1511 // based on the escape status of the AllocateNode.
1512 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1513 } // original reexecute is set back here
1514
1515 C->set_has_split_ifs(true); // Has chance for split-if optimization
1516 if (!stopped()) {
1517 set_result(newcopy);
1518 }
1519 clear_upper_avx();
1520
1521 return true;
1522 }
1523
1524 //------------------------inline_string_getCharsU--------------------------
1525 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1526 bool LibraryCallKit::inline_string_getCharsU() {
1527 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1528 return false;
1529 }
1530
1531 // Get the arguments.
1532 Node* src = argument(0);
1533 Node* src_begin = argument(1);
1534 Node* src_end = argument(2); // exclusive offset (i < src_end)
1535 Node* dst = argument(3);
1536 Node* dst_begin = argument(4);
1537
1538 // Check for allocation before we add nodes that would confuse
1539 // tightly_coupled_allocation()
1540 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1541
1542 // Check if a null path was taken unconditionally.
1543 src = null_check(src);
1544 dst = null_check(dst);
1545 if (stopped()) {
1546 return true;
1547 }
1548
1549 // Get length and convert char[] offset to byte[] offset
1550 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1551 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1552
1553 // Range checks
1554 generate_string_range_check(src, src_begin, length, true);
1555 generate_string_range_check(dst, dst_begin, length, false);
1556 if (stopped()) {
1557 return true;
1558 }
1559
1560 if (!stopped()) {
1561 // Calculate starting addresses.
1562 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1563 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1564
1565 // Check if array addresses are aligned to HeapWordSize
1566 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1567 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1568 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1569 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1570
1571 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1572 const char* copyfunc_name = "arraycopy";
1573 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1574 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1575 OptoRuntime::fast_arraycopy_Type(),
1576 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1577 src_start, dst_start, ConvI2X(length) XTOP);
1578 // Do not let reads from the cloned object float above the arraycopy.
1579 if (alloc != NULL) {
1580 if (alloc->maybe_set_complete(&_gvn)) {
1581 // "You break it, you buy it."
1582 InitializeNode* init = alloc->initialization();
1583 assert(init->is_complete(), "we just did this");
1584 init->set_complete_with_arraycopy();
1585 assert(dst->is_CheckCastPP(), "sanity");
1586 assert(dst->in(0)->in(0) == init, "dest pinned");
1587 }
1588 // Do not let stores that initialize this object be reordered with
1589 // a subsequent store that would make this object accessible by
1590 // other threads.
1591 // Record what AllocateNode this StoreStore protects so that
1592 // escape analysis can go from the MemBarStoreStoreNode to the
1593 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1594 // based on the escape status of the AllocateNode.
1595 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1596 } else {
1597 insert_mem_bar(Op_MemBarCPUOrder);
1598 }
1599 }
1600
1601 C->set_has_split_ifs(true); // Has chance for split-if optimization
1602 return true;
1603 }
1604
1605 //----------------------inline_string_char_access----------------------------
1606 // Store/Load char to/from byte[] array.
1607 // static void StringUTF16.putChar(byte[] val, int index, int c)
1608 // static char StringUTF16.getChar(byte[] val, int index)
1609 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1610 Node* value = argument(0);
1611 Node* index = argument(1);
1612 Node* ch = is_store ? argument(2) : NULL;
1613
1614 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1615 // correctly requires matched array shapes.
1616 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1617 "sanity: byte[] and char[] bases agree");
1618 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1619 "sanity: byte[] and char[] scales agree");
1620
1621 // Bail when getChar over constants is requested: constant folding would
1622 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1623 // Java method would constant fold nicely instead.
1624 if (!is_store && value->is_Con() && index->is_Con()) {
1625 return false;
1626 }
1627
1628 // Save state and restore on bailout
1629 uint old_sp = sp();
1630 SafePointNode* old_map = clone_map();
1631
1632 value = must_be_not_null(value, true);
1633
1634 Node* adr = array_element_address(value, index, T_CHAR);
1635 if (adr->is_top()) {
1636 set_map(old_map);
1637 set_sp(old_sp);
1638 return false;
1639 }
1640 old_map->destruct(&_gvn);
1641 if (is_store) {
1642 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1643 } else {
1644 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);
1645 set_result(ch);
1646 }
1647 return true;
1648 }
1649
1650 //--------------------------round_double_node--------------------------------
1651 // Round a double node if necessary.
1652 Node* LibraryCallKit::round_double_node(Node* n) {
1653 if (Matcher::strict_fp_requires_explicit_rounding) {
1654 #ifdef IA32
1655 if (UseSSE < 2) {
1656 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1657 }
1658 #else
1659 Unimplemented();
1660 #endif // IA32
1661 }
1662 return n;
1663 }
1664
1665 //------------------------------inline_math-----------------------------------
1666 // public static double Math.abs(double)
1667 // public static double Math.sqrt(double)
1668 // public static double Math.log(double)
1669 // public static double Math.log10(double)
1670 // public static double Math.round(double)
1671 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1672 Node* arg = round_double_node(argument(0));
1673 Node* n = NULL;
1674 switch (id) {
1675 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1676 case vmIntrinsics::_dsqrt:
1677 case vmIntrinsics::_dsqrt_strict:
1678 n = new SqrtDNode(C, control(), arg); break;
1679 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1680 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1681 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1682 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1683 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1684 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1685 default: fatal_unexpected_iid(id); break;
1686 }
1687 set_result(_gvn.transform(n));
1688 return true;
1689 }
1690
1691 //------------------------------inline_math-----------------------------------
1692 // public static float Math.abs(float)
1693 // public static int Math.abs(int)
1694 // public static long Math.abs(long)
1695 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1696 Node* arg = argument(0);
1697 Node* n = NULL;
1698 switch (id) {
1699 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1700 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1701 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1702 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1703 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1704 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1705 default: fatal_unexpected_iid(id); break;
1706 }
1707 set_result(_gvn.transform(n));
1708 return true;
1709 }
1710
1711 //------------------------------runtime_math-----------------------------
1712 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1713 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1714 "must be (DD)D or (D)D type");
1715
1716 // Inputs
1717 Node* a = round_double_node(argument(0));
1718 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1719
1720 const TypePtr* no_memory_effects = NULL;
1721 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1722 no_memory_effects,
1723 a, top(), b, b ? top() : NULL);
1724 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1725 #ifdef ASSERT
1726 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1727 assert(value_top == top(), "second value must be top");
1728 #endif
1729
1730 set_result(value);
1731 return true;
1732 }
1733
1734 //------------------------------inline_math_pow-----------------------------
1735 bool LibraryCallKit::inline_math_pow() {
1736 Node* exp = round_double_node(argument(2));
1737 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1738 if (d != NULL) {
1739 if (d->getd() == 2.0) {
1740 // Special case: pow(x, 2.0) => x * x
1741 Node* base = round_double_node(argument(0));
1742 set_result(_gvn.transform(new MulDNode(base, base)));
1743 return true;
1744 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1745 // Special case: pow(x, 0.5) => sqrt(x)
1746 Node* base = round_double_node(argument(0));
1747 Node* zero = _gvn.zerocon(T_DOUBLE);
1748
1749 RegionNode* region = new RegionNode(3);
1750 Node* phi = new PhiNode(region, Type::DOUBLE);
1751
1752 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1753 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1754 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1755 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1756 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1757
1758 Node* if_pow = generate_slow_guard(test, NULL);
1759 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1760 phi->init_req(1, value_sqrt);
1761 region->init_req(1, control());
1762
1763 if (if_pow != NULL) {
1764 set_control(if_pow);
1765 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1766 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1767 const TypePtr* no_memory_effects = NULL;
1768 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1769 no_memory_effects, base, top(), exp, top());
1770 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1771 #ifdef ASSERT
1772 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1773 assert(value_top == top(), "second value must be top");
1774 #endif
1775 phi->init_req(2, value_pow);
1776 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1777 }
1778
1779 C->set_has_split_ifs(true); // Has chance for split-if optimization
1780 set_control(_gvn.transform(region));
1781 record_for_igvn(region);
1782 set_result(_gvn.transform(phi));
1783
1784 return true;
1785 }
1786 }
1787
1788 return StubRoutines::dpow() != NULL ?
1789 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1790 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1791 }
1792
1793 //------------------------------inline_math_native-----------------------------
1794 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1795 switch (id) {
1796 case vmIntrinsics::_dsin:
1797 return StubRoutines::dsin() != NULL ?
1798 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1799 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1800 case vmIntrinsics::_dcos:
1801 return StubRoutines::dcos() != NULL ?
1802 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1803 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1804 case vmIntrinsics::_dtan:
1805 return StubRoutines::dtan() != NULL ?
1806 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1807 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1808 case vmIntrinsics::_dexp:
1809 return StubRoutines::dexp() != NULL ?
1810 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1811 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1812 case vmIntrinsics::_dlog:
1813 return StubRoutines::dlog() != NULL ?
1814 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1815 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1816 case vmIntrinsics::_dlog10:
1817 return StubRoutines::dlog10() != NULL ?
1818 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1819 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1820
1821 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1822 case vmIntrinsics::_ceil:
1823 case vmIntrinsics::_floor:
1824 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1825
1826 case vmIntrinsics::_dsqrt:
1827 case vmIntrinsics::_dsqrt_strict:
1828 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1829 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1830 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1831 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1832 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
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::scopedValueCache_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::scopedValueCache_offset()));
3368 // We cannot use immutable_memory() because we might flip onto a
3369 // different carrier thread, at which point we'll need to use that
3370 // carrier thread's cache.
3371 // return _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
3372 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3373 return make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3374 }
3375
3376 //------------------------inline_native_scopedValueCache------------------
3377 bool LibraryCallKit::inline_native_scopedValueCache() {
3378 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3379 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3380 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3381
3382 // Because we create the scopedValue cache lazily we have to make the
3383 // type of the result BotPTR.
3384 bool xk = etype->klass_is_exact();
3385 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3386 Node* cache_obj_handle = scopedValueCache_helper();
3387 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3388
3389 return true;
3390 }
3391
3392 //------------------------inline_native_setScopedValueCache------------------
3393 bool LibraryCallKit::inline_native_setScopedValueCache() {
3394 Node* arr = argument(0);
3395 Node* cache_obj_handle = scopedValueCache_helper();
3396
3397 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3398 store_to_memory(control(), cache_obj_handle, arr, T_OBJECT, adr_type,
3399 MemNode::unordered);
3400
3401 return true;
3402 }
3403
3404 //---------------------------load_mirror_from_klass----------------------------
3405 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3406 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3407 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3408 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3409 // mirror = ((OopHandle)mirror)->resolve();
3410 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3411 }
3412
3413 //-----------------------load_klass_from_mirror_common-------------------------
3414 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3415 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3416 // and branch to the given path on the region.
3417 // If never_see_null, take an uncommon trap on null, so we can optimistically
3418 // compile for the non-null case.
3419 // If the region is NULL, force never_see_null = true.
3420 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3421 bool never_see_null,
3422 RegionNode* region,
3423 int null_path,
3424 int offset) {
3425 if (region == NULL) never_see_null = true;
3426 Node* p = basic_plus_adr(mirror, offset);
3427 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3428 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3429 Node* null_ctl = top();
3430 kls = null_check_oop(kls, &null_ctl, never_see_null);
3431 if (region != NULL) {
3432 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3433 region->init_req(null_path, null_ctl);
3434 } else {
3435 assert(null_ctl == top(), "no loose ends");
3436 }
3437 return kls;
3438 }
3439
3440 //--------------------(inline_native_Class_query helpers)---------------------
3441 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3442 // Fall through if (mods & mask) == bits, take the guard otherwise.
3443 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3444 // Branch around if the given klass has the given modifier bit set.
3445 // Like generate_guard, adds a new path onto the region.
3446 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3447 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3448 Node* mask = intcon(modifier_mask);
3449 Node* bits = intcon(modifier_bits);
3450 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3451 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3452 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3453 return generate_fair_guard(bol, region);
3454 }
3455 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3456 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3457 }
3458 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3459 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3460 }
3461
3462 //-------------------------inline_native_Class_query-------------------
3463 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3464 const Type* return_type = TypeInt::BOOL;
3465 Node* prim_return_value = top(); // what happens if it's a primitive class?
3466 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3467 bool expect_prim = false; // most of these guys expect to work on refs
3468
3469 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3470
3471 Node* mirror = argument(0);
3472 Node* obj = top();
3473
3474 switch (id) {
3475 case vmIntrinsics::_isInstance:
3476 // nothing is an instance of a primitive type
3477 prim_return_value = intcon(0);
3478 obj = argument(1);
3479 break;
3480 case vmIntrinsics::_getModifiers:
3481 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3482 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3483 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3484 break;
3485 case vmIntrinsics::_isInterface:
3486 prim_return_value = intcon(0);
3487 break;
3488 case vmIntrinsics::_isArray:
3489 prim_return_value = intcon(0);
3490 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3491 break;
3492 case vmIntrinsics::_isPrimitive:
3493 prim_return_value = intcon(1);
3494 expect_prim = true; // obviously
3495 break;
3496 case vmIntrinsics::_isHidden:
3497 prim_return_value = intcon(0);
3498 break;
3499 case vmIntrinsics::_getSuperclass:
3500 prim_return_value = null();
3501 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3502 break;
3503 case vmIntrinsics::_getClassAccessFlags:
3504 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3505 return_type = TypeInt::INT; // not bool! 6297094
3506 break;
3507 default:
3508 fatal_unexpected_iid(id);
3509 break;
3510 }
3511
3512 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3513 if (mirror_con == NULL) return false; // cannot happen?
3514
3515 #ifndef PRODUCT
3516 if (C->print_intrinsics() || C->print_inlining()) {
3517 ciType* k = mirror_con->java_mirror_type();
3518 if (k) {
3519 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3520 k->print_name();
3521 tty->cr();
3522 }
3523 }
3524 #endif
3525
3526 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3527 RegionNode* region = new RegionNode(PATH_LIMIT);
3528 record_for_igvn(region);
3529 PhiNode* phi = new PhiNode(region, return_type);
3530
3531 // The mirror will never be null of Reflection.getClassAccessFlags, however
3532 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3533 // if it is. See bug 4774291.
3534
3535 // For Reflection.getClassAccessFlags(), the null check occurs in
3536 // the wrong place; see inline_unsafe_access(), above, for a similar
3537 // situation.
3538 mirror = null_check(mirror);
3539 // If mirror or obj is dead, only null-path is taken.
3540 if (stopped()) return true;
3541
3542 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3543
3544 // Now load the mirror's klass metaobject, and null-check it.
3545 // Side-effects region with the control path if the klass is null.
3546 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3547 // If kls is null, we have a primitive mirror.
3548 phi->init_req(_prim_path, prim_return_value);
3549 if (stopped()) { set_result(region, phi); return true; }
3550 bool safe_for_replace = (region->in(_prim_path) == top());
3551
3552 Node* p; // handy temp
3553 Node* null_ctl;
3554
3555 // Now that we have the non-null klass, we can perform the real query.
3556 // For constant classes, the query will constant-fold in LoadNode::Value.
3557 Node* query_value = top();
3558 switch (id) {
3559 case vmIntrinsics::_isInstance:
3560 // nothing is an instance of a primitive type
3561 query_value = gen_instanceof(obj, kls, safe_for_replace);
3562 break;
3563
3564 case vmIntrinsics::_getModifiers:
3565 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3566 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3567 break;
3568
3569 case vmIntrinsics::_isInterface:
3570 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3571 if (generate_interface_guard(kls, region) != NULL)
3572 // A guard was added. If the guard is taken, it was an interface.
3573 phi->add_req(intcon(1));
3574 // If we fall through, it's a plain class.
3575 query_value = intcon(0);
3576 break;
3577
3578 case vmIntrinsics::_isArray:
3579 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3580 if (generate_array_guard(kls, region) != NULL)
3581 // A guard was added. If the guard is taken, it was an array.
3582 phi->add_req(intcon(1));
3583 // If we fall through, it's a plain class.
3584 query_value = intcon(0);
3585 break;
3586
3587 case vmIntrinsics::_isPrimitive:
3588 query_value = intcon(0); // "normal" path produces false
3589 break;
3590
3591 case vmIntrinsics::_isHidden:
3592 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3593 if (generate_hidden_class_guard(kls, region) != NULL)
3594 // A guard was added. If the guard is taken, it was an hidden class.
3595 phi->add_req(intcon(1));
3596 // If we fall through, it's a plain class.
3597 query_value = intcon(0);
3598 break;
3599
3600
3601 case vmIntrinsics::_getSuperclass:
3602 // The rules here are somewhat unfortunate, but we can still do better
3603 // with random logic than with a JNI call.
3604 // Interfaces store null or Object as _super, but must report null.
3605 // Arrays store an intermediate super as _super, but must report Object.
3606 // Other types can report the actual _super.
3607 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3608 if (generate_interface_guard(kls, region) != NULL)
3609 // A guard was added. If the guard is taken, it was an interface.
3610 phi->add_req(null());
3611 if (generate_array_guard(kls, region) != NULL)
3612 // A guard was added. If the guard is taken, it was an array.
3613 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3614 // If we fall through, it's a plain class. Get its _super.
3615 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3616 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3617 null_ctl = top();
3618 kls = null_check_oop(kls, &null_ctl);
3619 if (null_ctl != top()) {
3620 // If the guard is taken, Object.superClass is null (both klass and mirror).
3621 region->add_req(null_ctl);
3622 phi ->add_req(null());
3623 }
3624 if (!stopped()) {
3625 query_value = load_mirror_from_klass(kls);
3626 }
3627 break;
3628
3629 case vmIntrinsics::_getClassAccessFlags:
3630 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3631 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3632 break;
3633
3634 default:
3635 fatal_unexpected_iid(id);
3636 break;
3637 }
3638
3639 // Fall-through is the normal case of a query to a real class.
3640 phi->init_req(1, query_value);
3641 region->init_req(1, control());
3642
3643 C->set_has_split_ifs(true); // Has chance for split-if optimization
3644 set_result(region, phi);
3645 return true;
3646 }
3647
3648 //-------------------------inline_Class_cast-------------------
3649 bool LibraryCallKit::inline_Class_cast() {
3650 Node* mirror = argument(0); // Class
3651 Node* obj = argument(1);
3652 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3653 if (mirror_con == NULL) {
3654 return false; // dead path (mirror->is_top()).
3655 }
3656 if (obj == NULL || obj->is_top()) {
3657 return false; // dead path
3658 }
3659 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3660
3661 // First, see if Class.cast() can be folded statically.
3662 // java_mirror_type() returns non-null for compile-time Class constants.
3663 ciType* tm = mirror_con->java_mirror_type();
3664 if (tm != NULL && tm->is_klass() &&
3665 tp != NULL) {
3666 if (!tp->is_loaded()) {
3667 // Don't use intrinsic when class is not loaded.
3668 return false;
3669 } else {
3670 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
3671 if (static_res == Compile::SSC_always_true) {
3672 // isInstance() is true - fold the code.
3673 set_result(obj);
3674 return true;
3675 } else if (static_res == Compile::SSC_always_false) {
3676 // Don't use intrinsic, have to throw ClassCastException.
3677 // If the reference is null, the non-intrinsic bytecode will
3678 // be optimized appropriately.
3679 return false;
3680 }
3681 }
3682 }
3683
3684 // Bailout intrinsic and do normal inlining if exception path is frequent.
3685 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3686 return false;
3687 }
3688
3689 // Generate dynamic checks.
3690 // Class.cast() is java implementation of _checkcast bytecode.
3691 // Do checkcast (Parse::do_checkcast()) optimizations here.
3692
3693 mirror = null_check(mirror);
3694 // If mirror is dead, only null-path is taken.
3695 if (stopped()) {
3696 return true;
3697 }
3698
3699 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3700 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3701 RegionNode* region = new RegionNode(PATH_LIMIT);
3702 record_for_igvn(region);
3703
3704 // Now load the mirror's klass metaobject, and null-check it.
3705 // If kls is null, we have a primitive mirror and
3706 // nothing is an instance of a primitive type.
3707 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3708
3709 Node* res = top();
3710 if (!stopped()) {
3711 Node* bad_type_ctrl = top();
3712 // Do checkcast optimizations.
3713 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3714 region->init_req(_bad_type_path, bad_type_ctrl);
3715 }
3716 if (region->in(_prim_path) != top() ||
3717 region->in(_bad_type_path) != top()) {
3718 // Let Interpreter throw ClassCastException.
3719 PreserveJVMState pjvms(this);
3720 set_control(_gvn.transform(region));
3721 uncommon_trap(Deoptimization::Reason_intrinsic,
3722 Deoptimization::Action_maybe_recompile);
3723 }
3724 if (!stopped()) {
3725 set_result(res);
3726 }
3727 return true;
3728 }
3729
3730
3731 //--------------------------inline_native_subtype_check------------------------
3732 // This intrinsic takes the JNI calls out of the heart of
3733 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3734 bool LibraryCallKit::inline_native_subtype_check() {
3735 // Pull both arguments off the stack.
3736 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3737 args[0] = argument(0);
3738 args[1] = argument(1);
3739 Node* klasses[2]; // corresponding Klasses: superk, subk
3740 klasses[0] = klasses[1] = top();
3741
3742 enum {
3743 // A full decision tree on {superc is prim, subc is prim}:
3744 _prim_0_path = 1, // {P,N} => false
3745 // {P,P} & superc!=subc => false
3746 _prim_same_path, // {P,P} & superc==subc => true
3747 _prim_1_path, // {N,P} => false
3748 _ref_subtype_path, // {N,N} & subtype check wins => true
3749 _both_ref_path, // {N,N} & subtype check loses => false
3750 PATH_LIMIT
3751 };
3752
3753 RegionNode* region = new RegionNode(PATH_LIMIT);
3754 Node* phi = new PhiNode(region, TypeInt::BOOL);
3755 record_for_igvn(region);
3756
3757 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3758 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3759 int class_klass_offset = java_lang_Class::klass_offset();
3760
3761 // First null-check both mirrors and load each mirror's klass metaobject.
3762 int which_arg;
3763 for (which_arg = 0; which_arg <= 1; which_arg++) {
3764 Node* arg = args[which_arg];
3765 arg = null_check(arg);
3766 if (stopped()) break;
3767 args[which_arg] = arg;
3768
3769 Node* p = basic_plus_adr(arg, class_klass_offset);
3770 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3771 klasses[which_arg] = _gvn.transform(kls);
3772 }
3773
3774 // Having loaded both klasses, test each for null.
3775 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3776 for (which_arg = 0; which_arg <= 1; which_arg++) {
3777 Node* kls = klasses[which_arg];
3778 Node* null_ctl = top();
3779 kls = null_check_oop(kls, &null_ctl, never_see_null);
3780 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3781 region->init_req(prim_path, null_ctl);
3782 if (stopped()) break;
3783 klasses[which_arg] = kls;
3784 }
3785
3786 if (!stopped()) {
3787 // now we have two reference types, in klasses[0..1]
3788 Node* subk = klasses[1]; // the argument to isAssignableFrom
3789 Node* superk = klasses[0]; // the receiver
3790 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3791 // now we have a successful reference subtype check
3792 region->set_req(_ref_subtype_path, control());
3793 }
3794
3795 // If both operands are primitive (both klasses null), then
3796 // we must return true when they are identical primitives.
3797 // It is convenient to test this after the first null klass check.
3798 set_control(region->in(_prim_0_path)); // go back to first null check
3799 if (!stopped()) {
3800 // Since superc is primitive, make a guard for the superc==subc case.
3801 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3802 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3803 generate_guard(bol_eq, region, PROB_FAIR);
3804 if (region->req() == PATH_LIMIT+1) {
3805 // A guard was added. If the added guard is taken, superc==subc.
3806 region->swap_edges(PATH_LIMIT, _prim_same_path);
3807 region->del_req(PATH_LIMIT);
3808 }
3809 region->set_req(_prim_0_path, control()); // Not equal after all.
3810 }
3811
3812 // these are the only paths that produce 'true':
3813 phi->set_req(_prim_same_path, intcon(1));
3814 phi->set_req(_ref_subtype_path, intcon(1));
3815
3816 // pull together the cases:
3817 assert(region->req() == PATH_LIMIT, "sane region");
3818 for (uint i = 1; i < region->req(); i++) {
3819 Node* ctl = region->in(i);
3820 if (ctl == NULL || ctl == top()) {
3821 region->set_req(i, top());
3822 phi ->set_req(i, top());
3823 } else if (phi->in(i) == NULL) {
3824 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3825 }
3826 }
3827
3828 set_control(_gvn.transform(region));
3829 set_result(_gvn.transform(phi));
3830 return true;
3831 }
3832
3833 //---------------------generate_array_guard_common------------------------
3834 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3835 bool obj_array, bool not_array) {
3836
3837 if (stopped()) {
3838 return NULL;
3839 }
3840
3841 // If obj_array/non_array==false/false:
3842 // Branch around if the given klass is in fact an array (either obj or prim).
3843 // If obj_array/non_array==false/true:
3844 // Branch around if the given klass is not an array klass of any kind.
3845 // If obj_array/non_array==true/true:
3846 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3847 // If obj_array/non_array==true/false:
3848 // Branch around if the kls is an oop array (Object[] or subtype)
3849 //
3850 // Like generate_guard, adds a new path onto the region.
3851 jint layout_con = 0;
3852 Node* layout_val = get_layout_helper(kls, layout_con);
3853 if (layout_val == NULL) {
3854 bool query = (obj_array
3855 ? Klass::layout_helper_is_objArray(layout_con)
3856 : Klass::layout_helper_is_array(layout_con));
3857 if (query == not_array) {
3858 return NULL; // never a branch
3859 } else { // always a branch
3860 Node* always_branch = control();
3861 if (region != NULL)
3862 region->add_req(always_branch);
3863 set_control(top());
3864 return always_branch;
3865 }
3866 }
3867 // Now test the correct condition.
3868 jint nval = (obj_array
3869 ? (jint)(Klass::_lh_array_tag_type_value
3870 << Klass::_lh_array_tag_shift)
3871 : Klass::_lh_neutral_value);
3872 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3873 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3874 // invert the test if we are looking for a non-array
3875 if (not_array) btest = BoolTest(btest).negate();
3876 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3877 return generate_fair_guard(bol, region);
3878 }
3879
3880
3881 //-----------------------inline_native_newArray--------------------------
3882 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3883 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3884 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3885 Node* mirror;
3886 Node* count_val;
3887 if (uninitialized) {
3888 mirror = argument(1);
3889 count_val = argument(2);
3890 } else {
3891 mirror = argument(0);
3892 count_val = argument(1);
3893 }
3894
3895 mirror = null_check(mirror);
3896 // If mirror or obj is dead, only null-path is taken.
3897 if (stopped()) return true;
3898
3899 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3900 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3901 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3902 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3903 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3904
3905 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3906 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3907 result_reg, _slow_path);
3908 Node* normal_ctl = control();
3909 Node* no_array_ctl = result_reg->in(_slow_path);
3910
3911 // Generate code for the slow case. We make a call to newArray().
3912 set_control(no_array_ctl);
3913 if (!stopped()) {
3914 // Either the input type is void.class, or else the
3915 // array klass has not yet been cached. Either the
3916 // ensuing call will throw an exception, or else it
3917 // will cache the array klass for next time.
3918 PreserveJVMState pjvms(this);
3919 CallJavaNode* slow_call = NULL;
3920 if (uninitialized) {
3921 // Generate optimized virtual call (holder class 'Unsafe' is final)
3922 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3923 } else {
3924 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3925 }
3926 Node* slow_result = set_results_for_java_call(slow_call);
3927 // this->control() comes from set_results_for_java_call
3928 result_reg->set_req(_slow_path, control());
3929 result_val->set_req(_slow_path, slow_result);
3930 result_io ->set_req(_slow_path, i_o());
3931 result_mem->set_req(_slow_path, reset_memory());
3932 }
3933
3934 set_control(normal_ctl);
3935 if (!stopped()) {
3936 // Normal case: The array type has been cached in the java.lang.Class.
3937 // The following call works fine even if the array type is polymorphic.
3938 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3939 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3940 result_reg->init_req(_normal_path, control());
3941 result_val->init_req(_normal_path, obj);
3942 result_io ->init_req(_normal_path, i_o());
3943 result_mem->init_req(_normal_path, reset_memory());
3944
3945 if (uninitialized) {
3946 // Mark the allocation so that zeroing is skipped
3947 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3948 alloc->maybe_set_complete(&_gvn);
3949 }
3950 }
3951
3952 // Return the combined state.
3953 set_i_o( _gvn.transform(result_io) );
3954 set_all_memory( _gvn.transform(result_mem));
3955
3956 C->set_has_split_ifs(true); // Has chance for split-if optimization
3957 set_result(result_reg, result_val);
3958 return true;
3959 }
3960
3961 //----------------------inline_native_getLength--------------------------
3962 // public static native int java.lang.reflect.Array.getLength(Object array);
3963 bool LibraryCallKit::inline_native_getLength() {
3964 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3965
3966 Node* array = null_check(argument(0));
3967 // If array is dead, only null-path is taken.
3968 if (stopped()) return true;
3969
3970 // Deoptimize if it is a non-array.
3971 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3972
3973 if (non_array != NULL) {
3974 PreserveJVMState pjvms(this);
3975 set_control(non_array);
3976 uncommon_trap(Deoptimization::Reason_intrinsic,
3977 Deoptimization::Action_maybe_recompile);
3978 }
3979
3980 // If control is dead, only non-array-path is taken.
3981 if (stopped()) return true;
3982
3983 // The works fine even if the array type is polymorphic.
3984 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3985 Node* result = load_array_length(array);
3986
3987 C->set_has_split_ifs(true); // Has chance for split-if optimization
3988 set_result(result);
3989 return true;
3990 }
3991
3992 //------------------------inline_array_copyOf----------------------------
3993 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3994 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3995 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3996 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3997
3998 // Get the arguments.
3999 Node* original = argument(0);
4000 Node* start = is_copyOfRange? argument(1): intcon(0);
4001 Node* end = is_copyOfRange? argument(2): argument(1);
4002 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4003
4004 Node* newcopy = NULL;
4005
4006 // Set the original stack and the reexecute bit for the interpreter to reexecute
4007 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4008 { PreserveReexecuteState preexecs(this);
4009 jvms()->set_should_reexecute(true);
4010
4011 array_type_mirror = null_check(array_type_mirror);
4012 original = null_check(original);
4013
4014 // Check if a null path was taken unconditionally.
4015 if (stopped()) return true;
4016
4017 Node* orig_length = load_array_length(original);
4018
4019 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4020 klass_node = null_check(klass_node);
4021
4022 RegionNode* bailout = new RegionNode(1);
4023 record_for_igvn(bailout);
4024
4025 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4026 // Bail out if that is so.
4027 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4028 if (not_objArray != NULL) {
4029 // Improve the klass node's type from the new optimistic assumption:
4030 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4031 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4032 Node* cast = new CastPPNode(klass_node, akls);
4033 cast->init_req(0, control());
4034 klass_node = _gvn.transform(cast);
4035 }
4036
4037 // Bail out if either start or end is negative.
4038 generate_negative_guard(start, bailout, &start);
4039 generate_negative_guard(end, bailout, &end);
4040
4041 Node* length = end;
4042 if (_gvn.type(start) != TypeInt::ZERO) {
4043 length = _gvn.transform(new SubINode(end, start));
4044 }
4045
4046 // Bail out if length is negative.
4047 // Without this the new_array would throw
4048 // NegativeArraySizeException but IllegalArgumentException is what
4049 // should be thrown
4050 generate_negative_guard(length, bailout, &length);
4051
4052 if (bailout->req() > 1) {
4053 PreserveJVMState pjvms(this);
4054 set_control(_gvn.transform(bailout));
4055 uncommon_trap(Deoptimization::Reason_intrinsic,
4056 Deoptimization::Action_maybe_recompile);
4057 }
4058
4059 if (!stopped()) {
4060 // How many elements will we copy from the original?
4061 // The answer is MinI(orig_length - start, length).
4062 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4063 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4064
4065 // Generate a direct call to the right arraycopy function(s).
4066 // We know the copy is disjoint but we might not know if the
4067 // oop stores need checking.
4068 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4069 // This will fail a store-check if x contains any non-nulls.
4070
4071 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4072 // loads/stores but it is legal only if we're sure the
4073 // Arrays.copyOf would succeed. So we need all input arguments
4074 // to the copyOf to be validated, including that the copy to the
4075 // new array won't trigger an ArrayStoreException. That subtype
4076 // check can be optimized if we know something on the type of
4077 // the input array from type speculation.
4078 if (_gvn.type(klass_node)->singleton()) {
4079 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4080 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4081
4082 int test = C->static_subtype_check(superk, subk);
4083 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4084 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4085 if (t_original->speculative_type() != NULL) {
4086 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4087 }
4088 }
4089 }
4090
4091 bool validated = false;
4092 // Reason_class_check rather than Reason_intrinsic because we
4093 // want to intrinsify even if this traps.
4094 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4095 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4096
4097 if (not_subtype_ctrl != top()) {
4098 PreserveJVMState pjvms(this);
4099 set_control(not_subtype_ctrl);
4100 uncommon_trap(Deoptimization::Reason_class_check,
4101 Deoptimization::Action_make_not_entrant);
4102 assert(stopped(), "Should be stopped");
4103 }
4104 validated = true;
4105 }
4106
4107 if (!stopped()) {
4108 newcopy = new_array(klass_node, length, 0); // no arguments to push
4109
4110 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4111 load_object_klass(original), klass_node);
4112 if (!is_copyOfRange) {
4113 ac->set_copyof(validated);
4114 } else {
4115 ac->set_copyofrange(validated);
4116 }
4117 Node* n = _gvn.transform(ac);
4118 if (n == ac) {
4119 ac->connect_outputs(this);
4120 } else {
4121 assert(validated, "shouldn't transform if all arguments not validated");
4122 set_all_memory(n);
4123 }
4124 }
4125 }
4126 } // original reexecute is set back here
4127
4128 C->set_has_split_ifs(true); // Has chance for split-if optimization
4129 if (!stopped()) {
4130 set_result(newcopy);
4131 }
4132 return true;
4133 }
4134
4135
4136 //----------------------generate_virtual_guard---------------------------
4137 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4138 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4139 RegionNode* slow_region) {
4140 ciMethod* method = callee();
4141 int vtable_index = method->vtable_index();
4142 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4143 "bad index %d", vtable_index);
4144 // Get the Method* out of the appropriate vtable entry.
4145 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4146 vtable_index*vtableEntry::size_in_bytes() +
4147 vtableEntry::method_offset_in_bytes();
4148 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4149 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4150
4151 // Compare the target method with the expected method (e.g., Object.hashCode).
4152 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4153
4154 Node* native_call = makecon(native_call_addr);
4155 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4156 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4157
4158 return generate_slow_guard(test_native, slow_region);
4159 }
4160
4161 //-----------------------generate_method_call----------------------------
4162 // Use generate_method_call to make a slow-call to the real
4163 // method if the fast path fails. An alternative would be to
4164 // use a stub like OptoRuntime::slow_arraycopy_Java.
4165 // This only works for expanding the current library call,
4166 // not another intrinsic. (E.g., don't use this for making an
4167 // arraycopy call inside of the copyOf intrinsic.)
4168 CallJavaNode*
4169 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4170 // When compiling the intrinsic method itself, do not use this technique.
4171 guarantee(callee() != C->method(), "cannot make slow-call to self");
4172
4173 ciMethod* method = callee();
4174 // ensure the JVMS we have will be correct for this call
4175 guarantee(method_id == method->intrinsic_id(), "must match");
4176
4177 const TypeFunc* tf = TypeFunc::make(method);
4178 CallJavaNode* slow_call;
4179 if (is_static) {
4180 assert(!is_virtual, "");
4181 slow_call = new CallStaticJavaNode(C, tf,
4182 SharedRuntime::get_resolve_static_call_stub(), method);
4183 } else if (is_virtual) {
4184 null_check_receiver();
4185 int vtable_index = Method::invalid_vtable_index;
4186 if (UseInlineCaches) {
4187 // Suppress the vtable call
4188 } else {
4189 // hashCode and clone are not a miranda methods,
4190 // so the vtable index is fixed.
4191 // No need to use the linkResolver to get it.
4192 vtable_index = method->vtable_index();
4193 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4194 "bad index %d", vtable_index);
4195 }
4196 slow_call = new CallDynamicJavaNode(tf,
4197 SharedRuntime::get_resolve_virtual_call_stub(),
4198 method, vtable_index);
4199 } else { // neither virtual nor static: opt_virtual
4200 null_check_receiver();
4201 slow_call = new CallStaticJavaNode(C, tf,
4202 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4203 slow_call->set_optimized_virtual(true);
4204 }
4205 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4206 // To be able to issue a direct call (optimized virtual or virtual)
4207 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4208 // about the method being invoked should be attached to the call site to
4209 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4210 slow_call->set_override_symbolic_info(true);
4211 }
4212 set_arguments_for_java_call(slow_call);
4213 set_edges_for_java_call(slow_call);
4214 return slow_call;
4215 }
4216
4217
4218 /**
4219 * Build special case code for calls to hashCode on an object. This call may
4220 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4221 * slightly different code.
4222 */
4223 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4224 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4225 assert(!(is_virtual && is_static), "either virtual, special, or static");
4226
4227 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4228
4229 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4230 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4231 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4232 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4233 Node* obj = NULL;
4234 if (!is_static) {
4235 // Check for hashing null object
4236 obj = null_check_receiver();
4237 if (stopped()) return true; // unconditionally null
4238 result_reg->init_req(_null_path, top());
4239 result_val->init_req(_null_path, top());
4240 } else {
4241 // Do a null check, and return zero if null.
4242 // System.identityHashCode(null) == 0
4243 obj = argument(0);
4244 Node* null_ctl = top();
4245 obj = null_check_oop(obj, &null_ctl);
4246 result_reg->init_req(_null_path, null_ctl);
4247 result_val->init_req(_null_path, _gvn.intcon(0));
4248 }
4249
4250 // Unconditionally null? Then return right away.
4251 if (stopped()) {
4252 set_control( result_reg->in(_null_path));
4253 if (!stopped())
4254 set_result(result_val->in(_null_path));
4255 return true;
4256 }
4257
4258 // We only go to the fast case code if we pass a number of guards. The
4259 // paths which do not pass are accumulated in the slow_region.
4260 RegionNode* slow_region = new RegionNode(1);
4261 record_for_igvn(slow_region);
4262
4263 // If this is a virtual call, we generate a funny guard. We pull out
4264 // the vtable entry corresponding to hashCode() from the target object.
4265 // If the target method which we are calling happens to be the native
4266 // Object hashCode() method, we pass the guard. We do not need this
4267 // guard for non-virtual calls -- the caller is known to be the native
4268 // Object hashCode().
4269 if (is_virtual) {
4270 // After null check, get the object's klass.
4271 Node* obj_klass = load_object_klass(obj);
4272 generate_virtual_guard(obj_klass, slow_region);
4273 }
4274
4275 // Get the header out of the object, use LoadMarkNode when available
4276 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4277 // The control of the load must be NULL. Otherwise, the load can move before
4278 // the null check after castPP removal.
4279 Node* no_ctrl = NULL;
4280 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4281
4282 // Test the header to see if it is unlocked.
4283 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4284 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4285 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4286 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4287 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4288
4289 generate_slow_guard(test_unlocked, slow_region);
4290
4291 // Get the hash value and check to see that it has been properly assigned.
4292 // We depend on hash_mask being at most 32 bits and avoid the use of
4293 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4294 // vm: see markWord.hpp.
4295 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4296 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4297 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4298 // This hack lets the hash bits live anywhere in the mark object now, as long
4299 // as the shift drops the relevant bits into the low 32 bits. Note that
4300 // Java spec says that HashCode is an int so there's no point in capturing
4301 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4302 hshifted_header = ConvX2I(hshifted_header);
4303 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4304
4305 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4306 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4307 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4308
4309 generate_slow_guard(test_assigned, slow_region);
4310
4311 Node* init_mem = reset_memory();
4312 // fill in the rest of the null path:
4313 result_io ->init_req(_null_path, i_o());
4314 result_mem->init_req(_null_path, init_mem);
4315
4316 result_val->init_req(_fast_path, hash_val);
4317 result_reg->init_req(_fast_path, control());
4318 result_io ->init_req(_fast_path, i_o());
4319 result_mem->init_req(_fast_path, init_mem);
4320
4321 // Generate code for the slow case. We make a call to hashCode().
4322 set_control(_gvn.transform(slow_region));
4323 if (!stopped()) {
4324 // No need for PreserveJVMState, because we're using up the present state.
4325 set_all_memory(init_mem);
4326 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4327 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4328 Node* slow_result = set_results_for_java_call(slow_call);
4329 // this->control() comes from set_results_for_java_call
4330 result_reg->init_req(_slow_path, control());
4331 result_val->init_req(_slow_path, slow_result);
4332 result_io ->set_req(_slow_path, i_o());
4333 result_mem ->set_req(_slow_path, reset_memory());
4334 }
4335
4336 // Return the combined state.
4337 set_i_o( _gvn.transform(result_io) );
4338 set_all_memory( _gvn.transform(result_mem));
4339
4340 set_result(result_reg, result_val);
4341 return true;
4342 }
4343
4344 //---------------------------inline_native_getClass----------------------------
4345 // public final native Class<?> java.lang.Object.getClass();
4346 //
4347 // Build special case code for calls to getClass on an object.
4348 bool LibraryCallKit::inline_native_getClass() {
4349 Node* obj = null_check_receiver();
4350 if (stopped()) return true;
4351 set_result(load_mirror_from_klass(load_object_klass(obj)));
4352 return true;
4353 }
4354
4355 //-----------------inline_native_Reflection_getCallerClass---------------------
4356 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4357 //
4358 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4359 //
4360 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4361 // in that it must skip particular security frames and checks for
4362 // caller sensitive methods.
4363 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4364 #ifndef PRODUCT
4365 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4366 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4367 }
4368 #endif
4369
4370 if (!jvms()->has_method()) {
4371 #ifndef PRODUCT
4372 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4373 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4374 }
4375 #endif
4376 return false;
4377 }
4378
4379 // Walk back up the JVM state to find the caller at the required
4380 // depth.
4381 JVMState* caller_jvms = jvms();
4382
4383 // Cf. JVM_GetCallerClass
4384 // NOTE: Start the loop at depth 1 because the current JVM state does
4385 // not include the Reflection.getCallerClass() frame.
4386 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4387 ciMethod* m = caller_jvms->method();
4388 switch (n) {
4389 case 0:
4390 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4391 break;
4392 case 1:
4393 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4394 if (!m->caller_sensitive()) {
4395 #ifndef PRODUCT
4396 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4397 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4398 }
4399 #endif
4400 return false; // bail-out; let JVM_GetCallerClass do the work
4401 }
4402 break;
4403 default:
4404 if (!m->is_ignored_by_security_stack_walk()) {
4405 // We have reached the desired frame; return the holder class.
4406 // Acquire method holder as java.lang.Class and push as constant.
4407 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4408 ciInstance* caller_mirror = caller_klass->java_mirror();
4409 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4410
4411 #ifndef PRODUCT
4412 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4413 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());
4414 tty->print_cr(" JVM state at this point:");
4415 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4416 ciMethod* m = jvms()->of_depth(i)->method();
4417 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4418 }
4419 }
4420 #endif
4421 return true;
4422 }
4423 break;
4424 }
4425 }
4426
4427 #ifndef PRODUCT
4428 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4429 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4430 tty->print_cr(" JVM state at this point:");
4431 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4432 ciMethod* m = jvms()->of_depth(i)->method();
4433 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4434 }
4435 }
4436 #endif
4437
4438 return false; // bail-out; let JVM_GetCallerClass do the work
4439 }
4440
4441 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4442 Node* arg = argument(0);
4443 Node* result = NULL;
4444
4445 switch (id) {
4446 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4447 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4448 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4449 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4450 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4451 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4452
4453 case vmIntrinsics::_doubleToLongBits: {
4454 // two paths (plus control) merge in a wood
4455 RegionNode *r = new RegionNode(3);
4456 Node *phi = new PhiNode(r, TypeLong::LONG);
4457
4458 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4459 // Build the boolean node
4460 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4461
4462 // Branch either way.
4463 // NaN case is less traveled, which makes all the difference.
4464 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4465 Node *opt_isnan = _gvn.transform(ifisnan);
4466 assert( opt_isnan->is_If(), "Expect an IfNode");
4467 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4468 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4469
4470 set_control(iftrue);
4471
4472 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4473 Node *slow_result = longcon(nan_bits); // return NaN
4474 phi->init_req(1, _gvn.transform( slow_result ));
4475 r->init_req(1, iftrue);
4476
4477 // Else fall through
4478 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4479 set_control(iffalse);
4480
4481 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4482 r->init_req(2, iffalse);
4483
4484 // Post merge
4485 set_control(_gvn.transform(r));
4486 record_for_igvn(r);
4487
4488 C->set_has_split_ifs(true); // Has chance for split-if optimization
4489 result = phi;
4490 assert(result->bottom_type()->isa_long(), "must be");
4491 break;
4492 }
4493
4494 case vmIntrinsics::_floatToIntBits: {
4495 // two paths (plus control) merge in a wood
4496 RegionNode *r = new RegionNode(3);
4497 Node *phi = new PhiNode(r, TypeInt::INT);
4498
4499 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4500 // Build the boolean node
4501 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4502
4503 // Branch either way.
4504 // NaN case is less traveled, which makes all the difference.
4505 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4506 Node *opt_isnan = _gvn.transform(ifisnan);
4507 assert( opt_isnan->is_If(), "Expect an IfNode");
4508 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4509 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4510
4511 set_control(iftrue);
4512
4513 static const jint nan_bits = 0x7fc00000;
4514 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4515 phi->init_req(1, _gvn.transform( slow_result ));
4516 r->init_req(1, iftrue);
4517
4518 // Else fall through
4519 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4520 set_control(iffalse);
4521
4522 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4523 r->init_req(2, iffalse);
4524
4525 // Post merge
4526 set_control(_gvn.transform(r));
4527 record_for_igvn(r);
4528
4529 C->set_has_split_ifs(true); // Has chance for split-if optimization
4530 result = phi;
4531 assert(result->bottom_type()->isa_int(), "must be");
4532 break;
4533 }
4534
4535 default:
4536 fatal_unexpected_iid(id);
4537 break;
4538 }
4539 set_result(_gvn.transform(result));
4540 return true;
4541 }
4542
4543 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4544 Node* arg = argument(0);
4545 Node* result = NULL;
4546
4547 switch (id) {
4548 case vmIntrinsics::_floatIsInfinite:
4549 result = new IsInfiniteFNode(arg);
4550 break;
4551 case vmIntrinsics::_floatIsFinite:
4552 result = new IsFiniteFNode(arg);
4553 break;
4554 case vmIntrinsics::_doubleIsInfinite:
4555 result = new IsInfiniteDNode(arg);
4556 break;
4557 case vmIntrinsics::_doubleIsFinite:
4558 result = new IsFiniteDNode(arg);
4559 break;
4560 default:
4561 fatal_unexpected_iid(id);
4562 break;
4563 }
4564 set_result(_gvn.transform(result));
4565 return true;
4566 }
4567
4568 //----------------------inline_unsafe_copyMemory-------------------------
4569 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4570
4571 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4572 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4573 const Type* base_t = gvn.type(base);
4574
4575 bool in_native = (base_t == TypePtr::NULL_PTR);
4576 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4577 bool is_mixed = !in_heap && !in_native;
4578
4579 if (is_mixed) {
4580 return true; // mixed accesses can touch both on-heap and off-heap memory
4581 }
4582 if (in_heap) {
4583 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4584 if (!is_prim_array) {
4585 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4586 // there's not enough type information available to determine proper memory slice for it.
4587 return true;
4588 }
4589 }
4590 return false;
4591 }
4592
4593 bool LibraryCallKit::inline_unsafe_copyMemory() {
4594 if (callee()->is_static()) return false; // caller must have the capability!
4595 null_check_receiver(); // null-check receiver
4596 if (stopped()) return true;
4597
4598 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4599
4600 Node* src_base = argument(1); // type: oop
4601 Node* src_off = ConvL2X(argument(2)); // type: long
4602 Node* dst_base = argument(4); // type: oop
4603 Node* dst_off = ConvL2X(argument(5)); // type: long
4604 Node* size = ConvL2X(argument(7)); // type: long
4605
4606 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4607 "fieldOffset must be byte-scaled");
4608
4609 Node* src_addr = make_unsafe_address(src_base, src_off);
4610 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4611
4612 Node* thread = _gvn.transform(new ThreadLocalNode());
4613 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4614 BasicType doing_unsafe_access_bt = T_BYTE;
4615 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4616
4617 // update volatile field
4618 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4619
4620 int flags = RC_LEAF | RC_NO_FP;
4621
4622 const TypePtr* dst_type = TypePtr::BOTTOM;
4623
4624 // Adjust memory effects of the runtime call based on input values.
4625 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4626 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4627 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4628
4629 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4630 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4631 flags |= RC_NARROW_MEM; // narrow in memory
4632 }
4633 }
4634
4635 // Call it. Note that the length argument is not scaled.
4636 make_runtime_call(flags,
4637 OptoRuntime::fast_arraycopy_Type(),
4638 StubRoutines::unsafe_arraycopy(),
4639 "unsafe_arraycopy",
4640 dst_type,
4641 src_addr, dst_addr, size XTOP);
4642
4643 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4644
4645 return true;
4646 }
4647
4648 #undef XTOP
4649
4650 //------------------------clone_coping-----------------------------------
4651 // Helper function for inline_native_clone.
4652 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4653 assert(obj_size != NULL, "");
4654 Node* raw_obj = alloc_obj->in(1);
4655 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4656
4657 AllocateNode* alloc = NULL;
4658 if (ReduceBulkZeroing) {
4659 // We will be completely responsible for initializing this object -
4660 // mark Initialize node as complete.
4661 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4662 // The object was just allocated - there should be no any stores!
4663 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4664 // Mark as complete_with_arraycopy so that on AllocateNode
4665 // expansion, we know this AllocateNode is initialized by an array
4666 // copy and a StoreStore barrier exists after the array copy.
4667 alloc->initialization()->set_complete_with_arraycopy();
4668 }
4669
4670 Node* size = _gvn.transform(obj_size);
4671 access_clone(obj, alloc_obj, size, is_array);
4672
4673 // Do not let reads from the cloned object float above the arraycopy.
4674 if (alloc != NULL) {
4675 // Do not let stores that initialize this object be reordered with
4676 // a subsequent store that would make this object accessible by
4677 // other threads.
4678 // Record what AllocateNode this StoreStore protects so that
4679 // escape analysis can go from the MemBarStoreStoreNode to the
4680 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4681 // based on the escape status of the AllocateNode.
4682 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4683 } else {
4684 insert_mem_bar(Op_MemBarCPUOrder);
4685 }
4686 }
4687
4688 //------------------------inline_native_clone----------------------------
4689 // protected native Object java.lang.Object.clone();
4690 //
4691 // Here are the simple edge cases:
4692 // null receiver => normal trap
4693 // virtual and clone was overridden => slow path to out-of-line clone
4694 // not cloneable or finalizer => slow path to out-of-line Object.clone
4695 //
4696 // The general case has two steps, allocation and copying.
4697 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4698 //
4699 // Copying also has two cases, oop arrays and everything else.
4700 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4701 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4702 //
4703 // These steps fold up nicely if and when the cloned object's klass
4704 // can be sharply typed as an object array, a type array, or an instance.
4705 //
4706 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4707 PhiNode* result_val;
4708
4709 // Set the reexecute bit for the interpreter to reexecute
4710 // the bytecode that invokes Object.clone if deoptimization happens.
4711 { PreserveReexecuteState preexecs(this);
4712 jvms()->set_should_reexecute(true);
4713
4714 Node* obj = null_check_receiver();
4715 if (stopped()) return true;
4716
4717 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4718
4719 // If we are going to clone an instance, we need its exact type to
4720 // know the number and types of fields to convert the clone to
4721 // loads/stores. Maybe a speculative type can help us.
4722 if (!obj_type->klass_is_exact() &&
4723 obj_type->speculative_type() != NULL &&
4724 obj_type->speculative_type()->is_instance_klass()) {
4725 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4726 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4727 !spec_ik->has_injected_fields()) {
4728 if (!obj_type->isa_instptr() ||
4729 obj_type->is_instptr()->instance_klass()->has_subklass()) {
4730 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4731 }
4732 }
4733 }
4734
4735 // Conservatively insert a memory barrier on all memory slices.
4736 // Do not let writes into the original float below the clone.
4737 insert_mem_bar(Op_MemBarCPUOrder);
4738
4739 // paths into result_reg:
4740 enum {
4741 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4742 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4743 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4744 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4745 PATH_LIMIT
4746 };
4747 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4748 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4749 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4750 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4751 record_for_igvn(result_reg);
4752
4753 Node* obj_klass = load_object_klass(obj);
4754 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4755 if (array_ctl != NULL) {
4756 // It's an array.
4757 PreserveJVMState pjvms(this);
4758 set_control(array_ctl);
4759 Node* obj_length = load_array_length(obj);
4760 Node* obj_size = NULL;
4761 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4762
4763 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4764 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4765 // If it is an oop array, it requires very special treatment,
4766 // because gc barriers are required when accessing the array.
4767 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4768 if (is_obja != NULL) {
4769 PreserveJVMState pjvms2(this);
4770 set_control(is_obja);
4771 // Generate a direct call to the right arraycopy function(s).
4772 // Clones are always tightly coupled.
4773 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4774 ac->set_clone_oop_array();
4775 Node* n = _gvn.transform(ac);
4776 assert(n == ac, "cannot disappear");
4777 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4778
4779 result_reg->init_req(_objArray_path, control());
4780 result_val->init_req(_objArray_path, alloc_obj);
4781 result_i_o ->set_req(_objArray_path, i_o());
4782 result_mem ->set_req(_objArray_path, reset_memory());
4783 }
4784 }
4785 // Otherwise, there are no barriers to worry about.
4786 // (We can dispense with card marks if we know the allocation
4787 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4788 // causes the non-eden paths to take compensating steps to
4789 // simulate a fresh allocation, so that no further
4790 // card marks are required in compiled code to initialize
4791 // the object.)
4792
4793 if (!stopped()) {
4794 copy_to_clone(obj, alloc_obj, obj_size, true);
4795
4796 // Present the results of the copy.
4797 result_reg->init_req(_array_path, control());
4798 result_val->init_req(_array_path, alloc_obj);
4799 result_i_o ->set_req(_array_path, i_o());
4800 result_mem ->set_req(_array_path, reset_memory());
4801 }
4802 }
4803
4804 // We only go to the instance fast case code if we pass a number of guards.
4805 // The paths which do not pass are accumulated in the slow_region.
4806 RegionNode* slow_region = new RegionNode(1);
4807 record_for_igvn(slow_region);
4808 if (!stopped()) {
4809 // It's an instance (we did array above). Make the slow-path tests.
4810 // If this is a virtual call, we generate a funny guard. We grab
4811 // the vtable entry corresponding to clone() from the target object.
4812 // If the target method which we are calling happens to be the
4813 // Object clone() method, we pass the guard. We do not need this
4814 // guard for non-virtual calls; the caller is known to be the native
4815 // Object clone().
4816 if (is_virtual) {
4817 generate_virtual_guard(obj_klass, slow_region);
4818 }
4819
4820 // The object must be easily cloneable and must not have a finalizer.
4821 // Both of these conditions may be checked in a single test.
4822 // We could optimize the test further, but we don't care.
4823 generate_access_flags_guard(obj_klass,
4824 // Test both conditions:
4825 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4826 // Must be cloneable but not finalizer:
4827 JVM_ACC_IS_CLONEABLE_FAST,
4828 slow_region);
4829 }
4830
4831 if (!stopped()) {
4832 // It's an instance, and it passed the slow-path tests.
4833 PreserveJVMState pjvms(this);
4834 Node* obj_size = NULL;
4835 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4836 // is reexecuted if deoptimization occurs and there could be problems when merging
4837 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4838 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4839
4840 copy_to_clone(obj, alloc_obj, obj_size, false);
4841
4842 // Present the results of the slow call.
4843 result_reg->init_req(_instance_path, control());
4844 result_val->init_req(_instance_path, alloc_obj);
4845 result_i_o ->set_req(_instance_path, i_o());
4846 result_mem ->set_req(_instance_path, reset_memory());
4847 }
4848
4849 // Generate code for the slow case. We make a call to clone().
4850 set_control(_gvn.transform(slow_region));
4851 if (!stopped()) {
4852 PreserveJVMState pjvms(this);
4853 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4854 // We need to deoptimize on exception (see comment above)
4855 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4856 // this->control() comes from set_results_for_java_call
4857 result_reg->init_req(_slow_path, control());
4858 result_val->init_req(_slow_path, slow_result);
4859 result_i_o ->set_req(_slow_path, i_o());
4860 result_mem ->set_req(_slow_path, reset_memory());
4861 }
4862
4863 // Return the combined state.
4864 set_control( _gvn.transform(result_reg));
4865 set_i_o( _gvn.transform(result_i_o));
4866 set_all_memory( _gvn.transform(result_mem));
4867 } // original reexecute is set back here
4868
4869 set_result(_gvn.transform(result_val));
4870 return true;
4871 }
4872
4873 // If we have a tightly coupled allocation, the arraycopy may take care
4874 // of the array initialization. If one of the guards we insert between
4875 // the allocation and the arraycopy causes a deoptimization, an
4876 // uninitialized array will escape the compiled method. To prevent that
4877 // we set the JVM state for uncommon traps between the allocation and
4878 // the arraycopy to the state before the allocation so, in case of
4879 // deoptimization, we'll reexecute the allocation and the
4880 // initialization.
4881 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4882 if (alloc != NULL) {
4883 ciMethod* trap_method = alloc->jvms()->method();
4884 int trap_bci = alloc->jvms()->bci();
4885
4886 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4887 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4888 // Make sure there's no store between the allocation and the
4889 // arraycopy otherwise visible side effects could be rexecuted
4890 // in case of deoptimization and cause incorrect execution.
4891 bool no_interfering_store = true;
4892 Node* mem = alloc->in(TypeFunc::Memory);
4893 if (mem->is_MergeMem()) {
4894 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4895 Node* n = mms.memory();
4896 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4897 assert(n->is_Store(), "what else?");
4898 no_interfering_store = false;
4899 break;
4900 }
4901 }
4902 } else {
4903 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4904 Node* n = mms.memory();
4905 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4906 assert(n->is_Store(), "what else?");
4907 no_interfering_store = false;
4908 break;
4909 }
4910 }
4911 }
4912
4913 if (no_interfering_store) {
4914 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4915 uint size = alloc->req();
4916 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4917 old_jvms->set_map(sfpt);
4918 for (uint i = 0; i < size; i++) {
4919 sfpt->init_req(i, alloc->in(i));
4920 }
4921 // re-push array length for deoptimization
4922 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4923 old_jvms->set_sp(old_jvms->sp()+1);
4924 old_jvms->set_monoff(old_jvms->monoff()+1);
4925 old_jvms->set_scloff(old_jvms->scloff()+1);
4926 old_jvms->set_endoff(old_jvms->endoff()+1);
4927 old_jvms->set_should_reexecute(true);
4928
4929 sfpt->set_i_o(map()->i_o());
4930 sfpt->set_memory(map()->memory());
4931 sfpt->set_control(map()->control());
4932
4933 JVMState* saved_jvms = jvms();
4934 saved_reexecute_sp = _reexecute_sp;
4935
4936 set_jvms(sfpt->jvms());
4937 _reexecute_sp = jvms()->sp();
4938
4939 return saved_jvms;
4940 }
4941 }
4942 }
4943 return NULL;
4944 }
4945
4946 // In case of a deoptimization, we restart execution at the
4947 // allocation, allocating a new array. We would leave an uninitialized
4948 // array in the heap that GCs wouldn't expect. Move the allocation
4949 // after the traps so we don't allocate the array if we
4950 // deoptimize. This is possible because tightly_coupled_allocation()
4951 // guarantees there's no observer of the allocated array at this point
4952 // and the control flow is simple enough.
4953 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4954 int saved_reexecute_sp, uint new_idx) {
4955 if (saved_jvms != NULL && !stopped()) {
4956 assert(alloc != NULL, "only with a tightly coupled allocation");
4957 // restore JVM state to the state at the arraycopy
4958 saved_jvms->map()->set_control(map()->control());
4959 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4960 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4961 // If we've improved the types of some nodes (null check) while
4962 // emitting the guards, propagate them to the current state
4963 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4964 set_jvms(saved_jvms);
4965 _reexecute_sp = saved_reexecute_sp;
4966
4967 // Remove the allocation from above the guards
4968 CallProjections callprojs;
4969 alloc->extract_projections(&callprojs, true);
4970 InitializeNode* init = alloc->initialization();
4971 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4972 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4973 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4974
4975 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4976 // the allocation (i.e. is only valid if the allocation succeeds):
4977 // 1) replace CastIINode with AllocateArrayNode's length here
4978 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4979 //
4980 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4981 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4982 Node* init_control = init->proj_out(TypeFunc::Control);
4983 Node* alloc_length = alloc->Ideal_length();
4984 #ifdef ASSERT
4985 Node* prev_cast = NULL;
4986 #endif
4987 for (uint i = 0; i < init_control->outcnt(); i++) {
4988 Node* init_out = init_control->raw_out(i);
4989 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4990 #ifdef ASSERT
4991 if (prev_cast == NULL) {
4992 prev_cast = init_out;
4993 } else {
4994 if (prev_cast->cmp(*init_out) == false) {
4995 prev_cast->dump();
4996 init_out->dump();
4997 assert(false, "not equal CastIINode");
4998 }
4999 }
5000 #endif
5001 C->gvn_replace_by(init_out, alloc_length);
5002 }
5003 }
5004 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5005
5006 // move the allocation here (after the guards)
5007 _gvn.hash_delete(alloc);
5008 alloc->set_req(TypeFunc::Control, control());
5009 alloc->set_req(TypeFunc::I_O, i_o());
5010 Node *mem = reset_memory();
5011 set_all_memory(mem);
5012 alloc->set_req(TypeFunc::Memory, mem);
5013 set_control(init->proj_out_or_null(TypeFunc::Control));
5014 set_i_o(callprojs.fallthrough_ioproj);
5015
5016 // Update memory as done in GraphKit::set_output_for_allocation()
5017 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5018 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5019 if (ary_type->isa_aryptr() && length_type != NULL) {
5020 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5021 }
5022 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5023 int elemidx = C->get_alias_index(telemref);
5024 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5025 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5026
5027 Node* allocx = _gvn.transform(alloc);
5028 assert(allocx == alloc, "where has the allocation gone?");
5029 assert(dest->is_CheckCastPP(), "not an allocation result?");
5030
5031 _gvn.hash_delete(dest);
5032 dest->set_req(0, control());
5033 Node* destx = _gvn.transform(dest);
5034 assert(destx == dest, "where has the allocation result gone?");
5035
5036 array_ideal_length(alloc, ary_type, true);
5037 }
5038 }
5039
5040
5041 //------------------------------inline_arraycopy-----------------------
5042 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5043 // Object dest, int destPos,
5044 // int length);
5045 bool LibraryCallKit::inline_arraycopy() {
5046 // Get the arguments.
5047 Node* src = argument(0); // type: oop
5048 Node* src_offset = argument(1); // type: int
5049 Node* dest = argument(2); // type: oop
5050 Node* dest_offset = argument(3); // type: int
5051 Node* length = argument(4); // type: int
5052
5053 uint new_idx = C->unique();
5054
5055 // Check for allocation before we add nodes that would confuse
5056 // tightly_coupled_allocation()
5057 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5058
5059 int saved_reexecute_sp = -1;
5060 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5061 // See arraycopy_restore_alloc_state() comment
5062 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5063 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5064 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5065 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5066
5067 // The following tests must be performed
5068 // (1) src and dest are arrays.
5069 // (2) src and dest arrays must have elements of the same BasicType
5070 // (3) src and dest must not be null.
5071 // (4) src_offset must not be negative.
5072 // (5) dest_offset must not be negative.
5073 // (6) length must not be negative.
5074 // (7) src_offset + length must not exceed length of src.
5075 // (8) dest_offset + length must not exceed length of dest.
5076 // (9) each element of an oop array must be assignable
5077
5078 // (3) src and dest must not be null.
5079 // always do this here because we need the JVM state for uncommon traps
5080 Node* null_ctl = top();
5081 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5082 assert(null_ctl->is_top(), "no null control here");
5083 dest = null_check(dest, T_ARRAY);
5084
5085 if (!can_emit_guards) {
5086 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5087 // guards but the arraycopy node could still take advantage of a
5088 // tightly allocated allocation. tightly_coupled_allocation() is
5089 // called again to make sure it takes the null check above into
5090 // account: the null check is mandatory and if it caused an
5091 // uncommon trap to be emitted then the allocation can't be
5092 // considered tightly coupled in this context.
5093 alloc = tightly_coupled_allocation(dest);
5094 }
5095
5096 bool validated = false;
5097
5098 const Type* src_type = _gvn.type(src);
5099 const Type* dest_type = _gvn.type(dest);
5100 const TypeAryPtr* top_src = src_type->isa_aryptr();
5101 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5102
5103 // Do we have the type of src?
5104 bool has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5105 // Do we have the type of dest?
5106 bool has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5107 // Is the type for src from speculation?
5108 bool src_spec = false;
5109 // Is the type for dest from speculation?
5110 bool dest_spec = false;
5111
5112 if ((!has_src || !has_dest) && can_emit_guards) {
5113 // We don't have sufficient type information, let's see if
5114 // speculative types can help. We need to have types for both src
5115 // and dest so that it pays off.
5116
5117 // Do we already have or could we have type information for src
5118 bool could_have_src = has_src;
5119 // Do we already have or could we have type information for dest
5120 bool could_have_dest = has_dest;
5121
5122 ciKlass* src_k = NULL;
5123 if (!has_src) {
5124 src_k = src_type->speculative_type_not_null();
5125 if (src_k != NULL && src_k->is_array_klass()) {
5126 could_have_src = true;
5127 }
5128 }
5129
5130 ciKlass* dest_k = NULL;
5131 if (!has_dest) {
5132 dest_k = dest_type->speculative_type_not_null();
5133 if (dest_k != NULL && dest_k->is_array_klass()) {
5134 could_have_dest = true;
5135 }
5136 }
5137
5138 if (could_have_src && could_have_dest) {
5139 // This is going to pay off so emit the required guards
5140 if (!has_src) {
5141 src = maybe_cast_profiled_obj(src, src_k, true);
5142 src_type = _gvn.type(src);
5143 top_src = src_type->isa_aryptr();
5144 has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5145 src_spec = true;
5146 }
5147 if (!has_dest) {
5148 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5149 dest_type = _gvn.type(dest);
5150 top_dest = dest_type->isa_aryptr();
5151 has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5152 dest_spec = true;
5153 }
5154 }
5155 }
5156
5157 if (has_src && has_dest && can_emit_guards) {
5158 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5159 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5160 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5161 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5162
5163 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5164 // If both arrays are object arrays then having the exact types
5165 // for both will remove the need for a subtype check at runtime
5166 // before the call and may make it possible to pick a faster copy
5167 // routine (without a subtype check on every element)
5168 // Do we have the exact type of src?
5169 bool could_have_src = src_spec;
5170 // Do we have the exact type of dest?
5171 bool could_have_dest = dest_spec;
5172 ciKlass* src_k = NULL;
5173 ciKlass* dest_k = NULL;
5174 if (!src_spec) {
5175 src_k = src_type->speculative_type_not_null();
5176 if (src_k != NULL && src_k->is_array_klass()) {
5177 could_have_src = true;
5178 }
5179 }
5180 if (!dest_spec) {
5181 dest_k = dest_type->speculative_type_not_null();
5182 if (dest_k != NULL && dest_k->is_array_klass()) {
5183 could_have_dest = true;
5184 }
5185 }
5186 if (could_have_src && could_have_dest) {
5187 // If we can have both exact types, emit the missing guards
5188 if (could_have_src && !src_spec) {
5189 src = maybe_cast_profiled_obj(src, src_k, true);
5190 }
5191 if (could_have_dest && !dest_spec) {
5192 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5193 }
5194 }
5195 }
5196 }
5197
5198 ciMethod* trap_method = method();
5199 int trap_bci = bci();
5200 if (saved_jvms != NULL) {
5201 trap_method = alloc->jvms()->method();
5202 trap_bci = alloc->jvms()->bci();
5203 }
5204
5205 bool negative_length_guard_generated = false;
5206
5207 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5208 can_emit_guards &&
5209 !src->is_top() && !dest->is_top()) {
5210 // validate arguments: enables transformation the ArrayCopyNode
5211 validated = true;
5212
5213 RegionNode* slow_region = new RegionNode(1);
5214 record_for_igvn(slow_region);
5215
5216 // (1) src and dest are arrays.
5217 generate_non_array_guard(load_object_klass(src), slow_region);
5218 generate_non_array_guard(load_object_klass(dest), slow_region);
5219
5220 // (2) src and dest arrays must have elements of the same BasicType
5221 // done at macro expansion or at Ideal transformation time
5222
5223 // (4) src_offset must not be negative.
5224 generate_negative_guard(src_offset, slow_region);
5225
5226 // (5) dest_offset must not be negative.
5227 generate_negative_guard(dest_offset, slow_region);
5228
5229 // (7) src_offset + length must not exceed length of src.
5230 generate_limit_guard(src_offset, length,
5231 load_array_length(src),
5232 slow_region);
5233
5234 // (8) dest_offset + length must not exceed length of dest.
5235 generate_limit_guard(dest_offset, length,
5236 load_array_length(dest),
5237 slow_region);
5238
5239 // (6) length must not be negative.
5240 // This is also checked in generate_arraycopy() during macro expansion, but
5241 // we also have to check it here for the case where the ArrayCopyNode will
5242 // be eliminated by Escape Analysis.
5243 if (EliminateAllocations) {
5244 generate_negative_guard(length, slow_region);
5245 negative_length_guard_generated = true;
5246 }
5247
5248 // (9) each element of an oop array must be assignable
5249 Node* dest_klass = load_object_klass(dest);
5250 if (src != dest) {
5251 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5252
5253 if (not_subtype_ctrl != top()) {
5254 PreserveJVMState pjvms(this);
5255 set_control(not_subtype_ctrl);
5256 uncommon_trap(Deoptimization::Reason_intrinsic,
5257 Deoptimization::Action_make_not_entrant);
5258 assert(stopped(), "Should be stopped");
5259 }
5260 }
5261 {
5262 PreserveJVMState pjvms(this);
5263 set_control(_gvn.transform(slow_region));
5264 uncommon_trap(Deoptimization::Reason_intrinsic,
5265 Deoptimization::Action_make_not_entrant);
5266 assert(stopped(), "Should be stopped");
5267 }
5268
5269 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5270 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5271 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5272 }
5273
5274 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5275
5276 if (stopped()) {
5277 return true;
5278 }
5279
5280 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5281 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5282 // so the compiler has a chance to eliminate them: during macro expansion,
5283 // we have to set their control (CastPP nodes are eliminated).
5284 load_object_klass(src), load_object_klass(dest),
5285 load_array_length(src), load_array_length(dest));
5286
5287 ac->set_arraycopy(validated);
5288
5289 Node* n = _gvn.transform(ac);
5290 if (n == ac) {
5291 ac->connect_outputs(this);
5292 } else {
5293 assert(validated, "shouldn't transform if all arguments not validated");
5294 set_all_memory(n);
5295 }
5296 clear_upper_avx();
5297
5298
5299 return true;
5300 }
5301
5302
5303 // Helper function which determines if an arraycopy immediately follows
5304 // an allocation, with no intervening tests or other escapes for the object.
5305 AllocateArrayNode*
5306 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5307 if (stopped()) return NULL; // no fast path
5308 if (!C->do_aliasing()) return NULL; // no MergeMems around
5309
5310 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5311 if (alloc == NULL) return NULL;
5312
5313 Node* rawmem = memory(Compile::AliasIdxRaw);
5314 // Is the allocation's memory state untouched?
5315 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5316 // Bail out if there have been raw-memory effects since the allocation.
5317 // (Example: There might have been a call or safepoint.)
5318 return NULL;
5319 }
5320 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5321 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5322 return NULL;
5323 }
5324
5325 // There must be no unexpected observers of this allocation.
5326 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5327 Node* obs = ptr->fast_out(i);
5328 if (obs != this->map()) {
5329 return NULL;
5330 }
5331 }
5332
5333 // This arraycopy must unconditionally follow the allocation of the ptr.
5334 Node* alloc_ctl = ptr->in(0);
5335 Node* ctl = control();
5336 while (ctl != alloc_ctl) {
5337 // There may be guards which feed into the slow_region.
5338 // Any other control flow means that we might not get a chance
5339 // to finish initializing the allocated object.
5340 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5341 IfNode* iff = ctl->in(0)->as_If();
5342 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5343 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5344 // One more try: Various low-level checks bottom out in
5345 // uncommon traps. If the debug-info of the trap omits
5346 // any reference to the allocation, as we've already
5347 // observed, then there can be no objection to the trap.
5348 bool found_trap = false;
5349 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5350 Node* obs = not_ctl->fast_out(j);
5351 if (obs->in(0) == not_ctl && obs->is_Call() &&
5352 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5353 found_trap = true; break;
5354 }
5355 }
5356 if (found_trap) {
5357 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5358 continue;
5359 }
5360 }
5361 return NULL;
5362 }
5363
5364 // If we get this far, we have an allocation which immediately
5365 // precedes the arraycopy, and we can take over zeroing the new object.
5366 // The arraycopy will finish the initialization, and provide
5367 // a new control state to which we will anchor the destination pointer.
5368
5369 return alloc;
5370 }
5371
5372 //-------------inline_encodeISOArray-----------------------------------
5373 // encode char[] to byte[] in ISO_8859_1 or ASCII
5374 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5375 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5376 // no receiver since it is static method
5377 Node *src = argument(0);
5378 Node *src_offset = argument(1);
5379 Node *dst = argument(2);
5380 Node *dst_offset = argument(3);
5381 Node *length = argument(4);
5382
5383 src = must_be_not_null(src, true);
5384 dst = must_be_not_null(dst, true);
5385
5386 const Type* src_type = src->Value(&_gvn);
5387 const Type* dst_type = dst->Value(&_gvn);
5388 const TypeAryPtr* top_src = src_type->isa_aryptr();
5389 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5390 if (top_src == NULL || top_src->elem() == Type::BOTTOM ||
5391 top_dest == NULL || top_dest->elem() == Type::BOTTOM) {
5392 // failed array check
5393 return false;
5394 }
5395
5396 // Figure out the size and type of the elements we will be copying.
5397 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5398 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
5399 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5400 return false;
5401 }
5402
5403 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5404 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5405 // 'src_start' points to src array + scaled offset
5406 // 'dst_start' points to dst array + scaled offset
5407
5408 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5409 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5410 enc = _gvn.transform(enc);
5411 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5412 set_memory(res_mem, mtype);
5413 set_result(enc);
5414 clear_upper_avx();
5415
5416 return true;
5417 }
5418
5419 //-------------inline_multiplyToLen-----------------------------------
5420 bool LibraryCallKit::inline_multiplyToLen() {
5421 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5422
5423 address stubAddr = StubRoutines::multiplyToLen();
5424 if (stubAddr == NULL) {
5425 return false; // Intrinsic's stub is not implemented on this platform
5426 }
5427 const char* stubName = "multiplyToLen";
5428
5429 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5430
5431 // no receiver because it is a static method
5432 Node* x = argument(0);
5433 Node* xlen = argument(1);
5434 Node* y = argument(2);
5435 Node* ylen = argument(3);
5436 Node* z = argument(4);
5437
5438 x = must_be_not_null(x, true);
5439 y = must_be_not_null(y, true);
5440
5441 const Type* x_type = x->Value(&_gvn);
5442 const Type* y_type = y->Value(&_gvn);
5443 const TypeAryPtr* top_x = x_type->isa_aryptr();
5444 const TypeAryPtr* top_y = y_type->isa_aryptr();
5445 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5446 top_y == NULL || top_y->elem() == Type::BOTTOM) {
5447 // failed array check
5448 return false;
5449 }
5450
5451 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5452 BasicType y_elem = y_type->isa_aryptr()->elem()->array_element_basic_type();
5453 if (x_elem != T_INT || y_elem != T_INT) {
5454 return false;
5455 }
5456
5457 // Set the original stack and the reexecute bit for the interpreter to reexecute
5458 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5459 // on the return from z array allocation in runtime.
5460 { PreserveReexecuteState preexecs(this);
5461 jvms()->set_should_reexecute(true);
5462
5463 Node* x_start = array_element_address(x, intcon(0), x_elem);
5464 Node* y_start = array_element_address(y, intcon(0), y_elem);
5465 // 'x_start' points to x array + scaled xlen
5466 // 'y_start' points to y array + scaled ylen
5467
5468 // Allocate the result array
5469 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5470 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5471 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5472
5473 IdealKit ideal(this);
5474
5475 #define __ ideal.
5476 Node* one = __ ConI(1);
5477 Node* zero = __ ConI(0);
5478 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5479 __ set(need_alloc, zero);
5480 __ set(z_alloc, z);
5481 __ if_then(z, BoolTest::eq, null()); {
5482 __ increment (need_alloc, one);
5483 } __ else_(); {
5484 // Update graphKit memory and control from IdealKit.
5485 sync_kit(ideal);
5486 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5487 cast->init_req(0, control());
5488 _gvn.set_type(cast, cast->bottom_type());
5489 C->record_for_igvn(cast);
5490
5491 Node* zlen_arg = load_array_length(cast);
5492 // Update IdealKit memory and control from graphKit.
5493 __ sync_kit(this);
5494 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5495 __ increment (need_alloc, one);
5496 } __ end_if();
5497 } __ end_if();
5498
5499 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5500 // Update graphKit memory and control from IdealKit.
5501 sync_kit(ideal);
5502 Node * narr = new_array(klass_node, zlen, 1);
5503 // Update IdealKit memory and control from graphKit.
5504 __ sync_kit(this);
5505 __ set(z_alloc, narr);
5506 } __ end_if();
5507
5508 sync_kit(ideal);
5509 z = __ value(z_alloc);
5510 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5511 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5512 // Final sync IdealKit and GraphKit.
5513 final_sync(ideal);
5514 #undef __
5515
5516 Node* z_start = array_element_address(z, intcon(0), T_INT);
5517
5518 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5519 OptoRuntime::multiplyToLen_Type(),
5520 stubAddr, stubName, TypePtr::BOTTOM,
5521 x_start, xlen, y_start, ylen, z_start, zlen);
5522 } // original reexecute is set back here
5523
5524 C->set_has_split_ifs(true); // Has chance for split-if optimization
5525 set_result(z);
5526 return true;
5527 }
5528
5529 //-------------inline_squareToLen------------------------------------
5530 bool LibraryCallKit::inline_squareToLen() {
5531 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5532
5533 address stubAddr = StubRoutines::squareToLen();
5534 if (stubAddr == NULL) {
5535 return false; // Intrinsic's stub is not implemented on this platform
5536 }
5537 const char* stubName = "squareToLen";
5538
5539 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5540
5541 Node* x = argument(0);
5542 Node* len = argument(1);
5543 Node* z = argument(2);
5544 Node* zlen = argument(3);
5545
5546 x = must_be_not_null(x, true);
5547 z = must_be_not_null(z, true);
5548
5549 const Type* x_type = x->Value(&_gvn);
5550 const Type* z_type = z->Value(&_gvn);
5551 const TypeAryPtr* top_x = x_type->isa_aryptr();
5552 const TypeAryPtr* top_z = z_type->isa_aryptr();
5553 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5554 top_z == NULL || top_z->elem() == Type::BOTTOM) {
5555 // failed array check
5556 return false;
5557 }
5558
5559 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5560 BasicType z_elem = z_type->isa_aryptr()->elem()->array_element_basic_type();
5561 if (x_elem != T_INT || z_elem != T_INT) {
5562 return false;
5563 }
5564
5565
5566 Node* x_start = array_element_address(x, intcon(0), x_elem);
5567 Node* z_start = array_element_address(z, intcon(0), z_elem);
5568
5569 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5570 OptoRuntime::squareToLen_Type(),
5571 stubAddr, stubName, TypePtr::BOTTOM,
5572 x_start, len, z_start, zlen);
5573
5574 set_result(z);
5575 return true;
5576 }
5577
5578 //-------------inline_mulAdd------------------------------------------
5579 bool LibraryCallKit::inline_mulAdd() {
5580 assert(UseMulAddIntrinsic, "not implemented on this platform");
5581
5582 address stubAddr = StubRoutines::mulAdd();
5583 if (stubAddr == NULL) {
5584 return false; // Intrinsic's stub is not implemented on this platform
5585 }
5586 const char* stubName = "mulAdd";
5587
5588 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5589
5590 Node* out = argument(0);
5591 Node* in = argument(1);
5592 Node* offset = argument(2);
5593 Node* len = argument(3);
5594 Node* k = argument(4);
5595
5596 out = must_be_not_null(out, true);
5597
5598 const Type* out_type = out->Value(&_gvn);
5599 const Type* in_type = in->Value(&_gvn);
5600 const TypeAryPtr* top_out = out_type->isa_aryptr();
5601 const TypeAryPtr* top_in = in_type->isa_aryptr();
5602 if (top_out == NULL || top_out->elem() == Type::BOTTOM ||
5603 top_in == NULL || top_in->elem() == Type::BOTTOM) {
5604 // failed array check
5605 return false;
5606 }
5607
5608 BasicType out_elem = out_type->isa_aryptr()->elem()->array_element_basic_type();
5609 BasicType in_elem = in_type->isa_aryptr()->elem()->array_element_basic_type();
5610 if (out_elem != T_INT || in_elem != T_INT) {
5611 return false;
5612 }
5613
5614 Node* outlen = load_array_length(out);
5615 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5616 Node* out_start = array_element_address(out, intcon(0), out_elem);
5617 Node* in_start = array_element_address(in, intcon(0), in_elem);
5618
5619 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5620 OptoRuntime::mulAdd_Type(),
5621 stubAddr, stubName, TypePtr::BOTTOM,
5622 out_start,in_start, new_offset, len, k);
5623 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5624 set_result(result);
5625 return true;
5626 }
5627
5628 //-------------inline_montgomeryMultiply-----------------------------------
5629 bool LibraryCallKit::inline_montgomeryMultiply() {
5630 address stubAddr = StubRoutines::montgomeryMultiply();
5631 if (stubAddr == NULL) {
5632 return false; // Intrinsic's stub is not implemented on this platform
5633 }
5634
5635 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5636 const char* stubName = "montgomery_multiply";
5637
5638 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5639
5640 Node* a = argument(0);
5641 Node* b = argument(1);
5642 Node* n = argument(2);
5643 Node* len = argument(3);
5644 Node* inv = argument(4);
5645 Node* m = argument(6);
5646
5647 const Type* a_type = a->Value(&_gvn);
5648 const TypeAryPtr* top_a = a_type->isa_aryptr();
5649 const Type* b_type = b->Value(&_gvn);
5650 const TypeAryPtr* top_b = b_type->isa_aryptr();
5651 const Type* n_type = a->Value(&_gvn);
5652 const TypeAryPtr* top_n = n_type->isa_aryptr();
5653 const Type* m_type = a->Value(&_gvn);
5654 const TypeAryPtr* top_m = m_type->isa_aryptr();
5655 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5656 top_b == NULL || top_b->elem() == Type::BOTTOM ||
5657 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5658 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5659 // failed array check
5660 return false;
5661 }
5662
5663 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5664 BasicType b_elem = b_type->isa_aryptr()->elem()->array_element_basic_type();
5665 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5666 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5667 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5668 return false;
5669 }
5670
5671 // Make the call
5672 {
5673 Node* a_start = array_element_address(a, intcon(0), a_elem);
5674 Node* b_start = array_element_address(b, intcon(0), b_elem);
5675 Node* n_start = array_element_address(n, intcon(0), n_elem);
5676 Node* m_start = array_element_address(m, intcon(0), m_elem);
5677
5678 Node* call = make_runtime_call(RC_LEAF,
5679 OptoRuntime::montgomeryMultiply_Type(),
5680 stubAddr, stubName, TypePtr::BOTTOM,
5681 a_start, b_start, n_start, len, inv, top(),
5682 m_start);
5683 set_result(m);
5684 }
5685
5686 return true;
5687 }
5688
5689 bool LibraryCallKit::inline_montgomerySquare() {
5690 address stubAddr = StubRoutines::montgomerySquare();
5691 if (stubAddr == NULL) {
5692 return false; // Intrinsic's stub is not implemented on this platform
5693 }
5694
5695 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5696 const char* stubName = "montgomery_square";
5697
5698 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5699
5700 Node* a = argument(0);
5701 Node* n = argument(1);
5702 Node* len = argument(2);
5703 Node* inv = argument(3);
5704 Node* m = argument(5);
5705
5706 const Type* a_type = a->Value(&_gvn);
5707 const TypeAryPtr* top_a = a_type->isa_aryptr();
5708 const Type* n_type = a->Value(&_gvn);
5709 const TypeAryPtr* top_n = n_type->isa_aryptr();
5710 const Type* m_type = a->Value(&_gvn);
5711 const TypeAryPtr* top_m = m_type->isa_aryptr();
5712 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5713 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5714 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5715 // failed array check
5716 return false;
5717 }
5718
5719 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5720 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5721 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5722 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5723 return false;
5724 }
5725
5726 // Make the call
5727 {
5728 Node* a_start = array_element_address(a, intcon(0), a_elem);
5729 Node* n_start = array_element_address(n, intcon(0), n_elem);
5730 Node* m_start = array_element_address(m, intcon(0), m_elem);
5731
5732 Node* call = make_runtime_call(RC_LEAF,
5733 OptoRuntime::montgomerySquare_Type(),
5734 stubAddr, stubName, TypePtr::BOTTOM,
5735 a_start, n_start, len, inv, top(),
5736 m_start);
5737 set_result(m);
5738 }
5739
5740 return true;
5741 }
5742
5743 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5744 address stubAddr = NULL;
5745 const char* stubName = NULL;
5746
5747 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5748 if (stubAddr == NULL) {
5749 return false; // Intrinsic's stub is not implemented on this platform
5750 }
5751
5752 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5753
5754 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5755
5756 Node* newArr = argument(0);
5757 Node* oldArr = argument(1);
5758 Node* newIdx = argument(2);
5759 Node* shiftCount = argument(3);
5760 Node* numIter = argument(4);
5761
5762 const Type* newArr_type = newArr->Value(&_gvn);
5763 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5764 const Type* oldArr_type = oldArr->Value(&_gvn);
5765 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5766 if (top_newArr == NULL || top_newArr->elem() == Type::BOTTOM || top_oldArr == NULL
5767 || top_oldArr->elem() == Type::BOTTOM) {
5768 return false;
5769 }
5770
5771 BasicType newArr_elem = newArr_type->isa_aryptr()->elem()->array_element_basic_type();
5772 BasicType oldArr_elem = oldArr_type->isa_aryptr()->elem()->array_element_basic_type();
5773 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5774 return false;
5775 }
5776
5777 // Make the call
5778 {
5779 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5780 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5781
5782 Node* call = make_runtime_call(RC_LEAF,
5783 OptoRuntime::bigIntegerShift_Type(),
5784 stubAddr,
5785 stubName,
5786 TypePtr::BOTTOM,
5787 newArr_start,
5788 oldArr_start,
5789 newIdx,
5790 shiftCount,
5791 numIter);
5792 }
5793
5794 return true;
5795 }
5796
5797 //-------------inline_vectorizedMismatch------------------------------
5798 bool LibraryCallKit::inline_vectorizedMismatch() {
5799 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5800
5801 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5802 Node* obja = argument(0); // Object
5803 Node* aoffset = argument(1); // long
5804 Node* objb = argument(3); // Object
5805 Node* boffset = argument(4); // long
5806 Node* length = argument(6); // int
5807 Node* scale = argument(7); // int
5808
5809 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5810 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5811 if (obja_t == NULL || obja_t->elem() == Type::BOTTOM ||
5812 objb_t == NULL || objb_t->elem() == Type::BOTTOM ||
5813 scale == top()) {
5814 return false; // failed input validation
5815 }
5816
5817 Node* obja_adr = make_unsafe_address(obja, aoffset);
5818 Node* objb_adr = make_unsafe_address(objb, boffset);
5819
5820 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5821 //
5822 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5823 // if (length <= inline_limit) {
5824 // inline_path:
5825 // vmask = VectorMaskGen length
5826 // vload1 = LoadVectorMasked obja, vmask
5827 // vload2 = LoadVectorMasked objb, vmask
5828 // result1 = VectorCmpMasked vload1, vload2, vmask
5829 // } else {
5830 // call_stub_path:
5831 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5832 // }
5833 // exit_block:
5834 // return Phi(result1, result2);
5835 //
5836 enum { inline_path = 1, // input is small enough to process it all at once
5837 stub_path = 2, // input is too large; call into the VM
5838 PATH_LIMIT = 3
5839 };
5840
5841 Node* exit_block = new RegionNode(PATH_LIMIT);
5842 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5843 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5844
5845 Node* call_stub_path = control();
5846
5847 BasicType elem_bt = T_ILLEGAL;
5848
5849 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5850 if (scale_t->is_con()) {
5851 switch (scale_t->get_con()) {
5852 case 0: elem_bt = T_BYTE; break;
5853 case 1: elem_bt = T_SHORT; break;
5854 case 2: elem_bt = T_INT; break;
5855 case 3: elem_bt = T_LONG; break;
5856
5857 default: elem_bt = T_ILLEGAL; break; // not supported
5858 }
5859 }
5860
5861 int inline_limit = 0;
5862 bool do_partial_inline = false;
5863
5864 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5865 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5866 do_partial_inline = inline_limit >= 16;
5867 }
5868
5869 if (do_partial_inline) {
5870 assert(elem_bt != T_ILLEGAL, "sanity");
5871
5872 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5873 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5874 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5875
5876 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5877 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5878 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5879
5880 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5881
5882 if (!stopped()) {
5883 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5884
5885 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5886 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5887 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5888 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5889
5890 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
5891 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5892 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5893 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5894
5895 exit_block->init_req(inline_path, control());
5896 memory_phi->init_req(inline_path, map()->memory());
5897 result_phi->init_req(inline_path, result);
5898
5899 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5900 clear_upper_avx();
5901 }
5902 }
5903 }
5904
5905 if (call_stub_path != NULL) {
5906 set_control(call_stub_path);
5907
5908 Node* call = make_runtime_call(RC_LEAF,
5909 OptoRuntime::vectorizedMismatch_Type(),
5910 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5911 obja_adr, objb_adr, length, scale);
5912
5913 exit_block->init_req(stub_path, control());
5914 memory_phi->init_req(stub_path, map()->memory());
5915 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5916 }
5917
5918 exit_block = _gvn.transform(exit_block);
5919 memory_phi = _gvn.transform(memory_phi);
5920 result_phi = _gvn.transform(result_phi);
5921
5922 set_control(exit_block);
5923 set_all_memory(memory_phi);
5924 set_result(result_phi);
5925
5926 return true;
5927 }
5928
5929 /**
5930 * Calculate CRC32 for byte.
5931 * int java.util.zip.CRC32.update(int crc, int b)
5932 */
5933 bool LibraryCallKit::inline_updateCRC32() {
5934 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5935 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5936 // no receiver since it is static method
5937 Node* crc = argument(0); // type: int
5938 Node* b = argument(1); // type: int
5939
5940 /*
5941 * int c = ~ crc;
5942 * b = timesXtoThe32[(b ^ c) & 0xFF];
5943 * b = b ^ (c >>> 8);
5944 * crc = ~b;
5945 */
5946
5947 Node* M1 = intcon(-1);
5948 crc = _gvn.transform(new XorINode(crc, M1));
5949 Node* result = _gvn.transform(new XorINode(crc, b));
5950 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5951
5952 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5953 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5954 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5955 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5956
5957 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5958 result = _gvn.transform(new XorINode(crc, result));
5959 result = _gvn.transform(new XorINode(result, M1));
5960 set_result(result);
5961 return true;
5962 }
5963
5964 /**
5965 * Calculate CRC32 for byte[] array.
5966 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5967 */
5968 bool LibraryCallKit::inline_updateBytesCRC32() {
5969 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5970 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5971 // no receiver since it is static method
5972 Node* crc = argument(0); // type: int
5973 Node* src = argument(1); // type: oop
5974 Node* offset = argument(2); // type: int
5975 Node* length = argument(3); // type: int
5976
5977 const Type* src_type = src->Value(&_gvn);
5978 const TypeAryPtr* top_src = src_type->isa_aryptr();
5979 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
5980 // failed array check
5981 return false;
5982 }
5983
5984 // Figure out the size and type of the elements we will be copying.
5985 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5986 if (src_elem != T_BYTE) {
5987 return false;
5988 }
5989
5990 // 'src_start' points to src array + scaled offset
5991 src = must_be_not_null(src, true);
5992 Node* src_start = array_element_address(src, offset, src_elem);
5993
5994 // We assume that range check is done by caller.
5995 // TODO: generate range check (offset+length < src.length) in debug VM.
5996
5997 // Call the stub.
5998 address stubAddr = StubRoutines::updateBytesCRC32();
5999 const char *stubName = "updateBytesCRC32";
6000
6001 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6002 stubAddr, stubName, TypePtr::BOTTOM,
6003 crc, src_start, length);
6004 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6005 set_result(result);
6006 return true;
6007 }
6008
6009 /**
6010 * Calculate CRC32 for ByteBuffer.
6011 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6012 */
6013 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6014 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6015 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6016 // no receiver since it is static method
6017 Node* crc = argument(0); // type: int
6018 Node* src = argument(1); // type: long
6019 Node* offset = argument(3); // type: int
6020 Node* length = argument(4); // type: int
6021
6022 src = ConvL2X(src); // adjust Java long to machine word
6023 Node* base = _gvn.transform(new CastX2PNode(src));
6024 offset = ConvI2X(offset);
6025
6026 // 'src_start' points to src array + scaled offset
6027 Node* src_start = basic_plus_adr(top(), base, offset);
6028
6029 // Call the stub.
6030 address stubAddr = StubRoutines::updateBytesCRC32();
6031 const char *stubName = "updateBytesCRC32";
6032
6033 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6034 stubAddr, stubName, TypePtr::BOTTOM,
6035 crc, src_start, length);
6036 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6037 set_result(result);
6038 return true;
6039 }
6040
6041 //------------------------------get_table_from_crc32c_class-----------------------
6042 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6043 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6044 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
6045
6046 return table;
6047 }
6048
6049 //------------------------------inline_updateBytesCRC32C-----------------------
6050 //
6051 // Calculate CRC32C for byte[] array.
6052 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6053 //
6054 bool LibraryCallKit::inline_updateBytesCRC32C() {
6055 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6056 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6057 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6058 // no receiver since it is a static method
6059 Node* crc = argument(0); // type: int
6060 Node* src = argument(1); // type: oop
6061 Node* offset = argument(2); // type: int
6062 Node* end = argument(3); // type: int
6063
6064 Node* length = _gvn.transform(new SubINode(end, offset));
6065
6066 const Type* src_type = src->Value(&_gvn);
6067 const TypeAryPtr* top_src = src_type->isa_aryptr();
6068 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6069 // failed array check
6070 return false;
6071 }
6072
6073 // Figure out the size and type of the elements we will be copying.
6074 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6075 if (src_elem != T_BYTE) {
6076 return false;
6077 }
6078
6079 // 'src_start' points to src array + scaled offset
6080 src = must_be_not_null(src, true);
6081 Node* src_start = array_element_address(src, offset, src_elem);
6082
6083 // static final int[] byteTable in class CRC32C
6084 Node* table = get_table_from_crc32c_class(callee()->holder());
6085 table = must_be_not_null(table, true);
6086 Node* table_start = array_element_address(table, intcon(0), T_INT);
6087
6088 // We assume that range check is done by caller.
6089 // TODO: generate range check (offset+length < src.length) in debug VM.
6090
6091 // Call the stub.
6092 address stubAddr = StubRoutines::updateBytesCRC32C();
6093 const char *stubName = "updateBytesCRC32C";
6094
6095 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6096 stubAddr, stubName, TypePtr::BOTTOM,
6097 crc, src_start, length, table_start);
6098 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6099 set_result(result);
6100 return true;
6101 }
6102
6103 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6104 //
6105 // Calculate CRC32C for DirectByteBuffer.
6106 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6107 //
6108 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6109 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6110 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6111 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6112 // no receiver since it is a static method
6113 Node* crc = argument(0); // type: int
6114 Node* src = argument(1); // type: long
6115 Node* offset = argument(3); // type: int
6116 Node* end = argument(4); // type: int
6117
6118 Node* length = _gvn.transform(new SubINode(end, offset));
6119
6120 src = ConvL2X(src); // adjust Java long to machine word
6121 Node* base = _gvn.transform(new CastX2PNode(src));
6122 offset = ConvI2X(offset);
6123
6124 // 'src_start' points to src array + scaled offset
6125 Node* src_start = basic_plus_adr(top(), base, offset);
6126
6127 // static final int[] byteTable in class CRC32C
6128 Node* table = get_table_from_crc32c_class(callee()->holder());
6129 table = must_be_not_null(table, true);
6130 Node* table_start = array_element_address(table, intcon(0), T_INT);
6131
6132 // Call the stub.
6133 address stubAddr = StubRoutines::updateBytesCRC32C();
6134 const char *stubName = "updateBytesCRC32C";
6135
6136 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6137 stubAddr, stubName, TypePtr::BOTTOM,
6138 crc, src_start, length, table_start);
6139 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6140 set_result(result);
6141 return true;
6142 }
6143
6144 //------------------------------inline_updateBytesAdler32----------------------
6145 //
6146 // Calculate Adler32 checksum for byte[] array.
6147 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6148 //
6149 bool LibraryCallKit::inline_updateBytesAdler32() {
6150 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6151 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6152 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6153 // no receiver since it is static method
6154 Node* crc = argument(0); // type: int
6155 Node* src = argument(1); // type: oop
6156 Node* offset = argument(2); // type: int
6157 Node* length = argument(3); // type: int
6158
6159 const Type* src_type = src->Value(&_gvn);
6160 const TypeAryPtr* top_src = src_type->isa_aryptr();
6161 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6162 // failed array check
6163 return false;
6164 }
6165
6166 // Figure out the size and type of the elements we will be copying.
6167 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6168 if (src_elem != T_BYTE) {
6169 return false;
6170 }
6171
6172 // 'src_start' points to src array + scaled offset
6173 Node* src_start = array_element_address(src, offset, src_elem);
6174
6175 // We assume that range check is done by caller.
6176 // TODO: generate range check (offset+length < src.length) in debug VM.
6177
6178 // Call the stub.
6179 address stubAddr = StubRoutines::updateBytesAdler32();
6180 const char *stubName = "updateBytesAdler32";
6181
6182 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6183 stubAddr, stubName, TypePtr::BOTTOM,
6184 crc, src_start, length);
6185 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6186 set_result(result);
6187 return true;
6188 }
6189
6190 //------------------------------inline_updateByteBufferAdler32---------------
6191 //
6192 // Calculate Adler32 checksum for DirectByteBuffer.
6193 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6194 //
6195 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6196 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6197 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6198 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6199 // no receiver since it is static method
6200 Node* crc = argument(0); // type: int
6201 Node* src = argument(1); // type: long
6202 Node* offset = argument(3); // type: int
6203 Node* length = argument(4); // type: int
6204
6205 src = ConvL2X(src); // adjust Java long to machine word
6206 Node* base = _gvn.transform(new CastX2PNode(src));
6207 offset = ConvI2X(offset);
6208
6209 // 'src_start' points to src array + scaled offset
6210 Node* src_start = basic_plus_adr(top(), base, offset);
6211
6212 // Call the stub.
6213 address stubAddr = StubRoutines::updateBytesAdler32();
6214 const char *stubName = "updateBytesAdler32";
6215
6216 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6217 stubAddr, stubName, TypePtr::BOTTOM,
6218 crc, src_start, length);
6219
6220 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6221 set_result(result);
6222 return true;
6223 }
6224
6225 //----------------------------inline_reference_get----------------------------
6226 // public T java.lang.ref.Reference.get();
6227 bool LibraryCallKit::inline_reference_get() {
6228 const int referent_offset = java_lang_ref_Reference::referent_offset();
6229
6230 // Get the argument:
6231 Node* reference_obj = null_check_receiver();
6232 if (stopped()) return true;
6233
6234 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6235 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6236 decorators, /*is_static*/ false, NULL);
6237 if (result == NULL) return false;
6238
6239 // Add memory barrier to prevent commoning reads from this field
6240 // across safepoint since GC can change its value.
6241 insert_mem_bar(Op_MemBarCPUOrder);
6242
6243 set_result(result);
6244 return true;
6245 }
6246
6247 //----------------------------inline_reference_refersTo0----------------------------
6248 // bool java.lang.ref.Reference.refersTo0();
6249 // bool java.lang.ref.PhantomReference.refersTo0();
6250 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6251 // Get arguments:
6252 Node* reference_obj = null_check_receiver();
6253 Node* other_obj = argument(1);
6254 if (stopped()) return true;
6255
6256 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6257 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6258 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6259 decorators, /*is_static*/ false, NULL);
6260 if (referent == NULL) return false;
6261
6262 // Add memory barrier to prevent commoning reads from this field
6263 // across safepoint since GC can change its value.
6264 insert_mem_bar(Op_MemBarCPUOrder);
6265
6266 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6267 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6268 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6269
6270 RegionNode* region = new RegionNode(3);
6271 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6272
6273 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6274 region->init_req(1, if_true);
6275 phi->init_req(1, intcon(1));
6276
6277 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6278 region->init_req(2, if_false);
6279 phi->init_req(2, intcon(0));
6280
6281 set_control(_gvn.transform(region));
6282 record_for_igvn(region);
6283 set_result(_gvn.transform(phi));
6284 return true;
6285 }
6286
6287
6288 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6289 DecoratorSet decorators, bool is_static,
6290 ciInstanceKlass* fromKls) {
6291 if (fromKls == NULL) {
6292 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6293 assert(tinst != NULL, "obj is null");
6294 assert(tinst->is_loaded(), "obj is not loaded");
6295 fromKls = tinst->instance_klass();
6296 } else {
6297 assert(is_static, "only for static field access");
6298 }
6299 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6300 ciSymbol::make(fieldTypeString),
6301 is_static);
6302
6303 assert(field != NULL, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6304 if (field == NULL) return (Node *) NULL;
6305
6306 if (is_static) {
6307 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6308 fromObj = makecon(tip);
6309 }
6310
6311 // Next code copied from Parse::do_get_xxx():
6312
6313 // Compute address and memory type.
6314 int offset = field->offset_in_bytes();
6315 bool is_vol = field->is_volatile();
6316 ciType* field_klass = field->type();
6317 assert(field_klass->is_loaded(), "should be loaded");
6318 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6319 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6320 BasicType bt = field->layout_type();
6321
6322 // Build the resultant type of the load
6323 const Type *type;
6324 if (bt == T_OBJECT) {
6325 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6326 } else {
6327 type = Type::get_const_basic_type(bt);
6328 }
6329
6330 if (is_vol) {
6331 decorators |= MO_SEQ_CST;
6332 }
6333
6334 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6335 }
6336
6337 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6338 bool is_exact /* true */, bool is_static /* false */,
6339 ciInstanceKlass * fromKls /* NULL */) {
6340 if (fromKls == NULL) {
6341 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6342 assert(tinst != NULL, "obj is null");
6343 assert(tinst->is_loaded(), "obj is not loaded");
6344 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6345 fromKls = tinst->instance_klass();
6346 }
6347 else {
6348 assert(is_static, "only for static field access");
6349 }
6350 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6351 ciSymbol::make(fieldTypeString),
6352 is_static);
6353
6354 assert(field != NULL, "undefined field");
6355 assert(!field->is_volatile(), "not defined for volatile fields");
6356
6357 if (is_static) {
6358 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6359 fromObj = makecon(tip);
6360 }
6361
6362 // Next code copied from Parse::do_get_xxx():
6363
6364 // Compute address and memory type.
6365 int offset = field->offset_in_bytes();
6366 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6367
6368 return adr;
6369 }
6370
6371 //------------------------------inline_aescrypt_Block-----------------------
6372 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6373 address stubAddr = NULL;
6374 const char *stubName;
6375 assert(UseAES, "need AES instruction support");
6376
6377 switch(id) {
6378 case vmIntrinsics::_aescrypt_encryptBlock:
6379 stubAddr = StubRoutines::aescrypt_encryptBlock();
6380 stubName = "aescrypt_encryptBlock";
6381 break;
6382 case vmIntrinsics::_aescrypt_decryptBlock:
6383 stubAddr = StubRoutines::aescrypt_decryptBlock();
6384 stubName = "aescrypt_decryptBlock";
6385 break;
6386 default:
6387 break;
6388 }
6389 if (stubAddr == NULL) return false;
6390
6391 Node* aescrypt_object = argument(0);
6392 Node* src = argument(1);
6393 Node* src_offset = argument(2);
6394 Node* dest = argument(3);
6395 Node* dest_offset = argument(4);
6396
6397 src = must_be_not_null(src, true);
6398 dest = must_be_not_null(dest, true);
6399
6400 // (1) src and dest are arrays.
6401 const Type* src_type = src->Value(&_gvn);
6402 const Type* dest_type = dest->Value(&_gvn);
6403 const TypeAryPtr* top_src = src_type->isa_aryptr();
6404 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6405 assert (top_src != NULL && top_src->elem() != Type::BOTTOM && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6406
6407 // for the quick and dirty code we will skip all the checks.
6408 // we are just trying to get the call to be generated.
6409 Node* src_start = src;
6410 Node* dest_start = dest;
6411 if (src_offset != NULL || dest_offset != NULL) {
6412 assert(src_offset != NULL && dest_offset != NULL, "");
6413 src_start = array_element_address(src, src_offset, T_BYTE);
6414 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6415 }
6416
6417 // now need to get the start of its expanded key array
6418 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6419 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6420 if (k_start == NULL) return false;
6421
6422 // Call the stub.
6423 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6424 stubAddr, stubName, TypePtr::BOTTOM,
6425 src_start, dest_start, k_start);
6426
6427 return true;
6428 }
6429
6430 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6431 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6432 address stubAddr = NULL;
6433 const char *stubName = NULL;
6434
6435 assert(UseAES, "need AES instruction support");
6436
6437 switch(id) {
6438 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6439 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6440 stubName = "cipherBlockChaining_encryptAESCrypt";
6441 break;
6442 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6443 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6444 stubName = "cipherBlockChaining_decryptAESCrypt";
6445 break;
6446 default:
6447 break;
6448 }
6449 if (stubAddr == NULL) return false;
6450
6451 Node* cipherBlockChaining_object = argument(0);
6452 Node* src = argument(1);
6453 Node* src_offset = argument(2);
6454 Node* len = argument(3);
6455 Node* dest = argument(4);
6456 Node* dest_offset = argument(5);
6457
6458 src = must_be_not_null(src, false);
6459 dest = must_be_not_null(dest, false);
6460
6461 // (1) src and dest are arrays.
6462 const Type* src_type = src->Value(&_gvn);
6463 const Type* dest_type = dest->Value(&_gvn);
6464 const TypeAryPtr* top_src = src_type->isa_aryptr();
6465 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6466 assert (top_src != NULL && top_src->elem() != Type::BOTTOM
6467 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6468
6469 // checks are the responsibility of the caller
6470 Node* src_start = src;
6471 Node* dest_start = dest;
6472 if (src_offset != NULL || dest_offset != NULL) {
6473 assert(src_offset != NULL && dest_offset != NULL, "");
6474 src_start = array_element_address(src, src_offset, T_BYTE);
6475 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6476 }
6477
6478 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6479 // (because of the predicated logic executed earlier).
6480 // so we cast it here safely.
6481 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6482
6483 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6484 if (embeddedCipherObj == NULL) return false;
6485
6486 // cast it to what we know it will be at runtime
6487 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6488 assert(tinst != NULL, "CBC obj is null");
6489 assert(tinst->is_loaded(), "CBC obj is not loaded");
6490 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6491 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6492
6493 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6494 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6495 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6496 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6497 aescrypt_object = _gvn.transform(aescrypt_object);
6498
6499 // we need to get the start of the aescrypt_object's expanded key array
6500 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6501 if (k_start == NULL) return false;
6502
6503 // similarly, get the start address of the r vector
6504 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6505 if (objRvec == NULL) return false;
6506 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6507
6508 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6509 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6510 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6511 stubAddr, stubName, TypePtr::BOTTOM,
6512 src_start, dest_start, k_start, r_start, len);
6513
6514 // return cipher length (int)
6515 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6516 set_result(retvalue);
6517 return true;
6518 }
6519
6520 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6521 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6522 address stubAddr = NULL;
6523 const char *stubName = NULL;
6524
6525 assert(UseAES, "need AES instruction support");
6526
6527 switch (id) {
6528 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6529 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6530 stubName = "electronicCodeBook_encryptAESCrypt";
6531 break;
6532 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6533 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6534 stubName = "electronicCodeBook_decryptAESCrypt";
6535 break;
6536 default:
6537 break;
6538 }
6539
6540 if (stubAddr == NULL) return false;
6541
6542 Node* electronicCodeBook_object = argument(0);
6543 Node* src = argument(1);
6544 Node* src_offset = argument(2);
6545 Node* len = argument(3);
6546 Node* dest = argument(4);
6547 Node* dest_offset = argument(5);
6548
6549 // (1) src and dest are arrays.
6550 const Type* src_type = src->Value(&_gvn);
6551 const Type* dest_type = dest->Value(&_gvn);
6552 const TypeAryPtr* top_src = src_type->isa_aryptr();
6553 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6554 assert(top_src != NULL && top_src->elem() != Type::BOTTOM
6555 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6556
6557 // checks are the responsibility of the caller
6558 Node* src_start = src;
6559 Node* dest_start = dest;
6560 if (src_offset != NULL || dest_offset != NULL) {
6561 assert(src_offset != NULL && dest_offset != NULL, "");
6562 src_start = array_element_address(src, src_offset, T_BYTE);
6563 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6564 }
6565
6566 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6567 // (because of the predicated logic executed earlier).
6568 // so we cast it here safely.
6569 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6570
6571 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6572 if (embeddedCipherObj == NULL) return false;
6573
6574 // cast it to what we know it will be at runtime
6575 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6576 assert(tinst != NULL, "ECB obj is null");
6577 assert(tinst->is_loaded(), "ECB obj is not loaded");
6578 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6579 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6580
6581 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6582 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6583 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6584 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6585 aescrypt_object = _gvn.transform(aescrypt_object);
6586
6587 // we need to get the start of the aescrypt_object's expanded key array
6588 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6589 if (k_start == NULL) return false;
6590
6591 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6592 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6593 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6594 stubAddr, stubName, TypePtr::BOTTOM,
6595 src_start, dest_start, k_start, len);
6596
6597 // return cipher length (int)
6598 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6599 set_result(retvalue);
6600 return true;
6601 }
6602
6603 //------------------------------inline_counterMode_AESCrypt-----------------------
6604 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6605 assert(UseAES, "need AES instruction support");
6606 if (!UseAESCTRIntrinsics) return false;
6607
6608 address stubAddr = NULL;
6609 const char *stubName = NULL;
6610 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6611 stubAddr = StubRoutines::counterMode_AESCrypt();
6612 stubName = "counterMode_AESCrypt";
6613 }
6614 if (stubAddr == NULL) return false;
6615
6616 Node* counterMode_object = argument(0);
6617 Node* src = argument(1);
6618 Node* src_offset = argument(2);
6619 Node* len = argument(3);
6620 Node* dest = argument(4);
6621 Node* dest_offset = argument(5);
6622
6623 // (1) src and dest are arrays.
6624 const Type* src_type = src->Value(&_gvn);
6625 const Type* dest_type = dest->Value(&_gvn);
6626 const TypeAryPtr* top_src = src_type->isa_aryptr();
6627 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6628 assert(top_src != NULL && top_src->elem() != Type::BOTTOM &&
6629 top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6630
6631 // checks are the responsibility of the caller
6632 Node* src_start = src;
6633 Node* dest_start = dest;
6634 if (src_offset != NULL || dest_offset != NULL) {
6635 assert(src_offset != NULL && dest_offset != NULL, "");
6636 src_start = array_element_address(src, src_offset, T_BYTE);
6637 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6638 }
6639
6640 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6641 // (because of the predicated logic executed earlier).
6642 // so we cast it here safely.
6643 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6644 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6645 if (embeddedCipherObj == NULL) return false;
6646 // cast it to what we know it will be at runtime
6647 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6648 assert(tinst != NULL, "CTR obj is null");
6649 assert(tinst->is_loaded(), "CTR obj is not loaded");
6650 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6651 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6652 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6653 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6654 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6655 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6656 aescrypt_object = _gvn.transform(aescrypt_object);
6657 // we need to get the start of the aescrypt_object's expanded key array
6658 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6659 if (k_start == NULL) return false;
6660 // similarly, get the start address of the r vector
6661 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6662 if (obj_counter == NULL) return false;
6663 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6664
6665 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6666 if (saved_encCounter == NULL) return false;
6667 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6668 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6669
6670 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6671 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6672 OptoRuntime::counterMode_aescrypt_Type(),
6673 stubAddr, stubName, TypePtr::BOTTOM,
6674 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6675
6676 // return cipher length (int)
6677 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6678 set_result(retvalue);
6679 return true;
6680 }
6681
6682 //------------------------------get_key_start_from_aescrypt_object-----------------------
6683 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6684 #if defined(PPC64) || defined(S390)
6685 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6686 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6687 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6688 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6689 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6690 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6691 if (objSessionK == NULL) {
6692 return (Node *) NULL;
6693 }
6694 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6695 #else
6696 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6697 #endif // PPC64
6698 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6699 if (objAESCryptKey == NULL) return (Node *) NULL;
6700
6701 // now have the array, need to get the start address of the K array
6702 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6703 return k_start;
6704 }
6705
6706 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6707 // Return node representing slow path of predicate check.
6708 // the pseudo code we want to emulate with this predicate is:
6709 // for encryption:
6710 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6711 // for decryption:
6712 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6713 // note cipher==plain is more conservative than the original java code but that's OK
6714 //
6715 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6716 // The receiver was checked for NULL already.
6717 Node* objCBC = argument(0);
6718
6719 Node* src = argument(1);
6720 Node* dest = argument(4);
6721
6722 // Load embeddedCipher field of CipherBlockChaining object.
6723 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6724
6725 // get AESCrypt klass for instanceOf check
6726 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6727 // will have same classloader as CipherBlockChaining object
6728 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6729 assert(tinst != NULL, "CBCobj is null");
6730 assert(tinst->is_loaded(), "CBCobj is not loaded");
6731
6732 // we want to do an instanceof comparison against the AESCrypt class
6733 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6734 if (!klass_AESCrypt->is_loaded()) {
6735 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6736 Node* ctrl = control();
6737 set_control(top()); // no regular fast path
6738 return ctrl;
6739 }
6740
6741 src = must_be_not_null(src, true);
6742 dest = must_be_not_null(dest, true);
6743
6744 // Resolve oops to stable for CmpP below.
6745 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6746
6747 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6748 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6749 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6750
6751 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6752
6753 // for encryption, we are done
6754 if (!decrypting)
6755 return instof_false; // even if it is NULL
6756
6757 // for decryption, we need to add a further check to avoid
6758 // taking the intrinsic path when cipher and plain are the same
6759 // see the original java code for why.
6760 RegionNode* region = new RegionNode(3);
6761 region->init_req(1, instof_false);
6762
6763 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6764 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6765 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6766 region->init_req(2, src_dest_conjoint);
6767
6768 record_for_igvn(region);
6769 return _gvn.transform(region);
6770 }
6771
6772 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6773 // Return node representing slow path of predicate check.
6774 // the pseudo code we want to emulate with this predicate is:
6775 // for encryption:
6776 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6777 // for decryption:
6778 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6779 // note cipher==plain is more conservative than the original java code but that's OK
6780 //
6781 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6782 // The receiver was checked for NULL already.
6783 Node* objECB = argument(0);
6784
6785 // Load embeddedCipher field of ElectronicCodeBook object.
6786 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6787
6788 // get AESCrypt klass for instanceOf check
6789 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6790 // will have same classloader as ElectronicCodeBook object
6791 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6792 assert(tinst != NULL, "ECBobj is null");
6793 assert(tinst->is_loaded(), "ECBobj is not loaded");
6794
6795 // we want to do an instanceof comparison against the AESCrypt class
6796 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6797 if (!klass_AESCrypt->is_loaded()) {
6798 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6799 Node* ctrl = control();
6800 set_control(top()); // no regular fast path
6801 return ctrl;
6802 }
6803 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6804
6805 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6806 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6807 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6808
6809 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6810
6811 // for encryption, we are done
6812 if (!decrypting)
6813 return instof_false; // even if it is NULL
6814
6815 // for decryption, we need to add a further check to avoid
6816 // taking the intrinsic path when cipher and plain are the same
6817 // see the original java code for why.
6818 RegionNode* region = new RegionNode(3);
6819 region->init_req(1, instof_false);
6820 Node* src = argument(1);
6821 Node* dest = argument(4);
6822 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6823 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6824 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6825 region->init_req(2, src_dest_conjoint);
6826
6827 record_for_igvn(region);
6828 return _gvn.transform(region);
6829 }
6830
6831 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6832 // Return node representing slow path of predicate check.
6833 // the pseudo code we want to emulate with this predicate is:
6834 // for encryption:
6835 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6836 // for decryption:
6837 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6838 // note cipher==plain is more conservative than the original java code but that's OK
6839 //
6840
6841 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6842 // The receiver was checked for NULL already.
6843 Node* objCTR = argument(0);
6844
6845 // Load embeddedCipher field of CipherBlockChaining object.
6846 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6847
6848 // get AESCrypt klass for instanceOf check
6849 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6850 // will have same classloader as CipherBlockChaining object
6851 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6852 assert(tinst != NULL, "CTRobj is null");
6853 assert(tinst->is_loaded(), "CTRobj is not loaded");
6854
6855 // we want to do an instanceof comparison against the AESCrypt class
6856 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6857 if (!klass_AESCrypt->is_loaded()) {
6858 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6859 Node* ctrl = control();
6860 set_control(top()); // no regular fast path
6861 return ctrl;
6862 }
6863
6864 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6865 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6866 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6867 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6868 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6869
6870 return instof_false; // even if it is NULL
6871 }
6872
6873 //------------------------------inline_ghash_processBlocks
6874 bool LibraryCallKit::inline_ghash_processBlocks() {
6875 address stubAddr;
6876 const char *stubName;
6877 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6878
6879 stubAddr = StubRoutines::ghash_processBlocks();
6880 stubName = "ghash_processBlocks";
6881
6882 Node* data = argument(0);
6883 Node* offset = argument(1);
6884 Node* len = argument(2);
6885 Node* state = argument(3);
6886 Node* subkeyH = argument(4);
6887
6888 state = must_be_not_null(state, true);
6889 subkeyH = must_be_not_null(subkeyH, true);
6890 data = must_be_not_null(data, true);
6891
6892 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6893 assert(state_start, "state is NULL");
6894 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6895 assert(subkeyH_start, "subkeyH is NULL");
6896 Node* data_start = array_element_address(data, offset, T_BYTE);
6897 assert(data_start, "data is NULL");
6898
6899 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6900 OptoRuntime::ghash_processBlocks_Type(),
6901 stubAddr, stubName, TypePtr::BOTTOM,
6902 state_start, subkeyH_start, data_start, len);
6903 return true;
6904 }
6905
6906 //------------------------------inline_chacha20Block
6907 bool LibraryCallKit::inline_chacha20Block() {
6908 address stubAddr;
6909 const char *stubName;
6910 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
6911
6912 stubAddr = StubRoutines::chacha20Block();
6913 stubName = "chacha20Block";
6914
6915 Node* state = argument(0);
6916 Node* result = argument(1);
6917
6918 state = must_be_not_null(state, true);
6919 result = must_be_not_null(result, true);
6920
6921 Node* state_start = array_element_address(state, intcon(0), T_INT);
6922 assert(state_start, "state is NULL");
6923 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
6924 assert(result_start, "result is NULL");
6925
6926 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
6927 OptoRuntime::chacha20Block_Type(),
6928 stubAddr, stubName, TypePtr::BOTTOM,
6929 state_start, result_start);
6930 // return key stream length (int)
6931 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
6932 set_result(retvalue);
6933 return true;
6934 }
6935
6936 bool LibraryCallKit::inline_base64_encodeBlock() {
6937 address stubAddr;
6938 const char *stubName;
6939 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6940 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6941 stubAddr = StubRoutines::base64_encodeBlock();
6942 stubName = "encodeBlock";
6943
6944 if (!stubAddr) return false;
6945 Node* base64obj = argument(0);
6946 Node* src = argument(1);
6947 Node* offset = argument(2);
6948 Node* len = argument(3);
6949 Node* dest = argument(4);
6950 Node* dp = argument(5);
6951 Node* isURL = argument(6);
6952
6953 src = must_be_not_null(src, true);
6954 dest = must_be_not_null(dest, true);
6955
6956 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6957 assert(src_start, "source array is NULL");
6958 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6959 assert(dest_start, "destination array is NULL");
6960
6961 Node* base64 = make_runtime_call(RC_LEAF,
6962 OptoRuntime::base64_encodeBlock_Type(),
6963 stubAddr, stubName, TypePtr::BOTTOM,
6964 src_start, offset, len, dest_start, dp, isURL);
6965 return true;
6966 }
6967
6968 bool LibraryCallKit::inline_base64_decodeBlock() {
6969 address stubAddr;
6970 const char *stubName;
6971 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6972 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6973 stubAddr = StubRoutines::base64_decodeBlock();
6974 stubName = "decodeBlock";
6975
6976 if (!stubAddr) return false;
6977 Node* base64obj = argument(0);
6978 Node* src = argument(1);
6979 Node* src_offset = argument(2);
6980 Node* len = argument(3);
6981 Node* dest = argument(4);
6982 Node* dest_offset = argument(5);
6983 Node* isURL = argument(6);
6984 Node* isMIME = argument(7);
6985
6986 src = must_be_not_null(src, true);
6987 dest = must_be_not_null(dest, true);
6988
6989 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6990 assert(src_start, "source array is NULL");
6991 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6992 assert(dest_start, "destination array is NULL");
6993
6994 Node* call = make_runtime_call(RC_LEAF,
6995 OptoRuntime::base64_decodeBlock_Type(),
6996 stubAddr, stubName, TypePtr::BOTTOM,
6997 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6998 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6999 set_result(result);
7000 return true;
7001 }
7002
7003 bool LibraryCallKit::inline_poly1305_processBlocks() {
7004 address stubAddr;
7005 const char *stubName;
7006 assert(UsePolyIntrinsics, "need Poly intrinsics support");
7007 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7008 stubAddr = StubRoutines::poly1305_processBlocks();
7009 stubName = "poly1305_processBlocks";
7010
7011 if (!stubAddr) return false;
7012 null_check_receiver(); // null-check receiver
7013 if (stopped()) return true;
7014
7015 Node* input = argument(1);
7016 Node* input_offset = argument(2);
7017 Node* len = argument(3);
7018 Node* alimbs = argument(4);
7019 Node* rlimbs = argument(5);
7020
7021 input = must_be_not_null(input, true);
7022 alimbs = must_be_not_null(alimbs, true);
7023 rlimbs = must_be_not_null(rlimbs, true);
7024
7025 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7026 assert(input_start, "input array is NULL");
7027 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7028 assert(acc_start, "acc array is NULL");
7029 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7030 assert(r_start, "r array is NULL");
7031
7032 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7033 OptoRuntime::poly1305_processBlocks_Type(),
7034 stubAddr, stubName, TypePtr::BOTTOM,
7035 input_start, len, acc_start, r_start);
7036 return true;
7037 }
7038
7039 //------------------------------inline_digestBase_implCompress-----------------------
7040 //
7041 // Calculate MD5 for single-block byte[] array.
7042 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7043 //
7044 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7045 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7046 //
7047 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7048 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7049 //
7050 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7051 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7052 //
7053 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7054 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7055 //
7056 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7057 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7058
7059 Node* digestBase_obj = argument(0);
7060 Node* src = argument(1); // type oop
7061 Node* ofs = argument(2); // type int
7062
7063 const Type* src_type = src->Value(&_gvn);
7064 const TypeAryPtr* top_src = src_type->isa_aryptr();
7065 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7066 // failed array check
7067 return false;
7068 }
7069 // Figure out the size and type of the elements we will be copying.
7070 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7071 if (src_elem != T_BYTE) {
7072 return false;
7073 }
7074 // 'src_start' points to src array + offset
7075 src = must_be_not_null(src, true);
7076 Node* src_start = array_element_address(src, ofs, src_elem);
7077 Node* state = NULL;
7078 Node* block_size = NULL;
7079 address stubAddr;
7080 const char *stubName;
7081
7082 switch(id) {
7083 case vmIntrinsics::_md5_implCompress:
7084 assert(UseMD5Intrinsics, "need MD5 instruction support");
7085 state = get_state_from_digest_object(digestBase_obj, T_INT);
7086 stubAddr = StubRoutines::md5_implCompress();
7087 stubName = "md5_implCompress";
7088 break;
7089 case vmIntrinsics::_sha_implCompress:
7090 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7091 state = get_state_from_digest_object(digestBase_obj, T_INT);
7092 stubAddr = StubRoutines::sha1_implCompress();
7093 stubName = "sha1_implCompress";
7094 break;
7095 case vmIntrinsics::_sha2_implCompress:
7096 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7097 state = get_state_from_digest_object(digestBase_obj, T_INT);
7098 stubAddr = StubRoutines::sha256_implCompress();
7099 stubName = "sha256_implCompress";
7100 break;
7101 case vmIntrinsics::_sha5_implCompress:
7102 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7103 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7104 stubAddr = StubRoutines::sha512_implCompress();
7105 stubName = "sha512_implCompress";
7106 break;
7107 case vmIntrinsics::_sha3_implCompress:
7108 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7109 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7110 stubAddr = StubRoutines::sha3_implCompress();
7111 stubName = "sha3_implCompress";
7112 block_size = get_block_size_from_digest_object(digestBase_obj);
7113 if (block_size == NULL) return false;
7114 break;
7115 default:
7116 fatal_unexpected_iid(id);
7117 return false;
7118 }
7119 if (state == NULL) return false;
7120
7121 assert(stubAddr != NULL, "Stub is generated");
7122 if (stubAddr == NULL) return false;
7123
7124 // Call the stub.
7125 Node* call;
7126 if (block_size == NULL) {
7127 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7128 stubAddr, stubName, TypePtr::BOTTOM,
7129 src_start, state);
7130 } else {
7131 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7132 stubAddr, stubName, TypePtr::BOTTOM,
7133 src_start, state, block_size);
7134 }
7135
7136 return true;
7137 }
7138
7139 //------------------------------inline_digestBase_implCompressMB-----------------------
7140 //
7141 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7142 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7143 //
7144 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7145 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7146 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7147 assert((uint)predicate < 5, "sanity");
7148 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7149
7150 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7151 Node* src = argument(1); // byte[] array
7152 Node* ofs = argument(2); // type int
7153 Node* limit = argument(3); // type int
7154
7155 const Type* src_type = src->Value(&_gvn);
7156 const TypeAryPtr* top_src = src_type->isa_aryptr();
7157 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7158 // failed array check
7159 return false;
7160 }
7161 // Figure out the size and type of the elements we will be copying.
7162 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7163 if (src_elem != T_BYTE) {
7164 return false;
7165 }
7166 // 'src_start' points to src array + offset
7167 src = must_be_not_null(src, false);
7168 Node* src_start = array_element_address(src, ofs, src_elem);
7169
7170 const char* klass_digestBase_name = NULL;
7171 const char* stub_name = NULL;
7172 address stub_addr = NULL;
7173 BasicType elem_type = T_INT;
7174
7175 switch (predicate) {
7176 case 0:
7177 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7178 klass_digestBase_name = "sun/security/provider/MD5";
7179 stub_name = "md5_implCompressMB";
7180 stub_addr = StubRoutines::md5_implCompressMB();
7181 }
7182 break;
7183 case 1:
7184 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7185 klass_digestBase_name = "sun/security/provider/SHA";
7186 stub_name = "sha1_implCompressMB";
7187 stub_addr = StubRoutines::sha1_implCompressMB();
7188 }
7189 break;
7190 case 2:
7191 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7192 klass_digestBase_name = "sun/security/provider/SHA2";
7193 stub_name = "sha256_implCompressMB";
7194 stub_addr = StubRoutines::sha256_implCompressMB();
7195 }
7196 break;
7197 case 3:
7198 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7199 klass_digestBase_name = "sun/security/provider/SHA5";
7200 stub_name = "sha512_implCompressMB";
7201 stub_addr = StubRoutines::sha512_implCompressMB();
7202 elem_type = T_LONG;
7203 }
7204 break;
7205 case 4:
7206 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7207 klass_digestBase_name = "sun/security/provider/SHA3";
7208 stub_name = "sha3_implCompressMB";
7209 stub_addr = StubRoutines::sha3_implCompressMB();
7210 elem_type = T_BYTE;
7211 }
7212 break;
7213 default:
7214 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7215 }
7216 if (klass_digestBase_name != NULL) {
7217 assert(stub_addr != NULL, "Stub is generated");
7218 if (stub_addr == NULL) return false;
7219
7220 // get DigestBase klass to lookup for SHA klass
7221 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7222 assert(tinst != NULL, "digestBase_obj is not instance???");
7223 assert(tinst->is_loaded(), "DigestBase is not loaded");
7224
7225 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7226 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7227 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7228 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7229 }
7230 return false;
7231 }
7232
7233 //------------------------------inline_digestBase_implCompressMB-----------------------
7234 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7235 BasicType elem_type, address stubAddr, const char *stubName,
7236 Node* src_start, Node* ofs, Node* limit) {
7237 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7238 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7239 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7240 digest_obj = _gvn.transform(digest_obj);
7241
7242 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7243 if (state == NULL) return false;
7244
7245 Node* block_size = NULL;
7246 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7247 block_size = get_block_size_from_digest_object(digest_obj);
7248 if (block_size == NULL) return false;
7249 }
7250
7251 // Call the stub.
7252 Node* call;
7253 if (block_size == NULL) {
7254 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7255 OptoRuntime::digestBase_implCompressMB_Type(false),
7256 stubAddr, stubName, TypePtr::BOTTOM,
7257 src_start, state, ofs, limit);
7258 } else {
7259 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7260 OptoRuntime::digestBase_implCompressMB_Type(true),
7261 stubAddr, stubName, TypePtr::BOTTOM,
7262 src_start, state, block_size, ofs, limit);
7263 }
7264
7265 // return ofs (int)
7266 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7267 set_result(result);
7268
7269 return true;
7270 }
7271
7272 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7273 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7274 assert(UseAES, "need AES instruction support");
7275 address stubAddr = NULL;
7276 const char *stubName = NULL;
7277 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7278 stubName = "galoisCounterMode_AESCrypt";
7279
7280 if (stubAddr == NULL) return false;
7281
7282 Node* in = argument(0);
7283 Node* inOfs = argument(1);
7284 Node* len = argument(2);
7285 Node* ct = argument(3);
7286 Node* ctOfs = argument(4);
7287 Node* out = argument(5);
7288 Node* outOfs = argument(6);
7289 Node* gctr_object = argument(7);
7290 Node* ghash_object = argument(8);
7291
7292 // (1) in, ct and out are arrays.
7293 const Type* in_type = in->Value(&_gvn);
7294 const Type* ct_type = ct->Value(&_gvn);
7295 const Type* out_type = out->Value(&_gvn);
7296 const TypeAryPtr* top_in = in_type->isa_aryptr();
7297 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7298 const TypeAryPtr* top_out = out_type->isa_aryptr();
7299 assert(top_in != NULL && top_in->elem() != Type::BOTTOM &&
7300 top_ct != NULL && top_ct->elem() != Type::BOTTOM &&
7301 top_out != NULL && top_out->elem() != Type::BOTTOM, "args are strange");
7302
7303 // checks are the responsibility of the caller
7304 Node* in_start = in;
7305 Node* ct_start = ct;
7306 Node* out_start = out;
7307 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7308 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7309 in_start = array_element_address(in, inOfs, T_BYTE);
7310 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7311 out_start = array_element_address(out, outOfs, T_BYTE);
7312 }
7313
7314 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7315 // (because of the predicated logic executed earlier).
7316 // so we cast it here safely.
7317 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7318 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7319 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7320 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7321 Node* state = load_field_from_object(ghash_object, "state", "[J");
7322
7323 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7324 return false;
7325 }
7326 // cast it to what we know it will be at runtime
7327 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7328 assert(tinst != NULL, "GCTR obj is null");
7329 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7330 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7331 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7332 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7333 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7334 const TypeOopPtr* xtype = aklass->as_instance_type();
7335 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7336 aescrypt_object = _gvn.transform(aescrypt_object);
7337 // we need to get the start of the aescrypt_object's expanded key array
7338 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7339 if (k_start == NULL) return false;
7340 // similarly, get the start address of the r vector
7341 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7342 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7343 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7344
7345
7346 // Call the stub, passing params
7347 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7348 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7349 stubAddr, stubName, TypePtr::BOTTOM,
7350 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7351
7352 // return cipher length (int)
7353 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7354 set_result(retvalue);
7355
7356 return true;
7357 }
7358
7359 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7360 // Return node representing slow path of predicate check.
7361 // the pseudo code we want to emulate with this predicate is:
7362 // for encryption:
7363 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7364 // for decryption:
7365 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7366 // note cipher==plain is more conservative than the original java code but that's OK
7367 //
7368
7369 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7370 // The receiver was checked for NULL already.
7371 Node* objGCTR = argument(7);
7372 // Load embeddedCipher field of GCTR object.
7373 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7374 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7375
7376 // get AESCrypt klass for instanceOf check
7377 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7378 // will have same classloader as CipherBlockChaining object
7379 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7380 assert(tinst != NULL, "GCTR obj is null");
7381 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7382
7383 // we want to do an instanceof comparison against the AESCrypt class
7384 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7385 if (!klass_AESCrypt->is_loaded()) {
7386 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7387 Node* ctrl = control();
7388 set_control(top()); // no regular fast path
7389 return ctrl;
7390 }
7391
7392 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7393 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7394 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7395 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7396 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7397
7398 return instof_false; // even if it is NULL
7399 }
7400
7401 //------------------------------get_state_from_digest_object-----------------------
7402 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7403 const char* state_type;
7404 switch (elem_type) {
7405 case T_BYTE: state_type = "[B"; break;
7406 case T_INT: state_type = "[I"; break;
7407 case T_LONG: state_type = "[J"; break;
7408 default: ShouldNotReachHere();
7409 }
7410 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7411 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7412 if (digest_state == NULL) return (Node *) NULL;
7413
7414 // now have the array, need to get the start address of the state array
7415 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7416 return state;
7417 }
7418
7419 //------------------------------get_block_size_from_sha3_object----------------------------------
7420 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
7421 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
7422 assert (block_size != NULL, "sanity");
7423 return block_size;
7424 }
7425
7426 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7427 // Return node representing slow path of predicate check.
7428 // the pseudo code we want to emulate with this predicate is:
7429 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7430 //
7431 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7432 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7433 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7434 assert((uint)predicate < 5, "sanity");
7435
7436 // The receiver was checked for NULL already.
7437 Node* digestBaseObj = argument(0);
7438
7439 // get DigestBase klass for instanceOf check
7440 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7441 assert(tinst != NULL, "digestBaseObj is null");
7442 assert(tinst->is_loaded(), "DigestBase is not loaded");
7443
7444 const char* klass_name = NULL;
7445 switch (predicate) {
7446 case 0:
7447 if (UseMD5Intrinsics) {
7448 // we want to do an instanceof comparison against the MD5 class
7449 klass_name = "sun/security/provider/MD5";
7450 }
7451 break;
7452 case 1:
7453 if (UseSHA1Intrinsics) {
7454 // we want to do an instanceof comparison against the SHA class
7455 klass_name = "sun/security/provider/SHA";
7456 }
7457 break;
7458 case 2:
7459 if (UseSHA256Intrinsics) {
7460 // we want to do an instanceof comparison against the SHA2 class
7461 klass_name = "sun/security/provider/SHA2";
7462 }
7463 break;
7464 case 3:
7465 if (UseSHA512Intrinsics) {
7466 // we want to do an instanceof comparison against the SHA5 class
7467 klass_name = "sun/security/provider/SHA5";
7468 }
7469 break;
7470 case 4:
7471 if (UseSHA3Intrinsics) {
7472 // we want to do an instanceof comparison against the SHA3 class
7473 klass_name = "sun/security/provider/SHA3";
7474 }
7475 break;
7476 default:
7477 fatal("unknown SHA intrinsic predicate: %d", predicate);
7478 }
7479
7480 ciKlass* klass = NULL;
7481 if (klass_name != NULL) {
7482 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7483 }
7484 if ((klass == NULL) || !klass->is_loaded()) {
7485 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7486 Node* ctrl = control();
7487 set_control(top()); // no intrinsic path
7488 return ctrl;
7489 }
7490 ciInstanceKlass* instklass = klass->as_instance_klass();
7491
7492 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7493 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7494 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7495 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7496
7497 return instof_false; // even if it is NULL
7498 }
7499
7500 //-------------inline_fma-----------------------------------
7501 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7502 Node *a = NULL;
7503 Node *b = NULL;
7504 Node *c = NULL;
7505 Node* result = NULL;
7506 switch (id) {
7507 case vmIntrinsics::_fmaD:
7508 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7509 // no receiver since it is static method
7510 a = round_double_node(argument(0));
7511 b = round_double_node(argument(2));
7512 c = round_double_node(argument(4));
7513 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7514 break;
7515 case vmIntrinsics::_fmaF:
7516 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7517 a = argument(0);
7518 b = argument(1);
7519 c = argument(2);
7520 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7521 break;
7522 default:
7523 fatal_unexpected_iid(id); break;
7524 }
7525 set_result(result);
7526 return true;
7527 }
7528
7529 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7530 // argument(0) is receiver
7531 Node* codePoint = argument(1);
7532 Node* n = NULL;
7533
7534 switch (id) {
7535 case vmIntrinsics::_isDigit :
7536 n = new DigitNode(control(), codePoint);
7537 break;
7538 case vmIntrinsics::_isLowerCase :
7539 n = new LowerCaseNode(control(), codePoint);
7540 break;
7541 case vmIntrinsics::_isUpperCase :
7542 n = new UpperCaseNode(control(), codePoint);
7543 break;
7544 case vmIntrinsics::_isWhitespace :
7545 n = new WhitespaceNode(control(), codePoint);
7546 break;
7547 default:
7548 fatal_unexpected_iid(id);
7549 }
7550
7551 set_result(_gvn.transform(n));
7552 return true;
7553 }
7554
7555 //------------------------------inline_fp_min_max------------------------------
7556 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7557 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7558
7559 // The intrinsic should be used only when the API branches aren't predictable,
7560 // the last one performing the most important comparison. The following heuristic
7561 // uses the branch statistics to eventually bail out if necessary.
7562
7563 ciMethodData *md = callee()->method_data();
7564
7565 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7566 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7567
7568 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7569 // Bail out if the call-site didn't contribute enough to the statistics.
7570 return false;
7571 }
7572
7573 uint taken = 0, not_taken = 0;
7574
7575 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7576 if (p->is_BranchData()) {
7577 taken = ((ciBranchData*)p)->taken();
7578 not_taken = ((ciBranchData*)p)->not_taken();
7579 }
7580 }
7581
7582 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7583 balance = balance < 0 ? -balance : balance;
7584 if ( balance > 0.2 ) {
7585 // Bail out if the most important branch is predictable enough.
7586 return false;
7587 }
7588 }
7589 */
7590
7591 Node *a = NULL;
7592 Node *b = NULL;
7593 Node *n = NULL;
7594 switch (id) {
7595 case vmIntrinsics::_maxF:
7596 case vmIntrinsics::_minF:
7597 case vmIntrinsics::_maxF_strict:
7598 case vmIntrinsics::_minF_strict:
7599 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7600 a = argument(0);
7601 b = argument(1);
7602 break;
7603 case vmIntrinsics::_maxD:
7604 case vmIntrinsics::_minD:
7605 case vmIntrinsics::_maxD_strict:
7606 case vmIntrinsics::_minD_strict:
7607 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7608 a = round_double_node(argument(0));
7609 b = round_double_node(argument(2));
7610 break;
7611 default:
7612 fatal_unexpected_iid(id);
7613 break;
7614 }
7615 switch (id) {
7616 case vmIntrinsics::_maxF:
7617 case vmIntrinsics::_maxF_strict:
7618 n = new MaxFNode(a, b);
7619 break;
7620 case vmIntrinsics::_minF:
7621 case vmIntrinsics::_minF_strict:
7622 n = new MinFNode(a, b);
7623 break;
7624 case vmIntrinsics::_maxD:
7625 case vmIntrinsics::_maxD_strict:
7626 n = new MaxDNode(a, b);
7627 break;
7628 case vmIntrinsics::_minD:
7629 case vmIntrinsics::_minD_strict:
7630 n = new MinDNode(a, b);
7631 break;
7632 default:
7633 fatal_unexpected_iid(id);
7634 break;
7635 }
7636 set_result(_gvn.transform(n));
7637 return true;
7638 }
7639
7640 bool LibraryCallKit::inline_profileBoolean() {
7641 Node* counts = argument(1);
7642 const TypeAryPtr* ary = NULL;
7643 ciArray* aobj = NULL;
7644 if (counts->is_Con()
7645 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7646 && (aobj = ary->const_oop()->as_array()) != NULL
7647 && (aobj->length() == 2)) {
7648 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7649 jint false_cnt = aobj->element_value(0).as_int();
7650 jint true_cnt = aobj->element_value(1).as_int();
7651
7652 if (C->log() != NULL) {
7653 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7654 false_cnt, true_cnt);
7655 }
7656
7657 if (false_cnt + true_cnt == 0) {
7658 // According to profile, never executed.
7659 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7660 Deoptimization::Action_reinterpret);
7661 return true;
7662 }
7663
7664 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7665 // is a number of each value occurrences.
7666 Node* result = argument(0);
7667 if (false_cnt == 0 || true_cnt == 0) {
7668 // According to profile, one value has been never seen.
7669 int expected_val = (false_cnt == 0) ? 1 : 0;
7670
7671 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7672 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7673
7674 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7675 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7676 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7677
7678 { // Slow path: uncommon trap for never seen value and then reexecute
7679 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7680 // the value has been seen at least once.
7681 PreserveJVMState pjvms(this);
7682 PreserveReexecuteState preexecs(this);
7683 jvms()->set_should_reexecute(true);
7684
7685 set_control(slow_path);
7686 set_i_o(i_o());
7687
7688 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7689 Deoptimization::Action_reinterpret);
7690 }
7691 // The guard for never seen value enables sharpening of the result and
7692 // returning a constant. It allows to eliminate branches on the same value
7693 // later on.
7694 set_control(fast_path);
7695 result = intcon(expected_val);
7696 }
7697 // Stop profiling.
7698 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7699 // By replacing method body with profile data (represented as ProfileBooleanNode
7700 // on IR level) we effectively disable profiling.
7701 // It enables full speed execution once optimized code is generated.
7702 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7703 C->record_for_igvn(profile);
7704 set_result(profile);
7705 return true;
7706 } else {
7707 // Continue profiling.
7708 // Profile data isn't available at the moment. So, execute method's bytecode version.
7709 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7710 // is compiled and counters aren't available since corresponding MethodHandle
7711 // isn't a compile-time constant.
7712 return false;
7713 }
7714 }
7715
7716 bool LibraryCallKit::inline_isCompileConstant() {
7717 Node* n = argument(0);
7718 set_result(n->is_Con() ? intcon(1) : intcon(0));
7719 return true;
7720 }
7721
7722 //------------------------------- inline_getObjectSize --------------------------------------
7723 //
7724 // Calculate the runtime size of the object/array.
7725 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7726 //
7727 bool LibraryCallKit::inline_getObjectSize() {
7728 Node* obj = argument(3);
7729 Node* klass_node = load_object_klass(obj);
7730
7731 jint layout_con = Klass::_lh_neutral_value;
7732 Node* layout_val = get_layout_helper(klass_node, layout_con);
7733 int layout_is_con = (layout_val == NULL);
7734
7735 if (layout_is_con) {
7736 // Layout helper is constant, can figure out things at compile time.
7737
7738 if (Klass::layout_helper_is_instance(layout_con)) {
7739 // Instance case: layout_con contains the size itself.
7740 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7741 set_result(size);
7742 } else {
7743 // Array case: size is round(header + element_size*arraylength).
7744 // Since arraylength is different for every array instance, we have to
7745 // compute the whole thing at runtime.
7746
7747 Node* arr_length = load_array_length(obj);
7748
7749 int round_mask = MinObjAlignmentInBytes - 1;
7750 int hsize = Klass::layout_helper_header_size(layout_con);
7751 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7752
7753 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7754 round_mask = 0; // strength-reduce it if it goes away completely
7755 }
7756 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7757 Node* header_size = intcon(hsize + round_mask);
7758
7759 Node* lengthx = ConvI2X(arr_length);
7760 Node* headerx = ConvI2X(header_size);
7761
7762 Node* abody = lengthx;
7763 if (eshift != 0) {
7764 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7765 }
7766 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7767 if (round_mask != 0) {
7768 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7769 }
7770 size = ConvX2L(size);
7771 set_result(size);
7772 }
7773 } else {
7774 // Layout helper is not constant, need to test for array-ness at runtime.
7775
7776 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7777 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7778 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7779 record_for_igvn(result_reg);
7780
7781 Node* array_ctl = generate_array_guard(klass_node, NULL);
7782 if (array_ctl != NULL) {
7783 // Array case: size is round(header + element_size*arraylength).
7784 // Since arraylength is different for every array instance, we have to
7785 // compute the whole thing at runtime.
7786
7787 PreserveJVMState pjvms(this);
7788 set_control(array_ctl);
7789 Node* arr_length = load_array_length(obj);
7790
7791 int round_mask = MinObjAlignmentInBytes - 1;
7792 Node* mask = intcon(round_mask);
7793
7794 Node* hss = intcon(Klass::_lh_header_size_shift);
7795 Node* hsm = intcon(Klass::_lh_header_size_mask);
7796 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7797 header_size = _gvn.transform(new AndINode(header_size, hsm));
7798 header_size = _gvn.transform(new AddINode(header_size, mask));
7799
7800 // There is no need to mask or shift this value.
7801 // The semantics of LShiftINode include an implicit mask to 0x1F.
7802 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7803 Node* elem_shift = layout_val;
7804
7805 Node* lengthx = ConvI2X(arr_length);
7806 Node* headerx = ConvI2X(header_size);
7807
7808 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7809 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7810 if (round_mask != 0) {
7811 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7812 }
7813 size = ConvX2L(size);
7814
7815 result_reg->init_req(_array_path, control());
7816 result_val->init_req(_array_path, size);
7817 }
7818
7819 if (!stopped()) {
7820 // Instance case: the layout helper gives us instance size almost directly,
7821 // but we need to mask out the _lh_instance_slow_path_bit.
7822 Node* size = ConvI2X(layout_val);
7823 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7824 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7825 size = _gvn.transform(new AndXNode(size, mask));
7826 size = ConvX2L(size);
7827
7828 result_reg->init_req(_instance_path, control());
7829 result_val->init_req(_instance_path, size);
7830 }
7831
7832 set_result(result_reg, result_val);
7833 }
7834
7835 return true;
7836 }
7837
7838 //------------------------------- inline_blackhole --------------------------------------
7839 //
7840 // Make sure all arguments to this node are alive.
7841 // This matches methods that were requested to be blackholed through compile commands.
7842 //
7843 bool LibraryCallKit::inline_blackhole() {
7844 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7845 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7846 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7847
7848 // Bind call arguments as blackhole arguments to keep them alive
7849 Node* bh = insert_mem_bar(Op_Blackhole);
7850 uint nargs = callee()->arg_size();
7851 for (uint i = 0; i < nargs; i++) {
7852 bh->add_req(argument(i));
7853 }
7854
7855 return true;
7856 }