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/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/unsafe.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/stubRoutines.hpp"
58 #include "utilities/macros.hpp"
59 #include "utilities/powerOfTwo.hpp"
60
61 //---------------------------make_vm_intrinsic----------------------------
62 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
63 vmIntrinsicID id = m->intrinsic_id();
64 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
65
66 if (!m->is_loaded()) {
67 // Do not attempt to inline unloaded methods.
68 return NULL;
69 }
70
71 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
72 bool is_available = false;
73
74 {
75 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
76 // the compiler must transition to '_thread_in_vm' state because both
77 // methods access VM-internal data.
78 VM_ENTRY_MARK;
79 methodHandle mh(THREAD, m->get_Method());
80 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
81 !C->directive()->is_intrinsic_disabled(mh) &&
82 !vmIntrinsics::is_disabled_by_flags(mh);
83
84 }
85
86 if (is_available) {
87 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
88 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
89 return new LibraryIntrinsic(m, is_virtual,
90 vmIntrinsics::predicates_needed(id),
91 vmIntrinsics::does_virtual_dispatch(id),
92 id);
93 } else {
94 return NULL;
95 }
96 }
97
98 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
99 LibraryCallKit kit(jvms, this);
100 Compile* C = kit.C;
101 int nodes = C->unique();
102 #ifndef PRODUCT
103 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
104 char buf[1000];
105 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
106 tty->print_cr("Intrinsic %s", str);
107 }
108 #endif
109 ciMethod* callee = kit.callee();
110 const int bci = kit.bci();
111 #ifdef ASSERT
112 Node* ctrl = kit.control();
113 #endif
114 // Try to inline the intrinsic.
115 if (callee->check_intrinsic_candidate() &&
116 kit.try_to_inline(_last_predicate)) {
117 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
118 : "(intrinsic)";
119 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
120 if (C->print_intrinsics() || C->print_inlining()) {
121 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
122 }
123 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
124 if (C->log()) {
125 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
126 vmIntrinsics::name_at(intrinsic_id()),
127 (is_virtual() ? " virtual='1'" : ""),
128 C->unique() - nodes);
129 }
130 // Push the result from the inlined method onto the stack.
131 kit.push_result();
132 C->print_inlining_update(this);
133 return kit.transfer_exceptions_into_jvms();
134 }
135
136 // The intrinsic bailed out
137 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
138 if (jvms->has_method()) {
139 // Not a root compile.
140 const char* msg;
141 if (callee->intrinsic_candidate()) {
142 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
143 } else {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
145 : "failed to inline (intrinsic), method not annotated";
146 }
147 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
148 if (C->print_intrinsics() || C->print_inlining()) {
149 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
150 }
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165 C->print_inlining_update(this);
166
167 return NULL;
168 }
169
170 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
171 LibraryCallKit kit(jvms, this);
172 Compile* C = kit.C;
173 int nodes = C->unique();
174 _last_predicate = predicate;
175 #ifndef PRODUCT
176 assert(is_predicated() && predicate < predicates_count(), "sanity");
177 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
178 char buf[1000];
179 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
180 tty->print_cr("Predicate for intrinsic %s", str);
181 }
182 #endif
183 ciMethod* callee = kit.callee();
184 const int bci = kit.bci();
185
186 Node* slow_ctl = kit.try_to_predicate(predicate);
187 if (!kit.failing()) {
188 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
189 : "(intrinsic, predicate)";
190 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
191 if (C->print_intrinsics() || C->print_inlining()) {
192 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
193 }
194 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
195 if (C->log()) {
196 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
197 vmIntrinsics::name_at(intrinsic_id()),
198 (is_virtual() ? " virtual='1'" : ""),
199 C->unique() - nodes);
200 }
201 return slow_ctl; // Could be NULL if the check folds.
202 }
203
204 // The intrinsic bailed out
205 if (jvms->has_method()) {
206 // Not a root compile.
207 const char* msg = "failed to generate predicate for intrinsic";
208 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
209 if (C->print_intrinsics() || C->print_inlining()) {
210 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
211 }
212 } else {
213 // Root compile
214 ResourceMark rm;
215 stringStream msg_stream;
216 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
217 vmIntrinsics::name_at(intrinsic_id()),
218 is_virtual() ? " (virtual)" : "", bci);
219 const char *msg = msg_stream.freeze();
220 log_debug(jit, inlining)("%s", msg);
221 if (C->print_intrinsics() || C->print_inlining()) {
222 C->print_inlining_stream()->print("%s", msg);
223 }
224 }
225 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
226 return NULL;
227 }
228
229 bool LibraryCallKit::try_to_inline(int predicate) {
230 // Handle symbolic names for otherwise undistinguished boolean switches:
231 const bool is_store = true;
232 const bool is_compress = true;
233 const bool is_static = true;
234 const bool is_volatile = true;
235
236 if (!jvms()->has_method()) {
237 // Root JVMState has a null method.
238 assert(map()->memory()->Opcode() == Op_Parm, "");
239 // Insert the memory aliasing node
240 set_all_memory(reset_memory());
241 }
242 assert(merged_memory(), "");
243
244 switch (intrinsic_id()) {
245 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
246 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
247 case vmIntrinsics::_getClass: return inline_native_getClass();
248
249 case vmIntrinsics::_ceil:
250 case vmIntrinsics::_floor:
251 case vmIntrinsics::_rint:
252 case vmIntrinsics::_dsin:
253 case vmIntrinsics::_dcos:
254 case vmIntrinsics::_dtan:
255 case vmIntrinsics::_dabs:
256 case vmIntrinsics::_fabs:
257 case vmIntrinsics::_iabs:
258 case vmIntrinsics::_labs:
259 case vmIntrinsics::_datan2:
260 case vmIntrinsics::_dsqrt:
261 case vmIntrinsics::_dsqrt_strict:
262 case vmIntrinsics::_dexp:
263 case vmIntrinsics::_dlog:
264 case vmIntrinsics::_dlog10:
265 case vmIntrinsics::_dpow:
266 case vmIntrinsics::_dcopySign:
267 case vmIntrinsics::_fcopySign:
268 case vmIntrinsics::_dsignum:
269 case vmIntrinsics::_roundF:
270 case vmIntrinsics::_roundD:
271 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
272
273 case vmIntrinsics::_notify:
274 case vmIntrinsics::_notifyAll:
275 return inline_notify(intrinsic_id());
276
277 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
278 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
279 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
280 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
281 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
282 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
283 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
284 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
285 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
286 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
287 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
288 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
289 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
290 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
291
292 case vmIntrinsics::_arraycopy: return inline_arraycopy();
293
294 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
295 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
296 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
297 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
298
299 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
306 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
307
308 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
309 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
310
311 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
312 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
313 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
314 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
315
316 case vmIntrinsics::_compressStringC:
317 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
318 case vmIntrinsics::_inflateStringC:
319 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
320
321 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
322 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
323 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
324 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
325 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
326 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
327 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
328 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
329 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
330 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
331 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
332 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_PRIMITIVE_OBJECT,Relaxed, false);
333
334 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
335 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
336 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
337 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
338 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
339 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
340 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
341 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
342 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
343 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_PRIMITIVE_OBJECT,Relaxed, false);
344
345 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
346 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
347 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
348 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
349 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
350 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
351 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
352 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
353 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
354
355 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
356 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
357 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
358 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
359 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
360 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
361 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
362 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
363 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
364
365 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
366 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
367 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
368 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
369
370 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
371 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
372 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
373 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
374
375 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
376 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
377 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
378 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
379 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
380 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
381 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
382 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
383 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
384
385 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
386 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
387 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
388 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
389 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
390 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
391 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
392 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
393 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
394
395 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
396 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
397 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
398 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
399 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
400 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
401 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
402 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
403 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
404
405 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
406 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
407 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
408 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
409 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
410 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
411 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
412 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
413 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
414
415 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
417 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
418 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
420
421 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
422 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
423 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
424 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
425 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
426 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
427 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
428 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
429 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
433 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
441
442 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
457
458 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
459 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
460 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
461 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
462
463 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
465 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
466 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
468
469 case vmIntrinsics::_loadFence:
470 case vmIntrinsics::_storeFence:
471 case vmIntrinsics::_storeStoreFence:
472 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
473
474 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
475
476 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
477 case vmIntrinsics::_currentThread: return inline_native_currentThread();
478 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
479
480 case vmIntrinsics::_extentLocalCache: return inline_native_extentLocalCache();
481 case vmIntrinsics::_setExtentLocalCache: return inline_native_setExtentLocalCache();
482
483 #ifdef JFR_HAVE_INTRINSICS
484 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
485 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
486 #endif
487 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
488 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
489 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
490 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
491 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
492 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
493 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
494 case vmIntrinsics::_getLength: return inline_native_getLength();
495 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
496 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
497 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
498 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
499 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
500 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
501 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
502
503 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
504 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
505
506 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
507
508 case vmIntrinsics::_isInstance:
509 case vmIntrinsics::_getModifiers:
510 case vmIntrinsics::_isInterface:
511 case vmIntrinsics::_isArray:
512 case vmIntrinsics::_isPrimitive:
513 case vmIntrinsics::_isHidden:
514 case vmIntrinsics::_getSuperclass:
515 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
516
517 case vmIntrinsics::_asPrimaryType:
518 case vmIntrinsics::_asPrimaryTypeArg:
519 case vmIntrinsics::_asValueType:
520 case vmIntrinsics::_asValueTypeArg: return inline_primitive_Class_conversion(intrinsic_id());
521
522 case vmIntrinsics::_floatToRawIntBits:
523 case vmIntrinsics::_floatToIntBits:
524 case vmIntrinsics::_intBitsToFloat:
525 case vmIntrinsics::_doubleToRawLongBits:
526 case vmIntrinsics::_doubleToLongBits:
527 case vmIntrinsics::_longBitsToDouble:
528 case vmIntrinsics::_floatToFloat16:
529 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
530
531 case vmIntrinsics::_floatIsFinite:
532 case vmIntrinsics::_floatIsInfinite:
533 case vmIntrinsics::_doubleIsFinite:
534 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
535
536 case vmIntrinsics::_numberOfLeadingZeros_i:
537 case vmIntrinsics::_numberOfLeadingZeros_l:
538 case vmIntrinsics::_numberOfTrailingZeros_i:
539 case vmIntrinsics::_numberOfTrailingZeros_l:
540 case vmIntrinsics::_bitCount_i:
541 case vmIntrinsics::_bitCount_l:
542 case vmIntrinsics::_reverse_i:
543 case vmIntrinsics::_reverse_l:
544 case vmIntrinsics::_reverseBytes_i:
545 case vmIntrinsics::_reverseBytes_l:
546 case vmIntrinsics::_reverseBytes_s:
547 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
548
549 case vmIntrinsics::_compress_i:
550 case vmIntrinsics::_compress_l:
551 case vmIntrinsics::_expand_i:
552 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
553
554 case vmIntrinsics::_compareUnsigned_i:
555 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
556
557 case vmIntrinsics::_divideUnsigned_i:
558 case vmIntrinsics::_divideUnsigned_l:
559 case vmIntrinsics::_remainderUnsigned_i:
560 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
561
562 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
563
564 case vmIntrinsics::_Reference_get: return inline_reference_get();
565 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
566 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
567
568 case vmIntrinsics::_Class_cast: return inline_Class_cast();
569
570 case vmIntrinsics::_aescrypt_encryptBlock:
571 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
572
573 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
574 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
575 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
576
577 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
578 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
579 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
580
581 case vmIntrinsics::_counterMode_AESCrypt:
582 return inline_counterMode_AESCrypt(intrinsic_id());
583
584 case vmIntrinsics::_galoisCounterMode_AESCrypt:
585 return inline_galoisCounterMode_AESCrypt();
586
587 case vmIntrinsics::_md5_implCompress:
588 case vmIntrinsics::_sha_implCompress:
589 case vmIntrinsics::_sha2_implCompress:
590 case vmIntrinsics::_sha5_implCompress:
591 case vmIntrinsics::_sha3_implCompress:
592 return inline_digestBase_implCompress(intrinsic_id());
593
594 case vmIntrinsics::_digestBase_implCompressMB:
595 return inline_digestBase_implCompressMB(predicate);
596
597 case vmIntrinsics::_multiplyToLen:
598 return inline_multiplyToLen();
599
600 case vmIntrinsics::_squareToLen:
601 return inline_squareToLen();
602
603 case vmIntrinsics::_mulAdd:
604 return inline_mulAdd();
605
606 case vmIntrinsics::_montgomeryMultiply:
607 return inline_montgomeryMultiply();
608 case vmIntrinsics::_montgomerySquare:
609 return inline_montgomerySquare();
610
611 case vmIntrinsics::_bigIntegerRightShiftWorker:
612 return inline_bigIntegerShift(true);
613 case vmIntrinsics::_bigIntegerLeftShiftWorker:
614 return inline_bigIntegerShift(false);
615
616 case vmIntrinsics::_vectorizedMismatch:
617 return inline_vectorizedMismatch();
618
619 case vmIntrinsics::_ghash_processBlocks:
620 return inline_ghash_processBlocks();
621 case vmIntrinsics::_base64_encodeBlock:
622 return inline_base64_encodeBlock();
623 case vmIntrinsics::_base64_decodeBlock:
624 return inline_base64_decodeBlock();
625
626 case vmIntrinsics::_encodeISOArray:
627 case vmIntrinsics::_encodeByteISOArray:
628 return inline_encodeISOArray(false);
629 case vmIntrinsics::_encodeAsciiArray:
630 return inline_encodeISOArray(true);
631
632 case vmIntrinsics::_updateCRC32:
633 return inline_updateCRC32();
634 case vmIntrinsics::_updateBytesCRC32:
635 return inline_updateBytesCRC32();
636 case vmIntrinsics::_updateByteBufferCRC32:
637 return inline_updateByteBufferCRC32();
638
639 case vmIntrinsics::_updateBytesCRC32C:
640 return inline_updateBytesCRC32C();
641 case vmIntrinsics::_updateDirectByteBufferCRC32C:
642 return inline_updateDirectByteBufferCRC32C();
643
644 case vmIntrinsics::_updateBytesAdler32:
645 return inline_updateBytesAdler32();
646 case vmIntrinsics::_updateByteBufferAdler32:
647 return inline_updateByteBufferAdler32();
648
649 case vmIntrinsics::_profileBoolean:
650 return inline_profileBoolean();
651 case vmIntrinsics::_isCompileConstant:
652 return inline_isCompileConstant();
653
654 case vmIntrinsics::_countPositives:
655 return inline_countPositives();
656
657 case vmIntrinsics::_fmaD:
658 case vmIntrinsics::_fmaF:
659 return inline_fma(intrinsic_id());
660
661 case vmIntrinsics::_isDigit:
662 case vmIntrinsics::_isLowerCase:
663 case vmIntrinsics::_isUpperCase:
664 case vmIntrinsics::_isWhitespace:
665 return inline_character_compare(intrinsic_id());
666
667 case vmIntrinsics::_min:
668 case vmIntrinsics::_max:
669 case vmIntrinsics::_min_strict:
670 case vmIntrinsics::_max_strict:
671 return inline_min_max(intrinsic_id());
672
673 case vmIntrinsics::_maxF:
674 case vmIntrinsics::_minF:
675 case vmIntrinsics::_maxD:
676 case vmIntrinsics::_minD:
677 case vmIntrinsics::_maxF_strict:
678 case vmIntrinsics::_minF_strict:
679 case vmIntrinsics::_maxD_strict:
680 case vmIntrinsics::_minD_strict:
681 return inline_fp_min_max(intrinsic_id());
682
683 case vmIntrinsics::_VectorUnaryOp:
684 return inline_vector_nary_operation(1);
685 case vmIntrinsics::_VectorBinaryOp:
686 return inline_vector_nary_operation(2);
687 case vmIntrinsics::_VectorTernaryOp:
688 return inline_vector_nary_operation(3);
689 case vmIntrinsics::_VectorFromBitsCoerced:
690 return inline_vector_frombits_coerced();
691 case vmIntrinsics::_VectorShuffleIota:
692 return inline_vector_shuffle_iota();
693 case vmIntrinsics::_VectorMaskOp:
694 return inline_vector_mask_operation();
695 case vmIntrinsics::_VectorShuffleToVector:
696 return inline_vector_shuffle_to_vector();
697 case vmIntrinsics::_VectorLoadOp:
698 return inline_vector_mem_operation(/*is_store=*/false);
699 case vmIntrinsics::_VectorLoadMaskedOp:
700 return inline_vector_mem_masked_operation(/*is_store*/false);
701 case vmIntrinsics::_VectorStoreOp:
702 return inline_vector_mem_operation(/*is_store=*/true);
703 case vmIntrinsics::_VectorStoreMaskedOp:
704 return inline_vector_mem_masked_operation(/*is_store=*/true);
705 case vmIntrinsics::_VectorGatherOp:
706 return inline_vector_gather_scatter(/*is_scatter*/ false);
707 case vmIntrinsics::_VectorScatterOp:
708 return inline_vector_gather_scatter(/*is_scatter*/ true);
709 case vmIntrinsics::_VectorReductionCoerced:
710 return inline_vector_reduction();
711 case vmIntrinsics::_VectorTest:
712 return inline_vector_test();
713 case vmIntrinsics::_VectorBlend:
714 return inline_vector_blend();
715 case vmIntrinsics::_VectorRearrange:
716 return inline_vector_rearrange();
717 case vmIntrinsics::_VectorCompare:
718 return inline_vector_compare();
719 case vmIntrinsics::_VectorBroadcastInt:
720 return inline_vector_broadcast_int();
721 case vmIntrinsics::_VectorConvert:
722 return inline_vector_convert();
723 case vmIntrinsics::_VectorInsert:
724 return inline_vector_insert();
725 case vmIntrinsics::_VectorExtract:
726 return inline_vector_extract();
727 case vmIntrinsics::_VectorCompressExpand:
728 return inline_vector_compress_expand();
729 case vmIntrinsics::_IndexVector:
730 return inline_index_vector();
731
732 case vmIntrinsics::_getObjectSize:
733 return inline_getObjectSize();
734
735 case vmIntrinsics::_blackhole:
736 return inline_blackhole();
737
738 default:
739 // If you get here, it may be that someone has added a new intrinsic
740 // to the list in vmIntrinsics.hpp without implementing it here.
741 #ifndef PRODUCT
742 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
743 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
744 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
745 }
746 #endif
747 return false;
748 }
749 }
750
751 Node* LibraryCallKit::try_to_predicate(int predicate) {
752 if (!jvms()->has_method()) {
753 // Root JVMState has a null method.
754 assert(map()->memory()->Opcode() == Op_Parm, "");
755 // Insert the memory aliasing node
756 set_all_memory(reset_memory());
757 }
758 assert(merged_memory(), "");
759
760 switch (intrinsic_id()) {
761 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
762 return inline_cipherBlockChaining_AESCrypt_predicate(false);
763 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
764 return inline_cipherBlockChaining_AESCrypt_predicate(true);
765 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
766 return inline_electronicCodeBook_AESCrypt_predicate(false);
767 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
768 return inline_electronicCodeBook_AESCrypt_predicate(true);
769 case vmIntrinsics::_counterMode_AESCrypt:
770 return inline_counterMode_AESCrypt_predicate();
771 case vmIntrinsics::_digestBase_implCompressMB:
772 return inline_digestBase_implCompressMB_predicate(predicate);
773 case vmIntrinsics::_galoisCounterMode_AESCrypt:
774 return inline_galoisCounterMode_AESCrypt_predicate();
775
776 default:
777 // If you get here, it may be that someone has added a new intrinsic
778 // to the list in vmIntrinsics.hpp without implementing it here.
779 #ifndef PRODUCT
780 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
781 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
782 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
783 }
784 #endif
785 Node* slow_ctl = control();
786 set_control(top()); // No fast path intrinsic
787 return slow_ctl;
788 }
789 }
790
791 //------------------------------set_result-------------------------------
792 // Helper function for finishing intrinsics.
793 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
794 record_for_igvn(region);
795 set_control(_gvn.transform(region));
796 set_result( _gvn.transform(value));
797 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
798 }
799
800 //------------------------------generate_guard---------------------------
801 // Helper function for generating guarded fast-slow graph structures.
802 // The given 'test', if true, guards a slow path. If the test fails
803 // then a fast path can be taken. (We generally hope it fails.)
804 // In all cases, GraphKit::control() is updated to the fast path.
805 // The returned value represents the control for the slow path.
806 // The return value is never 'top'; it is either a valid control
807 // or NULL if it is obvious that the slow path can never be taken.
808 // Also, if region and the slow control are not NULL, the slow edge
809 // is appended to the region.
810 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
811 if (stopped()) {
812 // Already short circuited.
813 return NULL;
814 }
815
816 // Build an if node and its projections.
817 // If test is true we take the slow path, which we assume is uncommon.
818 if (_gvn.type(test) == TypeInt::ZERO) {
819 // The slow branch is never taken. No need to build this guard.
820 return NULL;
821 }
822
823 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
824
825 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
826 if (if_slow == top()) {
827 // The slow branch is never taken. No need to build this guard.
828 return NULL;
829 }
830
831 if (region != NULL)
832 region->add_req(if_slow);
833
834 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
835 set_control(if_fast);
836
837 return if_slow;
838 }
839
840 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
841 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
842 }
843 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
844 return generate_guard(test, region, PROB_FAIR);
845 }
846
847 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
848 Node* *pos_index) {
849 if (stopped())
850 return NULL; // already stopped
851 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
852 return NULL; // index is already adequately typed
853 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
854 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
855 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
856 if (is_neg != NULL && pos_index != NULL) {
857 // Emulate effect of Parse::adjust_map_after_if.
858 Node* ccast = new CastIINode(index, TypeInt::POS);
859 ccast->set_req(0, control());
860 (*pos_index) = _gvn.transform(ccast);
861 }
862 return is_neg;
863 }
864
865 // Make sure that 'position' is a valid limit index, in [0..length].
866 // There are two equivalent plans for checking this:
867 // A. (offset + copyLength) unsigned<= arrayLength
868 // B. offset <= (arrayLength - copyLength)
869 // We require that all of the values above, except for the sum and
870 // difference, are already known to be non-negative.
871 // Plan A is robust in the face of overflow, if offset and copyLength
872 // are both hugely positive.
873 //
874 // Plan B is less direct and intuitive, but it does not overflow at
875 // all, since the difference of two non-negatives is always
876 // representable. Whenever Java methods must perform the equivalent
877 // check they generally use Plan B instead of Plan A.
878 // For the moment we use Plan A.
879 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
880 Node* subseq_length,
881 Node* array_length,
882 RegionNode* region) {
883 if (stopped())
884 return NULL; // already stopped
885 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
886 if (zero_offset && subseq_length->eqv_uncast(array_length))
887 return NULL; // common case of whole-array copy
888 Node* last = subseq_length;
889 if (!zero_offset) // last += offset
890 last = _gvn.transform(new AddINode(last, offset));
891 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
892 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
893 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
894 return is_over;
895 }
896
897 // Emit range checks for the given String.value byte array
898 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
899 if (stopped()) {
900 return; // already stopped
901 }
902 RegionNode* bailout = new RegionNode(1);
903 record_for_igvn(bailout);
904 if (char_count) {
905 // Convert char count to byte count
906 count = _gvn.transform(new LShiftINode(count, intcon(1)));
907 }
908
909 // Offset and count must not be negative
910 generate_negative_guard(offset, bailout);
911 generate_negative_guard(count, bailout);
912 // Offset + count must not exceed length of array
913 generate_limit_guard(offset, count, load_array_length(array), bailout);
914
915 if (bailout->req() > 1) {
916 PreserveJVMState pjvms(this);
917 set_control(_gvn.transform(bailout));
918 uncommon_trap(Deoptimization::Reason_intrinsic,
919 Deoptimization::Action_maybe_recompile);
920 }
921 }
922
923 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
924 bool is_immutable) {
925 ciKlass* thread_klass = env()->Thread_klass();
926 const Type* thread_type
927 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
928
929 Node* thread = _gvn.transform(new ThreadLocalNode());
930 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
931 tls_output = thread;
932
933 Node* thread_obj_handle
934 = (is_immutable
935 ? LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
936 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
937 : make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
938 thread_obj_handle = _gvn.transform(thread_obj_handle);
939
940 DecoratorSet decorators = IN_NATIVE;
941 if (is_immutable) {
942 decorators |= C2_IMMUTABLE_MEMORY;
943 }
944 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
945 }
946
947 //--------------------------generate_current_thread--------------------
948 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
949 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
950 /*is_immutable*/false);
951 }
952
953 //--------------------------generate_virtual_thread--------------------
954 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
955 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
956 !C->method()->changes_current_thread());
957 }
958
959 //------------------------------make_string_method_node------------------------
960 // Helper method for String intrinsic functions. This version is called with
961 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
962 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
963 // containing the lengths of str1 and str2.
964 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
965 Node* result = NULL;
966 switch (opcode) {
967 case Op_StrIndexOf:
968 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
969 str1_start, cnt1, str2_start, cnt2, ae);
970 break;
971 case Op_StrComp:
972 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
973 str1_start, cnt1, str2_start, cnt2, ae);
974 break;
975 case Op_StrEquals:
976 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
977 // Use the constant length if there is one because optimized match rule may exist.
978 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
979 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
980 break;
981 default:
982 ShouldNotReachHere();
983 return NULL;
984 }
985
986 // All these intrinsics have checks.
987 C->set_has_split_ifs(true); // Has chance for split-if optimization
988 clear_upper_avx();
989
990 return _gvn.transform(result);
991 }
992
993 //------------------------------inline_string_compareTo------------------------
994 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
995 Node* arg1 = argument(0);
996 Node* arg2 = argument(1);
997
998 arg1 = must_be_not_null(arg1, true);
999 arg2 = must_be_not_null(arg2, true);
1000
1001 // Get start addr and length of first argument
1002 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1003 Node* arg1_cnt = load_array_length(arg1);
1004
1005 // Get start addr and length of second argument
1006 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1007 Node* arg2_cnt = load_array_length(arg2);
1008
1009 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1010 set_result(result);
1011 return true;
1012 }
1013
1014 //------------------------------inline_string_equals------------------------
1015 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1016 Node* arg1 = argument(0);
1017 Node* arg2 = argument(1);
1018
1019 // paths (plus control) merge
1020 RegionNode* region = new RegionNode(3);
1021 Node* phi = new PhiNode(region, TypeInt::BOOL);
1022
1023 if (!stopped()) {
1024
1025 arg1 = must_be_not_null(arg1, true);
1026 arg2 = must_be_not_null(arg2, true);
1027
1028 // Get start addr and length of first argument
1029 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1030 Node* arg1_cnt = load_array_length(arg1);
1031
1032 // Get start addr and length of second argument
1033 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1034 Node* arg2_cnt = load_array_length(arg2);
1035
1036 // Check for arg1_cnt != arg2_cnt
1037 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1038 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1039 Node* if_ne = generate_slow_guard(bol, NULL);
1040 if (if_ne != NULL) {
1041 phi->init_req(2, intcon(0));
1042 region->init_req(2, if_ne);
1043 }
1044
1045 // Check for count == 0 is done by assembler code for StrEquals.
1046
1047 if (!stopped()) {
1048 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1049 phi->init_req(1, equals);
1050 region->init_req(1, control());
1051 }
1052 }
1053
1054 // post merge
1055 set_control(_gvn.transform(region));
1056 record_for_igvn(region);
1057
1058 set_result(_gvn.transform(phi));
1059 return true;
1060 }
1061
1062 //------------------------------inline_array_equals----------------------------
1063 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1064 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1065 Node* arg1 = argument(0);
1066 Node* arg2 = argument(1);
1067
1068 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1069 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1070 clear_upper_avx();
1071
1072 return true;
1073 }
1074
1075 //------------------------------inline_countPositives------------------------------
1076 bool LibraryCallKit::inline_countPositives() {
1077 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1078 return false;
1079 }
1080
1081 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1082 // no receiver since it is static method
1083 Node* ba = argument(0);
1084 Node* offset = argument(1);
1085 Node* len = argument(2);
1086
1087 ba = must_be_not_null(ba, true);
1088
1089 // Range checks
1090 generate_string_range_check(ba, offset, len, false);
1091 if (stopped()) {
1092 return true;
1093 }
1094 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1095 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1096 set_result(_gvn.transform(result));
1097 return true;
1098 }
1099
1100 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1101 Node* index = argument(0);
1102 Node* length = bt == T_INT ? argument(1) : argument(2);
1103 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1104 return false;
1105 }
1106
1107 // check that length is positive
1108 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1109 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1110
1111 {
1112 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1113 uncommon_trap(Deoptimization::Reason_intrinsic,
1114 Deoptimization::Action_make_not_entrant);
1115 }
1116
1117 if (stopped()) {
1118 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1119 return true;
1120 }
1121
1122 // length is now known positive, add a cast node to make this explicit
1123 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1124 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1125 casted_length = _gvn.transform(casted_length);
1126 replace_in_map(length, casted_length);
1127 length = casted_length;
1128
1129 // Use an unsigned comparison for the range check itself
1130 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1131 BoolTest::mask btest = BoolTest::lt;
1132 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1133 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1134 _gvn.set_type(rc, rc->Value(&_gvn));
1135 if (!rc_bool->is_Con()) {
1136 record_for_igvn(rc);
1137 }
1138 set_control(_gvn.transform(new IfTrueNode(rc)));
1139 {
1140 PreserveJVMState pjvms(this);
1141 set_control(_gvn.transform(new IfFalseNode(rc)));
1142 uncommon_trap(Deoptimization::Reason_range_check,
1143 Deoptimization::Action_make_not_entrant);
1144 }
1145
1146 if (stopped()) {
1147 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1148 return true;
1149 }
1150
1151 // index is now known to be >= 0 and < length, cast it
1152 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1153 result = _gvn.transform(result);
1154 set_result(result);
1155 replace_in_map(index, result);
1156 return true;
1157 }
1158
1159 //------------------------------inline_string_indexOf------------------------
1160 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1161 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1162 return false;
1163 }
1164 Node* src = argument(0);
1165 Node* tgt = argument(1);
1166
1167 // Make the merge point
1168 RegionNode* result_rgn = new RegionNode(4);
1169 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1170
1171 src = must_be_not_null(src, true);
1172 tgt = must_be_not_null(tgt, true);
1173
1174 // Get start addr and length of source string
1175 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1176 Node* src_count = load_array_length(src);
1177
1178 // Get start addr and length of substring
1179 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1180 Node* tgt_count = load_array_length(tgt);
1181
1182 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1183 // Divide src size by 2 if String is UTF16 encoded
1184 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1185 }
1186 if (ae == StrIntrinsicNode::UU) {
1187 // Divide substring size by 2 if String is UTF16 encoded
1188 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1189 }
1190
1191 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1192 if (result != NULL) {
1193 result_phi->init_req(3, result);
1194 result_rgn->init_req(3, control());
1195 }
1196 set_control(_gvn.transform(result_rgn));
1197 record_for_igvn(result_rgn);
1198 set_result(_gvn.transform(result_phi));
1199
1200 return true;
1201 }
1202
1203 //-----------------------------inline_string_indexOf-----------------------
1204 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1205 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1206 return false;
1207 }
1208 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1209 return false;
1210 }
1211 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1212 Node* src = argument(0); // byte[]
1213 Node* src_count = argument(1); // char count
1214 Node* tgt = argument(2); // byte[]
1215 Node* tgt_count = argument(3); // char count
1216 Node* from_index = argument(4); // char index
1217
1218 src = must_be_not_null(src, true);
1219 tgt = must_be_not_null(tgt, true);
1220
1221 // Multiply byte array index by 2 if String is UTF16 encoded
1222 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1223 src_count = _gvn.transform(new SubINode(src_count, from_index));
1224 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1225 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1226
1227 // Range checks
1228 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1229 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1230 if (stopped()) {
1231 return true;
1232 }
1233
1234 RegionNode* region = new RegionNode(5);
1235 Node* phi = new PhiNode(region, TypeInt::INT);
1236
1237 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1238 if (result != NULL) {
1239 // The result is index relative to from_index if substring was found, -1 otherwise.
1240 // Generate code which will fold into cmove.
1241 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1242 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1243
1244 Node* if_lt = generate_slow_guard(bol, NULL);
1245 if (if_lt != NULL) {
1246 // result == -1
1247 phi->init_req(3, result);
1248 region->init_req(3, if_lt);
1249 }
1250 if (!stopped()) {
1251 result = _gvn.transform(new AddINode(result, from_index));
1252 phi->init_req(4, result);
1253 region->init_req(4, control());
1254 }
1255 }
1256
1257 set_control(_gvn.transform(region));
1258 record_for_igvn(region);
1259 set_result(_gvn.transform(phi));
1260 clear_upper_avx();
1261
1262 return true;
1263 }
1264
1265 // Create StrIndexOfNode with fast path checks
1266 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1267 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1268 // Check for substr count > string count
1269 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1270 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1271 Node* if_gt = generate_slow_guard(bol, NULL);
1272 if (if_gt != NULL) {
1273 phi->init_req(1, intcon(-1));
1274 region->init_req(1, if_gt);
1275 }
1276 if (!stopped()) {
1277 // Check for substr count == 0
1278 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1279 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1280 Node* if_zero = generate_slow_guard(bol, NULL);
1281 if (if_zero != NULL) {
1282 phi->init_req(2, intcon(0));
1283 region->init_req(2, if_zero);
1284 }
1285 }
1286 if (!stopped()) {
1287 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1288 }
1289 return NULL;
1290 }
1291
1292 //-----------------------------inline_string_indexOfChar-----------------------
1293 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1294 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1295 return false;
1296 }
1297 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1298 return false;
1299 }
1300 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1301 Node* src = argument(0); // byte[]
1302 Node* tgt = argument(1); // tgt is int ch
1303 Node* from_index = argument(2);
1304 Node* max = argument(3);
1305
1306 src = must_be_not_null(src, true);
1307
1308 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1309 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1310 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1311
1312 // Range checks
1313 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1314 if (stopped()) {
1315 return true;
1316 }
1317
1318 RegionNode* region = new RegionNode(3);
1319 Node* phi = new PhiNode(region, TypeInt::INT);
1320
1321 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1322 C->set_has_split_ifs(true); // Has chance for split-if optimization
1323 _gvn.transform(result);
1324
1325 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1326 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1327
1328 Node* if_lt = generate_slow_guard(bol, NULL);
1329 if (if_lt != NULL) {
1330 // result == -1
1331 phi->init_req(2, result);
1332 region->init_req(2, if_lt);
1333 }
1334 if (!stopped()) {
1335 result = _gvn.transform(new AddINode(result, from_index));
1336 phi->init_req(1, result);
1337 region->init_req(1, control());
1338 }
1339 set_control(_gvn.transform(region));
1340 record_for_igvn(region);
1341 set_result(_gvn.transform(phi));
1342
1343 return true;
1344 }
1345 //---------------------------inline_string_copy---------------------
1346 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1347 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1348 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1349 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1350 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1351 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1352 bool LibraryCallKit::inline_string_copy(bool compress) {
1353 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1354 return false;
1355 }
1356 int nargs = 5; // 2 oops, 3 ints
1357 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1358
1359 Node* src = argument(0);
1360 Node* src_offset = argument(1);
1361 Node* dst = argument(2);
1362 Node* dst_offset = argument(3);
1363 Node* length = argument(4);
1364
1365 // Check for allocation before we add nodes that would confuse
1366 // tightly_coupled_allocation()
1367 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1368
1369 // Figure out the size and type of the elements we will be copying.
1370 const Type* src_type = src->Value(&_gvn);
1371 const Type* dst_type = dst->Value(&_gvn);
1372 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
1373 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
1374 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1375 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1376 "Unsupported array types for inline_string_copy");
1377
1378 src = must_be_not_null(src, true);
1379 dst = must_be_not_null(dst, true);
1380
1381 // Convert char[] offsets to byte[] offsets
1382 bool convert_src = (compress && src_elem == T_BYTE);
1383 bool convert_dst = (!compress && dst_elem == T_BYTE);
1384 if (convert_src) {
1385 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1386 } else if (convert_dst) {
1387 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1388 }
1389
1390 // Range checks
1391 generate_string_range_check(src, src_offset, length, convert_src);
1392 generate_string_range_check(dst, dst_offset, length, convert_dst);
1393 if (stopped()) {
1394 return true;
1395 }
1396
1397 Node* src_start = array_element_address(src, src_offset, src_elem);
1398 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1399 // 'src_start' points to src array + scaled offset
1400 // 'dst_start' points to dst array + scaled offset
1401 Node* count = NULL;
1402 if (compress) {
1403 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1404 } else {
1405 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1406 }
1407
1408 if (alloc != NULL) {
1409 if (alloc->maybe_set_complete(&_gvn)) {
1410 // "You break it, you buy it."
1411 InitializeNode* init = alloc->initialization();
1412 assert(init->is_complete(), "we just did this");
1413 init->set_complete_with_arraycopy();
1414 assert(dst->is_CheckCastPP(), "sanity");
1415 assert(dst->in(0)->in(0) == init, "dest pinned");
1416 }
1417 // Do not let stores that initialize this object be reordered with
1418 // a subsequent store that would make this object accessible by
1419 // other threads.
1420 // Record what AllocateNode this StoreStore protects so that
1421 // escape analysis can go from the MemBarStoreStoreNode to the
1422 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1423 // based on the escape status of the AllocateNode.
1424 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1425 }
1426 if (compress) {
1427 set_result(_gvn.transform(count));
1428 }
1429 clear_upper_avx();
1430
1431 return true;
1432 }
1433
1434 #ifdef _LP64
1435 #define XTOP ,top() /*additional argument*/
1436 #else //_LP64
1437 #define XTOP /*no additional argument*/
1438 #endif //_LP64
1439
1440 //------------------------inline_string_toBytesU--------------------------
1441 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1442 bool LibraryCallKit::inline_string_toBytesU() {
1443 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1444 return false;
1445 }
1446 // Get the arguments.
1447 Node* value = argument(0);
1448 Node* offset = argument(1);
1449 Node* length = argument(2);
1450
1451 Node* newcopy = NULL;
1452
1453 // Set the original stack and the reexecute bit for the interpreter to reexecute
1454 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1455 { PreserveReexecuteState preexecs(this);
1456 jvms()->set_should_reexecute(true);
1457
1458 // Check if a null path was taken unconditionally.
1459 value = null_check(value);
1460
1461 RegionNode* bailout = new RegionNode(1);
1462 record_for_igvn(bailout);
1463
1464 // Range checks
1465 generate_negative_guard(offset, bailout);
1466 generate_negative_guard(length, bailout);
1467 generate_limit_guard(offset, length, load_array_length(value), bailout);
1468 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1469 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1470
1471 if (bailout->req() > 1) {
1472 PreserveJVMState pjvms(this);
1473 set_control(_gvn.transform(bailout));
1474 uncommon_trap(Deoptimization::Reason_intrinsic,
1475 Deoptimization::Action_maybe_recompile);
1476 }
1477 if (stopped()) {
1478 return true;
1479 }
1480
1481 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1482 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1483 newcopy = new_array(klass_node, size, 0); // no arguments to push
1484 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1485 guarantee(alloc != NULL, "created above");
1486
1487 // Calculate starting addresses.
1488 Node* src_start = array_element_address(value, offset, T_CHAR);
1489 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1490
1491 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1492 const TypeInt* toffset = gvn().type(offset)->is_int();
1493 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1494
1495 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1496 const char* copyfunc_name = "arraycopy";
1497 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1498 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1499 OptoRuntime::fast_arraycopy_Type(),
1500 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1501 src_start, dst_start, ConvI2X(length) XTOP);
1502 // Do not let reads from the cloned object float above the arraycopy.
1503 if (alloc->maybe_set_complete(&_gvn)) {
1504 // "You break it, you buy it."
1505 InitializeNode* init = alloc->initialization();
1506 assert(init->is_complete(), "we just did this");
1507 init->set_complete_with_arraycopy();
1508 assert(newcopy->is_CheckCastPP(), "sanity");
1509 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1510 }
1511 // Do not let stores that initialize this object be reordered with
1512 // a subsequent store that would make this object accessible by
1513 // other threads.
1514 // Record what AllocateNode this StoreStore protects so that
1515 // escape analysis can go from the MemBarStoreStoreNode to the
1516 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1517 // based on the escape status of the AllocateNode.
1518 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1519 } // original reexecute is set back here
1520
1521 C->set_has_split_ifs(true); // Has chance for split-if optimization
1522 if (!stopped()) {
1523 set_result(newcopy);
1524 }
1525 clear_upper_avx();
1526
1527 return true;
1528 }
1529
1530 //------------------------inline_string_getCharsU--------------------------
1531 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1532 bool LibraryCallKit::inline_string_getCharsU() {
1533 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1534 return false;
1535 }
1536
1537 // Get the arguments.
1538 Node* src = argument(0);
1539 Node* src_begin = argument(1);
1540 Node* src_end = argument(2); // exclusive offset (i < src_end)
1541 Node* dst = argument(3);
1542 Node* dst_begin = argument(4);
1543
1544 // Check for allocation before we add nodes that would confuse
1545 // tightly_coupled_allocation()
1546 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1547
1548 // Check if a null path was taken unconditionally.
1549 src = null_check(src);
1550 dst = null_check(dst);
1551 if (stopped()) {
1552 return true;
1553 }
1554
1555 // Get length and convert char[] offset to byte[] offset
1556 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1557 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1558
1559 // Range checks
1560 generate_string_range_check(src, src_begin, length, true);
1561 generate_string_range_check(dst, dst_begin, length, false);
1562 if (stopped()) {
1563 return true;
1564 }
1565
1566 if (!stopped()) {
1567 // Calculate starting addresses.
1568 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1569 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1570
1571 // Check if array addresses are aligned to HeapWordSize
1572 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1573 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1574 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1575 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1576
1577 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1578 const char* copyfunc_name = "arraycopy";
1579 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1580 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1581 OptoRuntime::fast_arraycopy_Type(),
1582 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1583 src_start, dst_start, ConvI2X(length) XTOP);
1584 // Do not let reads from the cloned object float above the arraycopy.
1585 if (alloc != NULL) {
1586 if (alloc->maybe_set_complete(&_gvn)) {
1587 // "You break it, you buy it."
1588 InitializeNode* init = alloc->initialization();
1589 assert(init->is_complete(), "we just did this");
1590 init->set_complete_with_arraycopy();
1591 assert(dst->is_CheckCastPP(), "sanity");
1592 assert(dst->in(0)->in(0) == init, "dest pinned");
1593 }
1594 // Do not let stores that initialize this object be reordered with
1595 // a subsequent store that would make this object accessible by
1596 // other threads.
1597 // Record what AllocateNode this StoreStore protects so that
1598 // escape analysis can go from the MemBarStoreStoreNode to the
1599 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1600 // based on the escape status of the AllocateNode.
1601 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1602 } else {
1603 insert_mem_bar(Op_MemBarCPUOrder);
1604 }
1605 }
1606
1607 C->set_has_split_ifs(true); // Has chance for split-if optimization
1608 return true;
1609 }
1610
1611 //----------------------inline_string_char_access----------------------------
1612 // Store/Load char to/from byte[] array.
1613 // static void StringUTF16.putChar(byte[] val, int index, int c)
1614 // static char StringUTF16.getChar(byte[] val, int index)
1615 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1616 Node* value = argument(0);
1617 Node* index = argument(1);
1618 Node* ch = is_store ? argument(2) : NULL;
1619
1620 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1621 // correctly requires matched array shapes.
1622 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1623 "sanity: byte[] and char[] bases agree");
1624 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1625 "sanity: byte[] and char[] scales agree");
1626
1627 // Bail when getChar over constants is requested: constant folding would
1628 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1629 // Java method would constant fold nicely instead.
1630 if (!is_store && value->is_Con() && index->is_Con()) {
1631 return false;
1632 }
1633
1634 // Save state and restore on bailout
1635 uint old_sp = sp();
1636 SafePointNode* old_map = clone_map();
1637
1638 value = must_be_not_null(value, true);
1639
1640 Node* adr = array_element_address(value, index, T_CHAR);
1641 if (adr->is_top()) {
1642 set_map(old_map);
1643 set_sp(old_sp);
1644 return false;
1645 }
1646 old_map->destruct(&_gvn);
1647 if (is_store) {
1648 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1649 } else {
1650 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);
1651 set_result(ch);
1652 }
1653 return true;
1654 }
1655
1656 //--------------------------round_double_node--------------------------------
1657 // Round a double node if necessary.
1658 Node* LibraryCallKit::round_double_node(Node* n) {
1659 if (Matcher::strict_fp_requires_explicit_rounding) {
1660 #ifdef IA32
1661 if (UseSSE < 2) {
1662 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1663 }
1664 #else
1665 Unimplemented();
1666 #endif // IA32
1667 }
1668 return n;
1669 }
1670
1671 //------------------------------inline_math-----------------------------------
1672 // public static double Math.abs(double)
1673 // public static double Math.sqrt(double)
1674 // public static double Math.log(double)
1675 // public static double Math.log10(double)
1676 // public static double Math.round(double)
1677 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1678 Node* arg = round_double_node(argument(0));
1679 Node* n = NULL;
1680 switch (id) {
1681 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1682 case vmIntrinsics::_dsqrt:
1683 case vmIntrinsics::_dsqrt_strict:
1684 n = new SqrtDNode(C, control(), arg); break;
1685 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1686 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1687 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1688 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1689 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1690 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1691 default: fatal_unexpected_iid(id); break;
1692 }
1693 set_result(_gvn.transform(n));
1694 return true;
1695 }
1696
1697 //------------------------------inline_math-----------------------------------
1698 // public static float Math.abs(float)
1699 // public static int Math.abs(int)
1700 // public static long Math.abs(long)
1701 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1702 Node* arg = argument(0);
1703 Node* n = NULL;
1704 switch (id) {
1705 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1706 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1707 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1708 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1709 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1710 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1711 default: fatal_unexpected_iid(id); break;
1712 }
1713 set_result(_gvn.transform(n));
1714 return true;
1715 }
1716
1717 //------------------------------runtime_math-----------------------------
1718 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1719 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1720 "must be (DD)D or (D)D type");
1721
1722 // Inputs
1723 Node* a = round_double_node(argument(0));
1724 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1725
1726 const TypePtr* no_memory_effects = NULL;
1727 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1728 no_memory_effects,
1729 a, top(), b, b ? top() : NULL);
1730 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1731 #ifdef ASSERT
1732 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1733 assert(value_top == top(), "second value must be top");
1734 #endif
1735
1736 set_result(value);
1737 return true;
1738 }
1739
1740 //------------------------------inline_math_pow-----------------------------
1741 bool LibraryCallKit::inline_math_pow() {
1742 Node* exp = round_double_node(argument(2));
1743 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1744 if (d != NULL) {
1745 if (d->getd() == 2.0) {
1746 // Special case: pow(x, 2.0) => x * x
1747 Node* base = round_double_node(argument(0));
1748 set_result(_gvn.transform(new MulDNode(base, base)));
1749 return true;
1750 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1751 // Special case: pow(x, 0.5) => sqrt(x)
1752 Node* base = round_double_node(argument(0));
1753 Node* zero = _gvn.zerocon(T_DOUBLE);
1754
1755 RegionNode* region = new RegionNode(3);
1756 Node* phi = new PhiNode(region, Type::DOUBLE);
1757
1758 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1759 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1760 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1761 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1762 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1763
1764 Node* if_pow = generate_slow_guard(test, NULL);
1765 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1766 phi->init_req(1, value_sqrt);
1767 region->init_req(1, control());
1768
1769 if (if_pow != NULL) {
1770 set_control(if_pow);
1771 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1772 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1773 const TypePtr* no_memory_effects = NULL;
1774 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1775 no_memory_effects, base, top(), exp, top());
1776 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1777 #ifdef ASSERT
1778 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1779 assert(value_top == top(), "second value must be top");
1780 #endif
1781 phi->init_req(2, value_pow);
1782 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1783 }
1784
1785 C->set_has_split_ifs(true); // Has chance for split-if optimization
1786 set_control(_gvn.transform(region));
1787 record_for_igvn(region);
1788 set_result(_gvn.transform(phi));
1789
1790 return true;
1791 }
1792 }
1793
1794 return StubRoutines::dpow() != NULL ?
1795 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1796 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1797 }
1798
1799 //------------------------------inline_math_native-----------------------------
1800 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1801 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1802 switch (id) {
1803 // These intrinsics are not properly supported on all hardware
1804 case vmIntrinsics::_dsin:
1805 return StubRoutines::dsin() != NULL ?
1806 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1807 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1808 case vmIntrinsics::_dcos:
1809 return StubRoutines::dcos() != NULL ?
1810 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1811 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1812 case vmIntrinsics::_dtan:
1813 return StubRoutines::dtan() != NULL ?
1814 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1815 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1816 case vmIntrinsics::_dlog:
1817 return StubRoutines::dlog() != NULL ?
1818 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1819 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1820 case vmIntrinsics::_dlog10:
1821 return StubRoutines::dlog10() != NULL ?
1822 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1823 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1824
1825 // These intrinsics are supported on all hardware
1826 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1827 case vmIntrinsics::_ceil:
1828 case vmIntrinsics::_floor:
1829 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1830
1831 case vmIntrinsics::_dsqrt:
1832 case vmIntrinsics::_dsqrt_strict:
1833 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1834 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1835 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1836 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1837 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1838
1839 case vmIntrinsics::_dexp:
1840 return StubRoutines::dexp() != NULL ?
1841 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1842 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1843 #undef FN_PTR
1844
1845 case vmIntrinsics::_dpow: return inline_math_pow();
1846 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1847 case vmIntrinsics::_fcopySign: return inline_math(id);
1848 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1849 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1850 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1851
1852 // These intrinsics are not yet correctly implemented
1853 case vmIntrinsics::_datan2:
1854 return false;
1855
1856 default:
1857 fatal_unexpected_iid(id);
1858 return false;
1859 }
1860 }
1861
1862 //----------------------------inline_notify-----------------------------------*
1863 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1864 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1865 address func;
1866 if (id == vmIntrinsics::_notify) {
1867 func = OptoRuntime::monitor_notify_Java();
1868 } else {
1869 func = OptoRuntime::monitor_notifyAll_Java();
1870 }
1871 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1872 make_slow_call_ex(call, env()->Throwable_klass(), false);
1873 return true;
1874 }
1875
1876
1877 //----------------------------inline_min_max-----------------------------------
1878 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1879 set_result(generate_min_max(id, argument(0), argument(1)));
1880 return true;
1881 }
1882
1883 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1884 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1885 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1886 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1887 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1888
1889 {
1890 PreserveJVMState pjvms(this);
1891 PreserveReexecuteState preexecs(this);
1892 jvms()->set_should_reexecute(true);
1893
1894 set_control(slow_path);
1895 set_i_o(i_o());
1896
1897 uncommon_trap(Deoptimization::Reason_intrinsic,
1898 Deoptimization::Action_none);
1899 }
1900
1901 set_control(fast_path);
1902 set_result(math);
1903 }
1904
1905 template <typename OverflowOp>
1906 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1907 typedef typename OverflowOp::MathOp MathOp;
1908
1909 MathOp* mathOp = new MathOp(arg1, arg2);
1910 Node* operation = _gvn.transform( mathOp );
1911 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1912 inline_math_mathExact(operation, ofcheck);
1913 return true;
1914 }
1915
1916 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1917 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1918 }
1919
1920 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1921 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1922 }
1923
1924 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1925 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1926 }
1927
1928 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1929 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1930 }
1931
1932 bool LibraryCallKit::inline_math_negateExactI() {
1933 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1934 }
1935
1936 bool LibraryCallKit::inline_math_negateExactL() {
1937 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1938 }
1939
1940 bool LibraryCallKit::inline_math_multiplyExactI() {
1941 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1942 }
1943
1944 bool LibraryCallKit::inline_math_multiplyExactL() {
1945 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1946 }
1947
1948 bool LibraryCallKit::inline_math_multiplyHigh() {
1949 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1950 return true;
1951 }
1952
1953 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1954 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1955 return true;
1956 }
1957
1958 Node*
1959 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1960 Node* result_val = NULL;
1961 switch (id) {
1962 case vmIntrinsics::_min:
1963 case vmIntrinsics::_min_strict:
1964 result_val = _gvn.transform(new MinINode(x0, y0));
1965 break;
1966 case vmIntrinsics::_max:
1967 case vmIntrinsics::_max_strict:
1968 result_val = _gvn.transform(new MaxINode(x0, y0));
1969 break;
1970 default:
1971 fatal_unexpected_iid(id);
1972 break;
1973 }
1974 return result_val;
1975 }
1976
1977 inline int
1978 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
1979 const TypePtr* base_type = TypePtr::NULL_PTR;
1980 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
1981 if (base_type == NULL) {
1982 // Unknown type.
1983 return Type::AnyPtr;
1984 } else if (base_type == TypePtr::NULL_PTR) {
1985 // Since this is a NULL+long form, we have to switch to a rawptr.
1986 base = _gvn.transform(new CastX2PNode(offset));
1987 offset = MakeConX(0);
1988 return Type::RawPtr;
1989 } else if (base_type->base() == Type::RawPtr) {
1990 return Type::RawPtr;
1991 } else if (base_type->isa_oopptr()) {
1992 // Base is never null => always a heap address.
1993 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
1994 return Type::OopPtr;
1995 }
1996 // Offset is small => always a heap address.
1997 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
1998 if (offset_type != NULL &&
1999 base_type->offset() == 0 && // (should always be?)
2000 offset_type->_lo >= 0 &&
2001 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2002 return Type::OopPtr;
2003 } else if (type == T_OBJECT) {
2004 // off heap access to an oop doesn't make any sense. Has to be on
2005 // heap.
2006 return Type::OopPtr;
2007 }
2008 // Otherwise, it might either be oop+off or NULL+addr.
2009 return Type::AnyPtr;
2010 } else {
2011 // No information:
2012 return Type::AnyPtr;
2013 }
2014 }
2015
2016 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2017 Node* uncasted_base = base;
2018 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2019 if (kind == Type::RawPtr) {
2020 return basic_plus_adr(top(), uncasted_base, offset);
2021 } else if (kind == Type::AnyPtr) {
2022 assert(base == uncasted_base, "unexpected base change");
2023 if (can_cast) {
2024 if (!_gvn.type(base)->speculative_maybe_null() &&
2025 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2026 // According to profiling, this access is always on
2027 // heap. Casting the base to not null and thus avoiding membars
2028 // around the access should allow better optimizations
2029 Node* null_ctl = top();
2030 base = null_check_oop(base, &null_ctl, true, true, true);
2031 assert(null_ctl->is_top(), "no null control here");
2032 return basic_plus_adr(base, offset);
2033 } else if (_gvn.type(base)->speculative_always_null() &&
2034 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2035 // According to profiling, this access is always off
2036 // heap.
2037 base = null_assert(base);
2038 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2039 offset = MakeConX(0);
2040 return basic_plus_adr(top(), raw_base, offset);
2041 }
2042 }
2043 // We don't know if it's an on heap or off heap access. Fall back
2044 // to raw memory access.
2045 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2046 return basic_plus_adr(top(), raw, offset);
2047 } else {
2048 assert(base == uncasted_base, "unexpected base change");
2049 // We know it's an on heap access so base can't be null
2050 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2051 base = must_be_not_null(base, true);
2052 }
2053 return basic_plus_adr(base, offset);
2054 }
2055 }
2056
2057 //--------------------------inline_number_methods-----------------------------
2058 // inline int Integer.numberOfLeadingZeros(int)
2059 // inline int Long.numberOfLeadingZeros(long)
2060 //
2061 // inline int Integer.numberOfTrailingZeros(int)
2062 // inline int Long.numberOfTrailingZeros(long)
2063 //
2064 // inline int Integer.bitCount(int)
2065 // inline int Long.bitCount(long)
2066 //
2067 // inline char Character.reverseBytes(char)
2068 // inline short Short.reverseBytes(short)
2069 // inline int Integer.reverseBytes(int)
2070 // inline long Long.reverseBytes(long)
2071 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2072 Node* arg = argument(0);
2073 Node* n = NULL;
2074 switch (id) {
2075 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2076 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2077 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2078 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2079 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2080 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2081 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2082 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2083 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2084 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2085 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2086 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2087 default: fatal_unexpected_iid(id); break;
2088 }
2089 set_result(_gvn.transform(n));
2090 return true;
2091 }
2092
2093 //--------------------------inline_bitshuffle_methods-----------------------------
2094 // inline int Integer.compress(int, int)
2095 // inline int Integer.expand(int, int)
2096 // inline long Long.compress(long, long)
2097 // inline long Long.expand(long, long)
2098 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2099 Node* n = NULL;
2100 switch (id) {
2101 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2102 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2103 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2104 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2105 default: fatal_unexpected_iid(id); break;
2106 }
2107 set_result(_gvn.transform(n));
2108 return true;
2109 }
2110
2111 //--------------------------inline_number_methods-----------------------------
2112 // inline int Integer.compareUnsigned(int, int)
2113 // inline int Long.compareUnsigned(long, long)
2114 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2115 Node* arg1 = argument(0);
2116 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2117 Node* n = NULL;
2118 switch (id) {
2119 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2120 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2121 default: fatal_unexpected_iid(id); break;
2122 }
2123 set_result(_gvn.transform(n));
2124 return true;
2125 }
2126
2127 //--------------------------inline_unsigned_divmod_methods-----------------------------
2128 // inline int Integer.divideUnsigned(int, int)
2129 // inline int Integer.remainderUnsigned(int, int)
2130 // inline long Long.divideUnsigned(long, long)
2131 // inline long Long.remainderUnsigned(long, long)
2132 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2133 Node* n = NULL;
2134 switch (id) {
2135 case vmIntrinsics::_divideUnsigned_i: {
2136 zero_check_int(argument(1));
2137 // Compile-time detect of null-exception
2138 if (stopped()) {
2139 return true; // keep the graph constructed so far
2140 }
2141 n = new UDivINode(control(), argument(0), argument(1));
2142 break;
2143 }
2144 case vmIntrinsics::_divideUnsigned_l: {
2145 zero_check_long(argument(2));
2146 // Compile-time detect of null-exception
2147 if (stopped()) {
2148 return true; // keep the graph constructed so far
2149 }
2150 n = new UDivLNode(control(), argument(0), argument(2));
2151 break;
2152 }
2153 case vmIntrinsics::_remainderUnsigned_i: {
2154 zero_check_int(argument(1));
2155 // Compile-time detect of null-exception
2156 if (stopped()) {
2157 return true; // keep the graph constructed so far
2158 }
2159 n = new UModINode(control(), argument(0), argument(1));
2160 break;
2161 }
2162 case vmIntrinsics::_remainderUnsigned_l: {
2163 zero_check_long(argument(2));
2164 // Compile-time detect of null-exception
2165 if (stopped()) {
2166 return true; // keep the graph constructed so far
2167 }
2168 n = new UModLNode(control(), argument(0), argument(2));
2169 break;
2170 }
2171 default: fatal_unexpected_iid(id); break;
2172 }
2173 set_result(_gvn.transform(n));
2174 return true;
2175 }
2176
2177 //----------------------------inline_unsafe_access----------------------------
2178
2179 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2180 // Attempt to infer a sharper value type from the offset and base type.
2181 ciKlass* sharpened_klass = NULL;
2182 bool null_free = false;
2183
2184 // See if it is an instance field, with an object type.
2185 if (alias_type->field() != NULL) {
2186 if (alias_type->field()->type()->is_klass()) {
2187 sharpened_klass = alias_type->field()->type()->as_klass();
2188 null_free = alias_type->field()->is_null_free();
2189 }
2190 }
2191
2192 const TypeOopPtr* result = NULL;
2193 // See if it is a narrow oop array.
2194 if (adr_type->isa_aryptr()) {
2195 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2196 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2197 null_free = adr_type->is_aryptr()->is_null_free();
2198 if (elem_type != NULL && elem_type->is_loaded()) {
2199 // Sharpen the value type.
2200 result = elem_type;
2201 }
2202 }
2203 }
2204
2205 // The sharpened class might be unloaded if there is no class loader
2206 // contraint in place.
2207 if (result == NULL && sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2208 // Sharpen the value type.
2209 result = TypeOopPtr::make_from_klass(sharpened_klass);
2210 if (null_free) {
2211 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2212 }
2213 }
2214 if (result != NULL) {
2215 #ifndef PRODUCT
2216 if (C->print_intrinsics() || C->print_inlining()) {
2217 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2218 tty->print(" sharpened value: "); result->dump(); tty->cr();
2219 }
2220 #endif
2221 }
2222 return result;
2223 }
2224
2225 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2226 switch (kind) {
2227 case Relaxed:
2228 return MO_UNORDERED;
2229 case Opaque:
2230 return MO_RELAXED;
2231 case Acquire:
2232 return MO_ACQUIRE;
2233 case Release:
2234 return MO_RELEASE;
2235 case Volatile:
2236 return MO_SEQ_CST;
2237 default:
2238 ShouldNotReachHere();
2239 return 0;
2240 }
2241 }
2242
2243 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2244 if (callee()->is_static()) return false; // caller must have the capability!
2245 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2246 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2247 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2248 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2249
2250 if (is_reference_type(type)) {
2251 decorators |= ON_UNKNOWN_OOP_REF;
2252 }
2253
2254 if (unaligned) {
2255 decorators |= C2_UNALIGNED;
2256 }
2257
2258 #ifndef PRODUCT
2259 {
2260 ResourceMark rm;
2261 // Check the signatures.
2262 ciSignature* sig = callee()->signature();
2263 #ifdef ASSERT
2264 if (!is_store) {
2265 // Object getReference(Object base, int/long offset), etc.
2266 BasicType rtype = sig->return_type()->basic_type();
2267 assert(rtype == type || (rtype == T_OBJECT && type == T_PRIMITIVE_OBJECT), "getter must return the expected value");
2268 assert(sig->count() == 2 || (type == T_PRIMITIVE_OBJECT && sig->count() == 3), "oop getter has 2 or 3 arguments");
2269 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2270 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2271 } else {
2272 // void putReference(Object base, int/long offset, Object x), etc.
2273 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2274 assert(sig->count() == 3 || (type == T_PRIMITIVE_OBJECT && sig->count() == 4), "oop putter has 3 arguments");
2275 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2276 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2277 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2278 assert(vtype == type || (type == T_PRIMITIVE_OBJECT && vtype == T_OBJECT), "putter must accept the expected value");
2279 }
2280 #endif // ASSERT
2281 }
2282 #endif //PRODUCT
2283
2284 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2285
2286 Node* receiver = argument(0); // type: oop
2287
2288 // Build address expression.
2289 Node* heap_base_oop = top();
2290
2291 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2292 Node* base = argument(1); // type: oop
2293 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2294 Node* offset = argument(2); // type: long
2295 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2296 // to be plain byte offsets, which are also the same as those accepted
2297 // by oopDesc::field_addr.
2298 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2299 "fieldOffset must be byte-scaled");
2300
2301 ciInlineKlass* inline_klass = NULL;
2302 if (type == T_PRIMITIVE_OBJECT) {
2303 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2304 if (cls == NULL || cls->const_oop() == NULL) {
2305 return false;
2306 }
2307 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2308 if (!mirror_type->is_inlinetype()) {
2309 return false;
2310 }
2311 inline_klass = mirror_type->as_inline_klass();
2312 }
2313
2314 if (base->is_InlineType()) {
2315 InlineTypeNode* vt = base->as_InlineType();
2316 if (is_store) {
2317 if (!vt->is_allocated(&_gvn)) {
2318 return false;
2319 }
2320 base = vt->get_oop();
2321 } else {
2322 if (offset->is_Con()) {
2323 long off = find_long_con(offset, 0);
2324 ciInlineKlass* vk = vt->type()->inline_klass();
2325 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2326 return false;
2327 }
2328
2329 ciField* field = vk->get_non_flattened_field_by_offset(off);
2330 if (field != NULL) {
2331 BasicType bt = field->layout_type();
2332 if (bt == T_ARRAY || bt == T_NARROWOOP || (bt == T_PRIMITIVE_OBJECT && !field->is_flattened())) {
2333 bt = T_OBJECT;
2334 }
2335 if (bt == type && (bt != T_PRIMITIVE_OBJECT || field->type() == inline_klass)) {
2336 set_result(vt->field_value_by_offset(off, false));
2337 return true;
2338 }
2339 }
2340 }
2341 {
2342 // Re-execute the unsafe access if allocation triggers deoptimization.
2343 PreserveReexecuteState preexecs(this);
2344 jvms()->set_should_reexecute(true);
2345 vt = vt->buffer(this);
2346 }
2347 base = vt->get_oop();
2348 }
2349 }
2350
2351 // 32-bit machines ignore the high half!
2352 offset = ConvL2X(offset);
2353
2354 // Save state and restore on bailout
2355 uint old_sp = sp();
2356 SafePointNode* old_map = clone_map();
2357
2358 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2359
2360 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2361 if (type != T_OBJECT && (inline_klass == NULL || !inline_klass->has_object_fields())) {
2362 decorators |= IN_NATIVE; // off-heap primitive access
2363 } else {
2364 set_map(old_map);
2365 set_sp(old_sp);
2366 return false; // off-heap oop accesses are not supported
2367 }
2368 } else {
2369 heap_base_oop = base; // on-heap or mixed access
2370 }
2371
2372 // Can base be NULL? Otherwise, always on-heap access.
2373 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2374
2375 if (!can_access_non_heap) {
2376 decorators |= IN_HEAP;
2377 }
2378
2379 Node* val = is_store ? argument(4 + (type == T_PRIMITIVE_OBJECT ? 1 : 0)) : NULL;
2380
2381 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2382 if (adr_type == TypePtr::NULL_PTR) {
2383 set_map(old_map);
2384 set_sp(old_sp);
2385 return false; // off-heap access with zero address
2386 }
2387
2388 // Try to categorize the address.
2389 Compile::AliasType* alias_type = C->alias_type(adr_type);
2390 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2391
2392 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2393 alias_type->adr_type() == TypeAryPtr::RANGE) {
2394 set_map(old_map);
2395 set_sp(old_sp);
2396 return false; // not supported
2397 }
2398
2399 bool mismatched = false;
2400 BasicType bt = T_ILLEGAL;
2401 ciField* field = NULL;
2402 if (adr_type->isa_instptr()) {
2403 const TypeInstPtr* instptr = adr_type->is_instptr();
2404 ciInstanceKlass* k = instptr->instance_klass();
2405 int off = instptr->offset();
2406 if (instptr->const_oop() != NULL &&
2407 k == ciEnv::current()->Class_klass() &&
2408 instptr->offset() >= (k->size_helper() * wordSize)) {
2409 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2410 field = k->get_field_by_offset(off, true);
2411 } else {
2412 field = k->get_non_flattened_field_by_offset(off);
2413 }
2414 if (field != NULL) {
2415 bt = field->layout_type();
2416 }
2417 assert(bt == alias_type->basic_type() || bt == T_PRIMITIVE_OBJECT, "should match");
2418 if (field != NULL && bt == T_PRIMITIVE_OBJECT && !field->is_flattened()) {
2419 bt = T_OBJECT;
2420 }
2421 } else {
2422 bt = alias_type->basic_type();
2423 }
2424
2425 if (bt != T_ILLEGAL) {
2426 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2427 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2428 // Alias type doesn't differentiate between byte[] and boolean[]).
2429 // Use address type to get the element type.
2430 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2431 }
2432 if (bt != T_PRIMITIVE_OBJECT && is_reference_type(bt, true)) {
2433 // accessing an array field with getReference is not a mismatch
2434 bt = T_OBJECT;
2435 }
2436 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2437 // Don't intrinsify mismatched object accesses
2438 set_map(old_map);
2439 set_sp(old_sp);
2440 return false;
2441 }
2442 mismatched = (bt != type);
2443 } else if (alias_type->adr_type()->isa_oopptr()) {
2444 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2445 }
2446
2447 if (type == T_PRIMITIVE_OBJECT) {
2448 if (adr_type->isa_instptr()) {
2449 if (field == NULL || field->type() != inline_klass) {
2450 mismatched = true;
2451 }
2452 } else if (adr_type->isa_aryptr()) {
2453 const Type* elem = adr_type->is_aryptr()->elem();
2454 if (!elem->isa_inlinetype()) {
2455 mismatched = true;
2456 } else if (elem->inline_klass() != inline_klass) {
2457 mismatched = true;
2458 }
2459 } else {
2460 mismatched = true;
2461 }
2462 if (is_store) {
2463 const Type* val_t = _gvn.type(val);
2464 if (!(val_t->isa_inlinetype() || val_t->is_inlinetypeptr()) || val_t->inline_klass() != inline_klass) {
2465 set_map(old_map);
2466 set_sp(old_sp);
2467 return false;
2468 }
2469 }
2470 }
2471
2472 old_map->destruct(&_gvn);
2473 assert(!mismatched || type == T_PRIMITIVE_OBJECT || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2474
2475 if (mismatched) {
2476 decorators |= C2_MISMATCHED;
2477 }
2478
2479 // First guess at the value type.
2480 const Type *value_type = Type::get_const_basic_type(type);
2481
2482 // Figure out the memory ordering.
2483 decorators |= mo_decorator_for_access_kind(kind);
2484
2485 if (!is_store) {
2486 if (type == T_OBJECT) {
2487 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2488 if (tjp != NULL) {
2489 value_type = tjp;
2490 }
2491 } else if (type == T_PRIMITIVE_OBJECT) {
2492 value_type = NULL;
2493 }
2494 }
2495
2496 receiver = null_check(receiver);
2497 if (stopped()) {
2498 return true;
2499 }
2500 // Heap pointers get a null-check from the interpreter,
2501 // as a courtesy. However, this is not guaranteed by Unsafe,
2502 // and it is not possible to fully distinguish unintended nulls
2503 // from intended ones in this API.
2504
2505 if (!is_store) {
2506 Node* p = NULL;
2507 // Try to constant fold a load from a constant field
2508
2509 if (heap_base_oop != top() && field != NULL && field->is_constant() && !field->is_flattened() && !mismatched) {
2510 // final or stable field
2511 p = make_constant_from_field(field, heap_base_oop);
2512 }
2513
2514 if (p == NULL) { // Could not constant fold the load
2515 if (type == T_PRIMITIVE_OBJECT) {
2516 if (adr_type->isa_instptr() && !mismatched) {
2517 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2518 int offset = adr_type->is_instptr()->offset();
2519 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, base, holder, offset, decorators);
2520 } else {
2521 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, adr, NULL, 0, decorators);
2522 }
2523 } else {
2524 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2525 const TypeOopPtr* ptr = value_type->make_oopptr();
2526 if (ptr != NULL && ptr->is_inlinetypeptr()) {
2527 // Load a non-flattened inline type from memory
2528 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2529 }
2530 }
2531 // Normalize the value returned by getBoolean in the following cases
2532 if (type == T_BOOLEAN &&
2533 (mismatched ||
2534 heap_base_oop == top() || // - heap_base_oop is NULL or
2535 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2536 // and the unsafe access is made to large offset
2537 // (i.e., larger than the maximum offset necessary for any
2538 // field access)
2539 ) {
2540 IdealKit ideal = IdealKit(this);
2541 #define __ ideal.
2542 IdealVariable normalized_result(ideal);
2543 __ declarations_done();
2544 __ set(normalized_result, p);
2545 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2546 __ set(normalized_result, ideal.ConI(1));
2547 ideal.end_if();
2548 final_sync(ideal);
2549 p = __ value(normalized_result);
2550 #undef __
2551 }
2552 }
2553 if (type == T_ADDRESS) {
2554 p = gvn().transform(new CastP2XNode(NULL, p));
2555 p = ConvX2UL(p);
2556 }
2557 // The load node has the control of the preceding MemBarCPUOrder. All
2558 // following nodes will have the control of the MemBarCPUOrder inserted at
2559 // the end of this method. So, pushing the load onto the stack at a later
2560 // point is fine.
2561 set_result(p);
2562 } else {
2563 if (bt == T_ADDRESS) {
2564 // Repackage the long as a pointer.
2565 val = ConvL2X(val);
2566 val = gvn().transform(new CastX2PNode(val));
2567 }
2568 if (type == T_PRIMITIVE_OBJECT) {
2569 if (adr_type->isa_instptr() && !mismatched) {
2570 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2571 int offset = adr_type->is_instptr()->offset();
2572 val->as_InlineType()->store_flattened(this, base, base, holder, offset, decorators);
2573 } else {
2574 val->as_InlineType()->store_flattened(this, base, adr, NULL, 0, decorators);
2575 }
2576 } else {
2577 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2578 }
2579 }
2580
2581 if (argument(1)->is_InlineType() && is_store) {
2582 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2583 value = value->make_larval(this, false);
2584 replace_in_map(argument(1), value);
2585 }
2586
2587 return true;
2588 }
2589
2590 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2591 Node* receiver = argument(0);
2592 Node* value = argument(1);
2593 if (!value->is_InlineType()) {
2594 return false;
2595 }
2596
2597 receiver = null_check(receiver);
2598 if (stopped()) {
2599 return true;
2600 }
2601
2602 set_result(value->as_InlineType()->make_larval(this, true));
2603 return true;
2604 }
2605
2606 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2607 Node* receiver = argument(0);
2608 Node* buffer = argument(1);
2609 if (!buffer->is_InlineType()) {
2610 return false;
2611 }
2612 InlineTypeNode* vt = buffer->as_InlineType();
2613 if (!vt->is_allocated(&_gvn)) {
2614 return false;
2615 }
2616 // TODO 8239003 Why is this needed?
2617 if (AllocateNode::Ideal_allocation(vt->get_oop(), &_gvn) == NULL) {
2618 return false;
2619 }
2620
2621 receiver = null_check(receiver);
2622 if (stopped()) {
2623 return true;
2624 }
2625
2626 set_result(vt->finish_larval(this));
2627 return true;
2628 }
2629
2630 //----------------------------inline_unsafe_load_store----------------------------
2631 // This method serves a couple of different customers (depending on LoadStoreKind):
2632 //
2633 // LS_cmp_swap:
2634 //
2635 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2636 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2637 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2638 //
2639 // LS_cmp_swap_weak:
2640 //
2641 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2642 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2643 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2644 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2645 //
2646 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2647 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2648 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2649 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2650 //
2651 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2652 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2653 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2654 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2655 //
2656 // LS_cmp_exchange:
2657 //
2658 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2659 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2660 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2661 //
2662 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2663 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2664 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2665 //
2666 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2667 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2668 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2669 //
2670 // LS_get_add:
2671 //
2672 // int getAndAddInt( Object o, long offset, int delta)
2673 // long getAndAddLong(Object o, long offset, long delta)
2674 //
2675 // LS_get_set:
2676 //
2677 // int getAndSet(Object o, long offset, int newValue)
2678 // long getAndSet(Object o, long offset, long newValue)
2679 // Object getAndSet(Object o, long offset, Object newValue)
2680 //
2681 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2682 // This basic scheme here is the same as inline_unsafe_access, but
2683 // differs in enough details that combining them would make the code
2684 // overly confusing. (This is a true fact! I originally combined
2685 // them, but even I was confused by it!) As much code/comments as
2686 // possible are retained from inline_unsafe_access though to make
2687 // the correspondences clearer. - dl
2688
2689 if (callee()->is_static()) return false; // caller must have the capability!
2690
2691 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2692 decorators |= mo_decorator_for_access_kind(access_kind);
2693
2694 #ifndef PRODUCT
2695 BasicType rtype;
2696 {
2697 ResourceMark rm;
2698 // Check the signatures.
2699 ciSignature* sig = callee()->signature();
2700 rtype = sig->return_type()->basic_type();
2701 switch(kind) {
2702 case LS_get_add:
2703 case LS_get_set: {
2704 // Check the signatures.
2705 #ifdef ASSERT
2706 assert(rtype == type, "get and set must return the expected type");
2707 assert(sig->count() == 3, "get and set has 3 arguments");
2708 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2709 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2710 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2711 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2712 #endif // ASSERT
2713 break;
2714 }
2715 case LS_cmp_swap:
2716 case LS_cmp_swap_weak: {
2717 // Check the signatures.
2718 #ifdef ASSERT
2719 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2720 assert(sig->count() == 4, "CAS has 4 arguments");
2721 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2722 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2723 #endif // ASSERT
2724 break;
2725 }
2726 case LS_cmp_exchange: {
2727 // Check the signatures.
2728 #ifdef ASSERT
2729 assert(rtype == type, "CAS must return the expected type");
2730 assert(sig->count() == 4, "CAS has 4 arguments");
2731 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2732 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2733 #endif // ASSERT
2734 break;
2735 }
2736 default:
2737 ShouldNotReachHere();
2738 }
2739 }
2740 #endif //PRODUCT
2741
2742 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2743
2744 // Get arguments:
2745 Node* receiver = NULL;
2746 Node* base = NULL;
2747 Node* offset = NULL;
2748 Node* oldval = NULL;
2749 Node* newval = NULL;
2750 switch(kind) {
2751 case LS_cmp_swap:
2752 case LS_cmp_swap_weak:
2753 case LS_cmp_exchange: {
2754 const bool two_slot_type = type2size[type] == 2;
2755 receiver = argument(0); // type: oop
2756 base = argument(1); // type: oop
2757 offset = argument(2); // type: long
2758 oldval = argument(4); // type: oop, int, or long
2759 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2760 break;
2761 }
2762 case LS_get_add:
2763 case LS_get_set: {
2764 receiver = argument(0); // type: oop
2765 base = argument(1); // type: oop
2766 offset = argument(2); // type: long
2767 oldval = NULL;
2768 newval = argument(4); // type: oop, int, or long
2769 break;
2770 }
2771 default:
2772 ShouldNotReachHere();
2773 }
2774
2775 // Build field offset expression.
2776 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2777 // to be plain byte offsets, which are also the same as those accepted
2778 // by oopDesc::field_addr.
2779 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2780 // 32-bit machines ignore the high half of long offsets
2781 offset = ConvL2X(offset);
2782 // Save state and restore on bailout
2783 uint old_sp = sp();
2784 SafePointNode* old_map = clone_map();
2785 Node* adr = make_unsafe_address(base, offset,type, false);
2786 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2787
2788 Compile::AliasType* alias_type = C->alias_type(adr_type);
2789 BasicType bt = alias_type->basic_type();
2790 if (bt != T_ILLEGAL &&
2791 (is_reference_type(bt) != (type == T_OBJECT))) {
2792 // Don't intrinsify mismatched object accesses.
2793 set_map(old_map);
2794 set_sp(old_sp);
2795 return false;
2796 }
2797
2798 old_map->destruct(&_gvn);
2799
2800 // For CAS, unlike inline_unsafe_access, there seems no point in
2801 // trying to refine types. Just use the coarse types here.
2802 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2803 const Type *value_type = Type::get_const_basic_type(type);
2804
2805 switch (kind) {
2806 case LS_get_set:
2807 case LS_cmp_exchange: {
2808 if (type == T_OBJECT) {
2809 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2810 if (tjp != NULL) {
2811 value_type = tjp;
2812 }
2813 }
2814 break;
2815 }
2816 case LS_cmp_swap:
2817 case LS_cmp_swap_weak:
2818 case LS_get_add:
2819 break;
2820 default:
2821 ShouldNotReachHere();
2822 }
2823
2824 // Null check receiver.
2825 receiver = null_check(receiver);
2826 if (stopped()) {
2827 return true;
2828 }
2829
2830 int alias_idx = C->get_alias_index(adr_type);
2831
2832 if (is_reference_type(type)) {
2833 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2834
2835 if (oldval != NULL && oldval->is_InlineType()) {
2836 // Re-execute the unsafe access if allocation triggers deoptimization.
2837 PreserveReexecuteState preexecs(this);
2838 jvms()->set_should_reexecute(true);
2839 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2840 }
2841 if (newval != NULL && newval->is_InlineType()) {
2842 // Re-execute the unsafe access if allocation triggers deoptimization.
2843 PreserveReexecuteState preexecs(this);
2844 jvms()->set_should_reexecute(true);
2845 newval = newval->as_InlineType()->buffer(this)->get_oop();
2846 }
2847
2848 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2849 // could be delayed during Parse (for example, in adjust_map_after_if()).
2850 // Execute transformation here to avoid barrier generation in such case.
2851 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2852 newval = _gvn.makecon(TypePtr::NULL_PTR);
2853
2854 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2855 // Refine the value to a null constant, when it is known to be null
2856 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2857 }
2858 }
2859
2860 Node* result = NULL;
2861 switch (kind) {
2862 case LS_cmp_exchange: {
2863 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2864 oldval, newval, value_type, type, decorators);
2865 break;
2866 }
2867 case LS_cmp_swap_weak:
2868 decorators |= C2_WEAK_CMPXCHG;
2869 case LS_cmp_swap: {
2870 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2871 oldval, newval, value_type, type, decorators);
2872 break;
2873 }
2874 case LS_get_set: {
2875 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2876 newval, value_type, type, decorators);
2877 break;
2878 }
2879 case LS_get_add: {
2880 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2881 newval, value_type, type, decorators);
2882 break;
2883 }
2884 default:
2885 ShouldNotReachHere();
2886 }
2887
2888 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2889 set_result(result);
2890 return true;
2891 }
2892
2893 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2894 // Regardless of form, don't allow previous ld/st to move down,
2895 // then issue acquire, release, or volatile mem_bar.
2896 insert_mem_bar(Op_MemBarCPUOrder);
2897 switch(id) {
2898 case vmIntrinsics::_loadFence:
2899 insert_mem_bar(Op_LoadFence);
2900 return true;
2901 case vmIntrinsics::_storeFence:
2902 insert_mem_bar(Op_StoreFence);
2903 return true;
2904 case vmIntrinsics::_storeStoreFence:
2905 insert_mem_bar(Op_StoreStoreFence);
2906 return true;
2907 case vmIntrinsics::_fullFence:
2908 insert_mem_bar(Op_MemBarVolatile);
2909 return true;
2910 default:
2911 fatal_unexpected_iid(id);
2912 return false;
2913 }
2914 }
2915
2916 bool LibraryCallKit::inline_onspinwait() {
2917 insert_mem_bar(Op_OnSpinWait);
2918 return true;
2919 }
2920
2921 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2922 if (!kls->is_Con()) {
2923 return true;
2924 }
2925 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2926 if (klsptr == NULL) {
2927 return true;
2928 }
2929 ciInstanceKlass* ik = klsptr->instance_klass();
2930 // don't need a guard for a klass that is already initialized
2931 return !ik->is_initialized();
2932 }
2933
2934 //----------------------------inline_unsafe_writeback0-------------------------
2935 // public native void Unsafe.writeback0(long address)
2936 bool LibraryCallKit::inline_unsafe_writeback0() {
2937 if (!Matcher::has_match_rule(Op_CacheWB)) {
2938 return false;
2939 }
2940 #ifndef PRODUCT
2941 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2942 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2943 ciSignature* sig = callee()->signature();
2944 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2945 #endif
2946 null_check_receiver(); // null-check, then ignore
2947 Node *addr = argument(1);
2948 addr = new CastX2PNode(addr);
2949 addr = _gvn.transform(addr);
2950 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2951 flush = _gvn.transform(flush);
2952 set_memory(flush, TypeRawPtr::BOTTOM);
2953 return true;
2954 }
2955
2956 //----------------------------inline_unsafe_writeback0-------------------------
2957 // public native void Unsafe.writeback0(long address)
2958 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2959 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2960 return false;
2961 }
2962 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2963 return false;
2964 }
2965 #ifndef PRODUCT
2966 assert(Matcher::has_match_rule(Op_CacheWB),
2967 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2968 : "found match rule for CacheWBPostSync but not CacheWB"));
2969
2970 #endif
2971 null_check_receiver(); // null-check, then ignore
2972 Node *sync;
2973 if (is_pre) {
2974 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2975 } else {
2976 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2977 }
2978 sync = _gvn.transform(sync);
2979 set_memory(sync, TypeRawPtr::BOTTOM);
2980 return true;
2981 }
2982
2983 //----------------------------inline_unsafe_allocate---------------------------
2984 // public native Object Unsafe.allocateInstance(Class<?> cls);
2985 bool LibraryCallKit::inline_unsafe_allocate() {
2986 if (callee()->is_static()) return false; // caller must have the capability!
2987
2988 null_check_receiver(); // null-check, then ignore
2989 Node* cls = null_check(argument(1));
2990 if (stopped()) return true;
2991
2992 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2993 kls = null_check(kls);
2994 if (stopped()) return true; // argument was like int.class
2995
2996 Node* test = NULL;
2997 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2998 // Note: The argument might still be an illegal value like
2999 // Serializable.class or Object[].class. The runtime will handle it.
3000 // But we must make an explicit check for initialization.
3001 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3002 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3003 // can generate code to load it as unsigned byte.
3004 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3005 Node* bits = intcon(InstanceKlass::fully_initialized);
3006 test = _gvn.transform(new SubINode(inst, bits));
3007 // The 'test' is non-zero if we need to take a slow path.
3008 }
3009 Node* obj = NULL;
3010 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3011 if (tkls != NULL && tkls->instance_klass()->is_inlinetype()) {
3012 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3013 } else {
3014 obj = new_instance(kls, test);
3015 }
3016 set_result(obj);
3017 return true;
3018 }
3019
3020 //------------------------inline_native_time_funcs--------------
3021 // inline code for System.currentTimeMillis() and System.nanoTime()
3022 // these have the same type and signature
3023 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3024 const TypeFunc* tf = OptoRuntime::void_long_Type();
3025 const TypePtr* no_memory_effects = NULL;
3026 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3027 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3028 #ifdef ASSERT
3029 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3030 assert(value_top == top(), "second value must be top");
3031 #endif
3032 set_result(value);
3033 return true;
3034 }
3035
3036 #ifdef JFR_HAVE_INTRINSICS
3037
3038 /**
3039 * if oop->klass != null
3040 * // normal class
3041 * epoch = _epoch_state ? 2 : 1
3042 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3043 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3044 * }
3045 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3046 * else
3047 * // primitive class
3048 * if oop->array_klass != null
3049 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3050 * else
3051 * id = LAST_TYPE_ID + 1 // void class path
3052 * if (!signaled)
3053 * signaled = true
3054 */
3055 bool LibraryCallKit::inline_native_classID() {
3056 Node* cls = argument(0);
3057
3058 IdealKit ideal(this);
3059 #define __ ideal.
3060 IdealVariable result(ideal); __ declarations_done();
3061 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3062 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3063 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3064
3065
3066 __ if_then(kls, BoolTest::ne, null()); {
3067 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3068 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3069
3070 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3071 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3072 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3073 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3074 mask = _gvn.transform(new OrLNode(mask, epoch));
3075 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3076
3077 float unlikely = PROB_UNLIKELY(0.999);
3078 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3079 sync_kit(ideal);
3080 make_runtime_call(RC_LEAF,
3081 OptoRuntime::class_id_load_barrier_Type(),
3082 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3083 "class id load barrier",
3084 TypePtr::BOTTOM,
3085 kls);
3086 ideal.sync_kit(this);
3087 } __ end_if();
3088
3089 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3090 } __ else_(); {
3091 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3092 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3093 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3094 __ if_then(array_kls, BoolTest::ne, null()); {
3095 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3096 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3097 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3098 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3099 } __ else_(); {
3100 // void class case
3101 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3102 } __ end_if();
3103
3104 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3105 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3106 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3107 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3108 } __ end_if();
3109 } __ end_if();
3110
3111 final_sync(ideal);
3112 set_result(ideal.value(result));
3113 #undef __
3114 return true;
3115 }
3116
3117 /*
3118 * The intrinsic is a model of this pseudo-code:
3119 *
3120 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3121 * jobject h_event_writer = tl->java_event_writer();
3122 * if (h_event_writer == NULL) {
3123 * return NULL;
3124 * }
3125 * oop threadObj = Thread::threadObj();
3126 * oop vthread = java_lang_Thread::vthread(threadObj);
3127 * traceid tid;
3128 * bool excluded;
3129 * if (vthread != threadObj) { // i.e. current thread is virtual
3130 * tid = java_lang_Thread::tid(vthread);
3131 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3132 * excluded = vthread_epoch_raw & excluded_mask;
3133 * if (!excluded) {
3134 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3135 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3136 * if (vthread_epoch != current_epoch) {
3137 * write_checkpoint();
3138 * }
3139 * }
3140 * } else {
3141 * tid = java_lang_Thread::tid(threadObj);
3142 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3143 * excluded = thread_epoch_raw & excluded_mask;
3144 * }
3145 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3146 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3147 * if (tid_in_event_writer != tid) {
3148 * setField(event_writer, "threadID", tid);
3149 * setField(event_writer, "excluded", excluded);
3150 * }
3151 * return event_writer
3152 */
3153 bool LibraryCallKit::inline_native_getEventWriter() {
3154 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3155
3156 // Save input memory and i_o state.
3157 Node* input_memory_state = reset_memory();
3158 set_all_memory(input_memory_state);
3159 Node* input_io_state = i_o();
3160
3161 Node* excluded_mask = _gvn.intcon(32768);
3162 Node* epoch_mask = _gvn.intcon(32767);
3163
3164 // TLS
3165 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3166
3167 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3168 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3169
3170 // Load the eventwriter jobject handle.
3171 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3172
3173 // Null check the jobject handle.
3174 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3175 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3176 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3177
3178 // False path, jobj is null.
3179 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3180
3181 // True path, jobj is not null.
3182 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3183
3184 set_control(jobj_is_not_null);
3185
3186 // Load the threadObj for the CarrierThread.
3187 Node* threadObj = generate_current_thread(tls_ptr);
3188
3189 // Load the vthread.
3190 Node* vthread = generate_virtual_thread(tls_ptr);
3191
3192 // If vthread != threadObj, this is a virtual thread.
3193 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3194 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3195 IfNode* iff_vthread_not_equal_threadObj =
3196 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3197
3198 // False branch, fallback to threadObj.
3199 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3200 set_control(vthread_equal_threadObj);
3201
3202 // Load the tid field from the vthread object.
3203 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3204
3205 // Load the raw epoch value from the threadObj.
3206 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3207 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3208 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3209
3210 // Mask off the excluded information from the epoch.
3211 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3212
3213 // True branch, this is a virtual thread.
3214 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3215 set_control(vthread_not_equal_threadObj);
3216
3217 // Load the tid field from the vthread object.
3218 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3219
3220 // Load the raw epoch value from the vthread.
3221 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3222 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3223 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3224
3225 // Mask off the excluded information from the epoch.
3226 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3227
3228 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3229 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3230 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3231 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3232
3233 // False branch, vthread is excluded, no need to write epoch info.
3234 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3235
3236 // True branch, vthread is included, update epoch info.
3237 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3238 set_control(included);
3239
3240 // Get epoch value.
3241 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3242
3243 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3244 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3245 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3246
3247 // Compare the epoch in the vthread to the current epoch generation.
3248 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3249 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3250 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3251
3252 // False path, epoch is equal, checkpoint information is valid.
3253 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3254
3255 // True path, epoch is not equal, write a checkpoint for the vthread.
3256 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3257
3258 set_control(epoch_is_not_equal);
3259
3260 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3261 // The call also updates the native thread local thread id and the vthread with the current epoch.
3262 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3263 OptoRuntime::jfr_write_checkpoint_Type(),
3264 StubRoutines::jfr_write_checkpoint(),
3265 "write_checkpoint", TypePtr::BOTTOM);
3266 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3267
3268 // vthread epoch != current epoch
3269 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3270 record_for_igvn(epoch_compare_rgn);
3271 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3272 record_for_igvn(epoch_compare_mem);
3273 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3274 record_for_igvn(epoch_compare_io);
3275
3276 // Update control and phi nodes.
3277 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3278 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3279 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3280 epoch_compare_mem->init_req(_false_path, input_memory_state);
3281 epoch_compare_io->init_req(_true_path, i_o());
3282 epoch_compare_io->init_req(_false_path, input_io_state);
3283
3284 // excluded != true
3285 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3286 record_for_igvn(exclude_compare_rgn);
3287 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3288 record_for_igvn(exclude_compare_mem);
3289 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3290 record_for_igvn(exclude_compare_io);
3291
3292 // Update control and phi nodes.
3293 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3294 exclude_compare_rgn->init_req(_false_path, excluded);
3295 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3296 exclude_compare_mem->init_req(_false_path, input_memory_state);
3297 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3298 exclude_compare_io->init_req(_false_path, input_io_state);
3299
3300 // vthread != threadObj
3301 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3302 record_for_igvn(vthread_compare_rgn);
3303 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3304 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3305 record_for_igvn(vthread_compare_io);
3306 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3307 record_for_igvn(tid);
3308 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3309 record_for_igvn(exclusion);
3310
3311 // Update control and phi nodes.
3312 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3313 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3314 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3315 vthread_compare_mem->init_req(_false_path, input_memory_state);
3316 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3317 vthread_compare_io->init_req(_false_path, input_io_state);
3318 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3319 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3320 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3321 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3322
3323 // Update branch state.
3324 set_control(_gvn.transform(vthread_compare_rgn));
3325 set_all_memory(_gvn.transform(vthread_compare_mem));
3326 set_i_o(_gvn.transform(vthread_compare_io));
3327
3328 // Load the event writer oop by dereferencing the jobject handle.
3329 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3330 assert(klass_EventWriter->is_loaded(), "invariant");
3331 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3332 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3333 const TypeOopPtr* const xtype = aklass->as_instance_type();
3334 Node* event_writer = access_load(jobj, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3335
3336 // Load the current thread id from the event writer object.
3337 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3338 // Get the field offset to, conditionally, store an updated tid value later.
3339 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3340 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3341 // Get the field offset to, conditionally, store an updated exclusion value later.
3342 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3343 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3344
3345 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3346 record_for_igvn(event_writer_tid_compare_rgn);
3347 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3348 record_for_igvn(event_writer_tid_compare_mem);
3349 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3350 record_for_igvn(event_writer_tid_compare_io);
3351
3352 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3353 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3354 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3355 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3356
3357 // False path, tids are the same.
3358 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3359
3360 // True path, tid is not equal, need to update the tid in the event writer.
3361 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3362 record_for_igvn(tid_is_not_equal);
3363
3364 // Store the exclusion state to the event writer.
3365 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3366
3367 // Store the tid to the event writer.
3368 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3369
3370 // Update control and phi nodes.
3371 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3372 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3373 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3374 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3375 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3376 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3377
3378 // Result of top level CFG, Memory, IO and Value.
3379 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3380 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3381 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3382 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3383
3384 // Result control.
3385 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3386 result_rgn->init_req(_false_path, jobj_is_null);
3387
3388 // Result memory.
3389 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3390 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3391
3392 // Result IO.
3393 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3394 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3395
3396 // Result value.
3397 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3398 result_value->init_req(_false_path, null()); // return NULL
3399
3400 // Set output state.
3401 set_control(_gvn.transform(result_rgn));
3402 set_all_memory(_gvn.transform(result_mem));
3403 set_i_o(_gvn.transform(result_io));
3404 set_result(result_rgn, result_value);
3405 return true;
3406 }
3407
3408 /*
3409 * The intrinsic is a model of this pseudo-code:
3410 *
3411 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3412 * if (carrierThread != thread) { // is virtual thread
3413 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3414 * bool excluded = vthread_epoch_raw & excluded_mask;
3415 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3416 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3417 * if (!excluded) {
3418 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3419 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3420 * }
3421 * Atomic::release_store(&tl->_vthread, true);
3422 * return;
3423 * }
3424 * Atomic::release_store(&tl->_vthread, false);
3425 */
3426 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3427 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3428
3429 Node* input_memory_state = reset_memory();
3430 set_all_memory(input_memory_state);
3431
3432 Node* excluded_mask = _gvn.intcon(32768);
3433 Node* epoch_mask = _gvn.intcon(32767);
3434
3435 Node* const carrierThread = generate_current_thread(jt);
3436 // If thread != carrierThread, this is a virtual thread.
3437 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3438 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3439 IfNode* iff_thread_not_equal_carrierThread =
3440 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3441
3442 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3443
3444 // False branch, is carrierThread.
3445 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3446 // Store release
3447 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3448
3449 set_all_memory(input_memory_state);
3450
3451 // True branch, is virtual thread.
3452 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3453 set_control(thread_not_equal_carrierThread);
3454
3455 // Load the raw epoch value from the vthread.
3456 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3457 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3458 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3459
3460 // Mask off the excluded information from the epoch.
3461 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3462
3463 // Load the tid field from the thread.
3464 Node* tid = load_field_from_object(thread, "tid", "J");
3465
3466 // Store the vthread tid to the jfr thread local.
3467 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3468 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3469
3470 // Branch is_excluded to conditionalize updating the epoch .
3471 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3472 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3473 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3474
3475 // True branch, vthread is excluded, no need to write epoch info.
3476 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3477 set_control(excluded);
3478 Node* vthread_is_excluded = _gvn.intcon(1);
3479
3480 // False branch, vthread is included, update epoch info.
3481 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3482 set_control(included);
3483 Node* vthread_is_included = _gvn.intcon(0);
3484
3485 // Get epoch value.
3486 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3487
3488 // Store the vthread epoch to the jfr thread local.
3489 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3490 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3491
3492 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3493 record_for_igvn(excluded_rgn);
3494 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3495 record_for_igvn(excluded_mem);
3496 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3497 record_for_igvn(exclusion);
3498
3499 // Merge the excluded control and memory.
3500 excluded_rgn->init_req(_true_path, excluded);
3501 excluded_rgn->init_req(_false_path, included);
3502 excluded_mem->init_req(_true_path, tid_memory);
3503 excluded_mem->init_req(_false_path, included_memory);
3504 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3505 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3506
3507 // Set intermediate state.
3508 set_control(_gvn.transform(excluded_rgn));
3509 set_all_memory(excluded_mem);
3510
3511 // Store the vthread exclusion state to the jfr thread local.
3512 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3513 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3514
3515 // Store release
3516 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3517
3518 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3519 record_for_igvn(thread_compare_rgn);
3520 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3521 record_for_igvn(thread_compare_mem);
3522 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3523 record_for_igvn(vthread);
3524
3525 // Merge the thread_compare control and memory.
3526 thread_compare_rgn->init_req(_true_path, control());
3527 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3528 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3529 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3530
3531 // Set output state.
3532 set_control(_gvn.transform(thread_compare_rgn));
3533 set_all_memory(_gvn.transform(thread_compare_mem));
3534 }
3535
3536 #endif // JFR_HAVE_INTRINSICS
3537
3538 //------------------------inline_native_currentCarrierThread------------------
3539 bool LibraryCallKit::inline_native_currentCarrierThread() {
3540 Node* junk = NULL;
3541 set_result(generate_current_thread(junk));
3542 return true;
3543 }
3544
3545 //------------------------inline_native_currentThread------------------
3546 bool LibraryCallKit::inline_native_currentThread() {
3547 Node* junk = NULL;
3548 set_result(generate_virtual_thread(junk));
3549 return true;
3550 }
3551
3552 //------------------------inline_native_setVthread------------------
3553 bool LibraryCallKit::inline_native_setCurrentThread() {
3554 assert(C->method()->changes_current_thread(),
3555 "method changes current Thread but is not annotated ChangesCurrentThread");
3556 Node* arr = argument(1);
3557 Node* thread = _gvn.transform(new ThreadLocalNode());
3558 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3559 Node* thread_obj_handle
3560 = make_load(NULL, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3561 thread_obj_handle = _gvn.transform(thread_obj_handle);
3562 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3563 // Stores of oops to native memory not supported yet by BarrierSetC2::store_at_resolved
3564 // access_store_at(NULL, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3565 store_to_memory(control(), thread_obj_handle, arr, T_OBJECT, adr_type, MemNode::unordered);
3566 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3567 return true;
3568 }
3569
3570 Node* LibraryCallKit::extentLocalCache_helper() {
3571 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3572 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3573
3574 bool xk = etype->klass_is_exact();
3575
3576 Node* thread = _gvn.transform(new ThreadLocalNode());
3577 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::extentLocalCache_offset()));
3578 return _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
3579 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3580 }
3581
3582 //------------------------inline_native_extentLocalCache------------------
3583 bool LibraryCallKit::inline_native_extentLocalCache() {
3584 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3585 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3586 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3587
3588 // Because we create the extentLocal cache lazily we have to make the
3589 // type of the result BotPTR.
3590 bool xk = etype->klass_is_exact();
3591 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3592 Node* cache_obj_handle = extentLocalCache_helper();
3593 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3594
3595 return true;
3596 }
3597
3598 //------------------------inline_native_setExtentLocalCache------------------
3599 bool LibraryCallKit::inline_native_setExtentLocalCache() {
3600 Node* arr = argument(0);
3601 Node* cache_obj_handle = extentLocalCache_helper();
3602
3603 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3604 store_to_memory(control(), cache_obj_handle, arr, T_OBJECT, adr_type,
3605 MemNode::unordered);
3606
3607 return true;
3608 }
3609
3610 //-----------------------load_klass_from_mirror_common-------------------------
3611 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3612 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3613 // and branch to the given path on the region.
3614 // If never_see_null, take an uncommon trap on null, so we can optimistically
3615 // compile for the non-null case.
3616 // If the region is NULL, force never_see_null = true.
3617 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3618 bool never_see_null,
3619 RegionNode* region,
3620 int null_path,
3621 int offset) {
3622 if (region == NULL) never_see_null = true;
3623 Node* p = basic_plus_adr(mirror, offset);
3624 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3625 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3626 Node* null_ctl = top();
3627 kls = null_check_oop(kls, &null_ctl, never_see_null);
3628 if (region != NULL) {
3629 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3630 region->init_req(null_path, null_ctl);
3631 } else {
3632 assert(null_ctl == top(), "no loose ends");
3633 }
3634 return kls;
3635 }
3636
3637 //--------------------(inline_native_Class_query helpers)---------------------
3638 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3639 // Fall through if (mods & mask) == bits, take the guard otherwise.
3640 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3641 // Branch around if the given klass has the given modifier bit set.
3642 // Like generate_guard, adds a new path onto the region.
3643 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3644 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3645 Node* mask = intcon(modifier_mask);
3646 Node* bits = intcon(modifier_bits);
3647 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3648 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3649 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3650 return generate_fair_guard(bol, region);
3651 }
3652
3653 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3654 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3655 }
3656 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3657 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3658 }
3659
3660 //-------------------------inline_native_Class_query-------------------
3661 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3662 const Type* return_type = TypeInt::BOOL;
3663 Node* prim_return_value = top(); // what happens if it's a primitive class?
3664 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3665 bool expect_prim = false; // most of these guys expect to work on refs
3666
3667 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3668
3669 Node* mirror = argument(0);
3670 Node* obj = top();
3671
3672 switch (id) {
3673 case vmIntrinsics::_isInstance:
3674 // nothing is an instance of a primitive type
3675 prim_return_value = intcon(0);
3676 obj = argument(1);
3677 break;
3678 case vmIntrinsics::_getModifiers:
3679 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3680 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3681 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3682 break;
3683 case vmIntrinsics::_isInterface:
3684 prim_return_value = intcon(0);
3685 break;
3686 case vmIntrinsics::_isArray:
3687 prim_return_value = intcon(0);
3688 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3689 break;
3690 case vmIntrinsics::_isPrimitive:
3691 prim_return_value = intcon(1);
3692 expect_prim = true; // obviously
3693 break;
3694 case vmIntrinsics::_isHidden:
3695 prim_return_value = intcon(0);
3696 break;
3697 case vmIntrinsics::_getSuperclass:
3698 prim_return_value = null();
3699 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3700 break;
3701 case vmIntrinsics::_getClassAccessFlags:
3702 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3703 return_type = TypeInt::INT; // not bool! 6297094
3704 break;
3705 default:
3706 fatal_unexpected_iid(id);
3707 break;
3708 }
3709
3710 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3711 if (mirror_con == NULL) return false; // cannot happen?
3712
3713 #ifndef PRODUCT
3714 if (C->print_intrinsics() || C->print_inlining()) {
3715 ciType* k = mirror_con->java_mirror_type();
3716 if (k) {
3717 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3718 k->print_name();
3719 tty->cr();
3720 }
3721 }
3722 #endif
3723
3724 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3725 RegionNode* region = new RegionNode(PATH_LIMIT);
3726 record_for_igvn(region);
3727 PhiNode* phi = new PhiNode(region, return_type);
3728
3729 // The mirror will never be null of Reflection.getClassAccessFlags, however
3730 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3731 // if it is. See bug 4774291.
3732
3733 // For Reflection.getClassAccessFlags(), the null check occurs in
3734 // the wrong place; see inline_unsafe_access(), above, for a similar
3735 // situation.
3736 mirror = null_check(mirror);
3737 // If mirror or obj is dead, only null-path is taken.
3738 if (stopped()) return true;
3739
3740 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3741
3742 // Now load the mirror's klass metaobject, and null-check it.
3743 // Side-effects region with the control path if the klass is null.
3744 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3745 // If kls is null, we have a primitive mirror.
3746 phi->init_req(_prim_path, prim_return_value);
3747 if (stopped()) { set_result(region, phi); return true; }
3748 bool safe_for_replace = (region->in(_prim_path) == top());
3749
3750 Node* p; // handy temp
3751 Node* null_ctl;
3752
3753 // Now that we have the non-null klass, we can perform the real query.
3754 // For constant classes, the query will constant-fold in LoadNode::Value.
3755 Node* query_value = top();
3756 switch (id) {
3757 case vmIntrinsics::_isInstance:
3758 // nothing is an instance of a primitive type
3759 query_value = gen_instanceof(obj, kls, safe_for_replace);
3760 break;
3761
3762 case vmIntrinsics::_getModifiers:
3763 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3764 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3765 break;
3766
3767 case vmIntrinsics::_isInterface:
3768 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3769 if (generate_interface_guard(kls, region) != NULL)
3770 // A guard was added. If the guard is taken, it was an interface.
3771 phi->add_req(intcon(1));
3772 // If we fall through, it's a plain class.
3773 query_value = intcon(0);
3774 break;
3775
3776 case vmIntrinsics::_isArray:
3777 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3778 if (generate_array_guard(kls, region) != NULL)
3779 // A guard was added. If the guard is taken, it was an array.
3780 phi->add_req(intcon(1));
3781 // If we fall through, it's a plain class.
3782 query_value = intcon(0);
3783 break;
3784
3785 case vmIntrinsics::_isPrimitive:
3786 query_value = intcon(0); // "normal" path produces false
3787 break;
3788
3789 case vmIntrinsics::_isHidden:
3790 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3791 if (generate_hidden_class_guard(kls, region) != NULL)
3792 // A guard was added. If the guard is taken, it was an hidden class.
3793 phi->add_req(intcon(1));
3794 // If we fall through, it's a plain class.
3795 query_value = intcon(0);
3796 break;
3797
3798
3799 case vmIntrinsics::_getSuperclass:
3800 // The rules here are somewhat unfortunate, but we can still do better
3801 // with random logic than with a JNI call.
3802 // Interfaces store null or Object as _super, but must report null.
3803 // Arrays store an intermediate super as _super, but must report Object.
3804 // Other types can report the actual _super.
3805 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3806 if (generate_interface_guard(kls, region) != NULL)
3807 // A guard was added. If the guard is taken, it was an interface.
3808 phi->add_req(null());
3809 if (generate_array_guard(kls, region) != NULL)
3810 // A guard was added. If the guard is taken, it was an array.
3811 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3812 // If we fall through, it's a plain class. Get its _super.
3813 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3814 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3815 null_ctl = top();
3816 kls = null_check_oop(kls, &null_ctl);
3817 if (null_ctl != top()) {
3818 // If the guard is taken, Object.superClass is null (both klass and mirror).
3819 region->add_req(null_ctl);
3820 phi ->add_req(null());
3821 }
3822 if (!stopped()) {
3823 query_value = load_mirror_from_klass(kls);
3824 }
3825 break;
3826
3827 case vmIntrinsics::_getClassAccessFlags:
3828 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3829 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3830 break;
3831
3832 default:
3833 fatal_unexpected_iid(id);
3834 break;
3835 }
3836
3837 // Fall-through is the normal case of a query to a real class.
3838 phi->init_req(1, query_value);
3839 region->init_req(1, control());
3840
3841 C->set_has_split_ifs(true); // Has chance for split-if optimization
3842 set_result(region, phi);
3843 return true;
3844 }
3845
3846 //-------------------------inline_primitive_Class_conversion-------------------
3847 // Class<T> java.lang.Class .asPrimaryType()
3848 // public static Class<T> jdk.internal.value.PrimitiveClass.asPrimaryType(Class<T>)
3849 // Class<T> java.lang.Class .asValueType()
3850 // public static Class<T> jdk.internal.value.PrimitiveClass.asValueType(Class<T>)
3851 bool LibraryCallKit::inline_primitive_Class_conversion(vmIntrinsics::ID id) {
3852 Node* mirror = argument(0); // Receiver/argument Class
3853 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3854 if (mirror_con == NULL) {
3855 return false;
3856 }
3857
3858 bool is_val_mirror = true;
3859 ciType* tm = mirror_con->java_mirror_type(&is_val_mirror);
3860 if (tm != NULL) {
3861 Node* result = mirror;
3862 if ((id == vmIntrinsics::_asPrimaryType || id == vmIntrinsics::_asPrimaryTypeArg) && is_val_mirror) {
3863 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->ref_mirror()));
3864 } else if (id == vmIntrinsics::_asValueType || id == vmIntrinsics::_asValueTypeArg) {
3865 if (!tm->is_inlinetype()) {
3866 return false; // Throw UnsupportedOperationException
3867 } else if (!is_val_mirror) {
3868 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->val_mirror()));
3869 }
3870 }
3871 set_result(result);
3872 return true;
3873 }
3874 return false;
3875 }
3876
3877 //-------------------------inline_Class_cast-------------------
3878 bool LibraryCallKit::inline_Class_cast() {
3879 Node* mirror = argument(0); // Class
3880 Node* obj = argument(1);
3881 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3882 if (mirror_con == NULL) {
3883 return false; // dead path (mirror->is_top()).
3884 }
3885 if (obj == NULL || obj->is_top()) {
3886 return false; // dead path
3887 }
3888 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3889
3890 // First, see if Class.cast() can be folded statically.
3891 // java_mirror_type() returns non-null for compile-time Class constants.
3892 bool requires_null_check = false;
3893 ciType* tm = mirror_con->java_mirror_type(&requires_null_check);
3894 if (tm != NULL && tm->is_klass() &&
3895 tp != NULL) {
3896 if (!tp->is_loaded()) {
3897 // Don't use intrinsic when class is not loaded.
3898 return false;
3899 } else {
3900 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass()), tp->as_klass_type());
3901 if (static_res == Compile::SSC_always_true) {
3902 // isInstance() is true - fold the code.
3903 if (requires_null_check) {
3904 obj = null_check(obj);
3905 }
3906 set_result(obj);
3907 return true;
3908 } else if (static_res == Compile::SSC_always_false) {
3909 // Don't use intrinsic, have to throw ClassCastException.
3910 // If the reference is null, the non-intrinsic bytecode will
3911 // be optimized appropriately.
3912 return false;
3913 }
3914 }
3915 }
3916
3917 // Bailout intrinsic and do normal inlining if exception path is frequent.
3918 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3919 return false;
3920 }
3921
3922 // Generate dynamic checks.
3923 // Class.cast() is java implementation of _checkcast bytecode.
3924 // Do checkcast (Parse::do_checkcast()) optimizations here.
3925
3926 if (requires_null_check) {
3927 obj = null_check(obj);
3928 }
3929 mirror = null_check(mirror);
3930 // If mirror is dead, only null-path is taken.
3931 if (stopped()) {
3932 return true;
3933 }
3934
3935 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3936 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
3937 RegionNode* region = new RegionNode(PATH_LIMIT);
3938 record_for_igvn(region);
3939
3940 // Now load the mirror's klass metaobject, and null-check it.
3941 // If kls is null, we have a primitive mirror and
3942 // nothing is an instance of a primitive type.
3943 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3944
3945 Node* res = top();
3946 Node* io = i_o();
3947 Node* mem = merged_memory();
3948 if (!stopped()) {
3949 if (EnableValhalla && !requires_null_check) {
3950 // Check if we are casting to QMyValue
3951 Node* ctrl_val_mirror = generate_fair_guard(is_val_mirror(mirror), NULL);
3952 if (ctrl_val_mirror != NULL) {
3953 RegionNode* r = new RegionNode(3);
3954 record_for_igvn(r);
3955 r->init_req(1, control());
3956
3957 // Casting to QMyValue, check for null
3958 set_control(ctrl_val_mirror);
3959 { // PreserveJVMState because null check replaces obj in map
3960 PreserveJVMState pjvms(this);
3961 Node* null_ctr = top();
3962 null_check_oop(obj, &null_ctr);
3963 region->init_req(_npe_path, null_ctr);
3964 r->init_req(2, control());
3965 }
3966 set_control(_gvn.transform(r));
3967 }
3968 }
3969
3970 Node* bad_type_ctrl = top();
3971 // Do checkcast optimizations.
3972 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3973 region->init_req(_bad_type_path, bad_type_ctrl);
3974 }
3975 if (region->in(_prim_path) != top() ||
3976 region->in(_bad_type_path) != top() ||
3977 region->in(_npe_path) != top()) {
3978 // Let Interpreter throw ClassCastException.
3979 PreserveJVMState pjvms(this);
3980 set_control(_gvn.transform(region));
3981 // Set IO and memory because gen_checkcast may override them when buffering inline types
3982 set_i_o(io);
3983 set_all_memory(mem);
3984 uncommon_trap(Deoptimization::Reason_intrinsic,
3985 Deoptimization::Action_maybe_recompile);
3986 }
3987 if (!stopped()) {
3988 set_result(res);
3989 }
3990 return true;
3991 }
3992
3993
3994 //--------------------------inline_native_subtype_check------------------------
3995 // This intrinsic takes the JNI calls out of the heart of
3996 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3997 bool LibraryCallKit::inline_native_subtype_check() {
3998 // Pull both arguments off the stack.
3999 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4000 args[0] = argument(0);
4001 args[1] = argument(1);
4002 Node* klasses[2]; // corresponding Klasses: superk, subk
4003 klasses[0] = klasses[1] = top();
4004
4005 enum {
4006 // A full decision tree on {superc is prim, subc is prim}:
4007 _prim_0_path = 1, // {P,N} => false
4008 // {P,P} & superc!=subc => false
4009 _prim_same_path, // {P,P} & superc==subc => true
4010 _prim_1_path, // {N,P} => false
4011 _ref_subtype_path, // {N,N} & subtype check wins => true
4012 _both_ref_path, // {N,N} & subtype check loses => false
4013 PATH_LIMIT
4014 };
4015
4016 RegionNode* region = new RegionNode(PATH_LIMIT);
4017 RegionNode* prim_region = new RegionNode(2);
4018 Node* phi = new PhiNode(region, TypeInt::BOOL);
4019 record_for_igvn(region);
4020 record_for_igvn(prim_region);
4021
4022 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4023 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4024 int class_klass_offset = java_lang_Class::klass_offset();
4025
4026 // First null-check both mirrors and load each mirror's klass metaobject.
4027 int which_arg;
4028 for (which_arg = 0; which_arg <= 1; which_arg++) {
4029 Node* arg = args[which_arg];
4030 arg = null_check(arg);
4031 if (stopped()) break;
4032 args[which_arg] = arg;
4033
4034 Node* p = basic_plus_adr(arg, class_klass_offset);
4035 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
4036 klasses[which_arg] = _gvn.transform(kls);
4037 }
4038
4039 // Having loaded both klasses, test each for null.
4040 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4041 for (which_arg = 0; which_arg <= 1; which_arg++) {
4042 Node* kls = klasses[which_arg];
4043 Node* null_ctl = top();
4044 kls = null_check_oop(kls, &null_ctl, never_see_null);
4045 if (which_arg == 0) {
4046 prim_region->init_req(1, null_ctl);
4047 } else {
4048 region->init_req(_prim_1_path, null_ctl);
4049 }
4050 if (stopped()) break;
4051 klasses[which_arg] = kls;
4052 }
4053
4054 if (!stopped()) {
4055 // now we have two reference types, in klasses[0..1]
4056 Node* subk = klasses[1]; // the argument to isAssignableFrom
4057 Node* superk = klasses[0]; // the receiver
4058 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4059 // If superc is an inline mirror, we also need to check if superc == subc because LMyValue
4060 // is not a subtype of QMyValue but due to subk == superk the subtype check will pass.
4061 generate_fair_guard(is_val_mirror(args[0]), prim_region);
4062 // now we have a successful reference subtype check
4063 region->set_req(_ref_subtype_path, control());
4064 }
4065
4066 // If both operands are primitive (both klasses null), then
4067 // we must return true when they are identical primitives.
4068 // It is convenient to test this after the first null klass check.
4069 // This path is also used if superc is a value mirror.
4070 set_control(_gvn.transform(prim_region));
4071 if (!stopped()) {
4072 // Since superc is primitive, make a guard for the superc==subc case.
4073 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4074 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4075 generate_fair_guard(bol_eq, region);
4076 if (region->req() == PATH_LIMIT+1) {
4077 // A guard was added. If the added guard is taken, superc==subc.
4078 region->swap_edges(PATH_LIMIT, _prim_same_path);
4079 region->del_req(PATH_LIMIT);
4080 }
4081 region->set_req(_prim_0_path, control()); // Not equal after all.
4082 }
4083
4084 // these are the only paths that produce 'true':
4085 phi->set_req(_prim_same_path, intcon(1));
4086 phi->set_req(_ref_subtype_path, intcon(1));
4087
4088 // pull together the cases:
4089 assert(region->req() == PATH_LIMIT, "sane region");
4090 for (uint i = 1; i < region->req(); i++) {
4091 Node* ctl = region->in(i);
4092 if (ctl == NULL || ctl == top()) {
4093 region->set_req(i, top());
4094 phi ->set_req(i, top());
4095 } else if (phi->in(i) == NULL) {
4096 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4097 }
4098 }
4099
4100 set_control(_gvn.transform(region));
4101 set_result(_gvn.transform(phi));
4102 return true;
4103 }
4104
4105 //---------------------generate_array_guard_common------------------------
4106 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4107
4108 if (stopped()) {
4109 return NULL;
4110 }
4111
4112 // Like generate_guard, adds a new path onto the region.
4113 jint layout_con = 0;
4114 Node* layout_val = get_layout_helper(kls, layout_con);
4115 if (layout_val == NULL) {
4116 bool query = 0;
4117 switch(kind) {
4118 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4119 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4120 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4121 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4122 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4123 default:
4124 ShouldNotReachHere();
4125 }
4126 if (!query) {
4127 return NULL; // never a branch
4128 } else { // always a branch
4129 Node* always_branch = control();
4130 if (region != NULL)
4131 region->add_req(always_branch);
4132 set_control(top());
4133 return always_branch;
4134 }
4135 }
4136 unsigned int value = 0;
4137 BoolTest::mask btest = BoolTest::illegal;
4138 switch(kind) {
4139 case ObjectArray:
4140 case NonObjectArray: {
4141 value = Klass::_lh_array_tag_obj_value;
4142 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4143 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4144 break;
4145 }
4146 case TypeArray: {
4147 value = Klass::_lh_array_tag_type_value;
4148 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4149 btest = BoolTest::eq;
4150 break;
4151 }
4152 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4153 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4154 default:
4155 ShouldNotReachHere();
4156 }
4157 // Now test the correct condition.
4158 jint nval = (jint)value;
4159 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4160 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4161 return generate_fair_guard(bol, region);
4162 }
4163
4164
4165 //-----------------------inline_native_newArray--------------------------
4166 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4167 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4168 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4169 Node* mirror;
4170 Node* count_val;
4171 if (uninitialized) {
4172 mirror = argument(1);
4173 count_val = argument(2);
4174 } else {
4175 mirror = argument(0);
4176 count_val = argument(1);
4177 }
4178
4179 mirror = null_check(mirror);
4180 // If mirror or obj is dead, only null-path is taken.
4181 if (stopped()) return true;
4182
4183 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4184 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4185 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4186 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4187 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4188
4189 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4190 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4191 result_reg, _slow_path);
4192 Node* normal_ctl = control();
4193 Node* no_array_ctl = result_reg->in(_slow_path);
4194
4195 // Generate code for the slow case. We make a call to newArray().
4196 set_control(no_array_ctl);
4197 if (!stopped()) {
4198 // Either the input type is void.class, or else the
4199 // array klass has not yet been cached. Either the
4200 // ensuing call will throw an exception, or else it
4201 // will cache the array klass for next time.
4202 PreserveJVMState pjvms(this);
4203 CallJavaNode* slow_call = NULL;
4204 if (uninitialized) {
4205 // Generate optimized virtual call (holder class 'Unsafe' is final)
4206 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
4207 } else {
4208 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
4209 }
4210 Node* slow_result = set_results_for_java_call(slow_call);
4211 // this->control() comes from set_results_for_java_call
4212 result_reg->set_req(_slow_path, control());
4213 result_val->set_req(_slow_path, slow_result);
4214 result_io ->set_req(_slow_path, i_o());
4215 result_mem->set_req(_slow_path, reset_memory());
4216 }
4217
4218 set_control(normal_ctl);
4219 if (!stopped()) {
4220 // Normal case: The array type has been cached in the java.lang.Class.
4221 // The following call works fine even if the array type is polymorphic.
4222 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4223 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4224 result_reg->init_req(_normal_path, control());
4225 result_val->init_req(_normal_path, obj);
4226 result_io ->init_req(_normal_path, i_o());
4227 result_mem->init_req(_normal_path, reset_memory());
4228
4229 if (uninitialized) {
4230 // Mark the allocation so that zeroing is skipped
4231 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
4232 alloc->maybe_set_complete(&_gvn);
4233 }
4234 }
4235
4236 // Return the combined state.
4237 set_i_o( _gvn.transform(result_io) );
4238 set_all_memory( _gvn.transform(result_mem));
4239
4240 C->set_has_split_ifs(true); // Has chance for split-if optimization
4241 set_result(result_reg, result_val);
4242 return true;
4243 }
4244
4245 //----------------------inline_native_getLength--------------------------
4246 // public static native int java.lang.reflect.Array.getLength(Object array);
4247 bool LibraryCallKit::inline_native_getLength() {
4248 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4249
4250 Node* array = null_check(argument(0));
4251 // If array is dead, only null-path is taken.
4252 if (stopped()) return true;
4253
4254 // Deoptimize if it is a non-array.
4255 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
4256
4257 if (non_array != NULL) {
4258 PreserveJVMState pjvms(this);
4259 set_control(non_array);
4260 uncommon_trap(Deoptimization::Reason_intrinsic,
4261 Deoptimization::Action_maybe_recompile);
4262 }
4263
4264 // If control is dead, only non-array-path is taken.
4265 if (stopped()) return true;
4266
4267 // The works fine even if the array type is polymorphic.
4268 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4269 Node* result = load_array_length(array);
4270
4271 C->set_has_split_ifs(true); // Has chance for split-if optimization
4272 set_result(result);
4273 return true;
4274 }
4275
4276 //------------------------inline_array_copyOf----------------------------
4277 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4278 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4279 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4280 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4281
4282 // Get the arguments.
4283 Node* original = argument(0);
4284 Node* start = is_copyOfRange? argument(1): intcon(0);
4285 Node* end = is_copyOfRange? argument(2): argument(1);
4286 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4287
4288 Node* newcopy = NULL;
4289
4290 // Set the original stack and the reexecute bit for the interpreter to reexecute
4291 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4292 { PreserveReexecuteState preexecs(this);
4293 jvms()->set_should_reexecute(true);
4294
4295 array_type_mirror = null_check(array_type_mirror);
4296 original = null_check(original);
4297
4298 // Check if a null path was taken unconditionally.
4299 if (stopped()) return true;
4300
4301 Node* orig_length = load_array_length(original);
4302
4303 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4304 klass_node = null_check(klass_node);
4305
4306 RegionNode* bailout = new RegionNode(1);
4307 record_for_igvn(bailout);
4308
4309 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4310 // Bail out if that is so.
4311 // Inline type array may have object field that would require a
4312 // write barrier. Conservatively, go to slow path.
4313 // TODO 8251971: Optimize for the case when flat src/dst are later found
4314 // to not contain oops (i.e., move this check to the macro expansion phase).
4315 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4316 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4317 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4318 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4319 // Can src array be flat and contain oops?
4320 (orig_t == NULL || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4321 // Can dest array be flat and contain oops?
4322 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4323 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4324 if (not_objArray != NULL) {
4325 // Improve the klass node's type from the new optimistic assumption:
4326 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4327 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4328 Node* cast = new CastPPNode(klass_node, akls);
4329 cast->init_req(0, control());
4330 klass_node = _gvn.transform(cast);
4331 }
4332
4333 // Bail out if either start or end is negative.
4334 generate_negative_guard(start, bailout, &start);
4335 generate_negative_guard(end, bailout, &end);
4336
4337 Node* length = end;
4338 if (_gvn.type(start) != TypeInt::ZERO) {
4339 length = _gvn.transform(new SubINode(end, start));
4340 }
4341
4342 // Bail out if length is negative.
4343 // Without this the new_array would throw
4344 // NegativeArraySizeException but IllegalArgumentException is what
4345 // should be thrown
4346 generate_negative_guard(length, bailout, &length);
4347
4348 // Handle inline type arrays
4349 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4350 if (!stopped()) {
4351 orig_t = _gvn.type(original)->isa_aryptr();
4352 if (orig_t != NULL && orig_t->is_flat()) {
4353 // Src is flat, check that dest is flat as well
4354 if (exclude_flat) {
4355 // Dest can't be flat, bail out
4356 bailout->add_req(control());
4357 set_control(top());
4358 } else {
4359 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4360 }
4361 } else if (UseFlatArray && (orig_t == NULL || !orig_t->is_not_flat()) &&
4362 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4363 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4364 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4365 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4366 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4367 if (orig_t != NULL) {
4368 orig_t = orig_t->cast_to_not_flat();
4369 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4370 }
4371 }
4372 if (!can_validate) {
4373 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4374 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4375 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4376 generate_fair_guard(null_free_array_test(klass_node), bailout);
4377 }
4378 }
4379
4380 if (bailout->req() > 1) {
4381 PreserveJVMState pjvms(this);
4382 set_control(_gvn.transform(bailout));
4383 uncommon_trap(Deoptimization::Reason_intrinsic,
4384 Deoptimization::Action_maybe_recompile);
4385 }
4386
4387 if (!stopped()) {
4388 // How many elements will we copy from the original?
4389 // The answer is MinI(orig_length - start, length).
4390 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4391 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4392
4393 // Generate a direct call to the right arraycopy function(s).
4394 // We know the copy is disjoint but we might not know if the
4395 // oop stores need checking.
4396 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4397 // This will fail a store-check if x contains any non-nulls.
4398
4399 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4400 // loads/stores but it is legal only if we're sure the
4401 // Arrays.copyOf would succeed. So we need all input arguments
4402 // to the copyOf to be validated, including that the copy to the
4403 // new array won't trigger an ArrayStoreException. That subtype
4404 // check can be optimized if we know something on the type of
4405 // the input array from type speculation.
4406 if (_gvn.type(klass_node)->singleton()) {
4407 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4408 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4409
4410 int test = C->static_subtype_check(superk, subk);
4411 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4412 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4413 if (t_original->speculative_type() != NULL) {
4414 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4415 }
4416 }
4417 }
4418
4419 bool validated = false;
4420 // Reason_class_check rather than Reason_intrinsic because we
4421 // want to intrinsify even if this traps.
4422 if (can_validate) {
4423 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4424
4425 if (not_subtype_ctrl != top()) {
4426 PreserveJVMState pjvms(this);
4427 set_control(not_subtype_ctrl);
4428 uncommon_trap(Deoptimization::Reason_class_check,
4429 Deoptimization::Action_make_not_entrant);
4430 assert(stopped(), "Should be stopped");
4431 }
4432 validated = true;
4433 }
4434
4435 if (!stopped()) {
4436 newcopy = new_array(klass_node, length, 0); // no arguments to push
4437
4438 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4439 load_object_klass(original), klass_node);
4440 if (!is_copyOfRange) {
4441 ac->set_copyof(validated);
4442 } else {
4443 ac->set_copyofrange(validated);
4444 }
4445 Node* n = _gvn.transform(ac);
4446 if (n == ac) {
4447 ac->connect_outputs(this);
4448 } else {
4449 assert(validated, "shouldn't transform if all arguments not validated");
4450 set_all_memory(n);
4451 }
4452 }
4453 }
4454 } // original reexecute is set back here
4455
4456 C->set_has_split_ifs(true); // Has chance for split-if optimization
4457 if (!stopped()) {
4458 set_result(newcopy);
4459 }
4460 return true;
4461 }
4462
4463
4464 //----------------------generate_virtual_guard---------------------------
4465 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4466 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4467 RegionNode* slow_region) {
4468 ciMethod* method = callee();
4469 int vtable_index = method->vtable_index();
4470 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4471 "bad index %d", vtable_index);
4472 // Get the Method* out of the appropriate vtable entry.
4473 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4474 vtable_index*vtableEntry::size_in_bytes() +
4475 vtableEntry::method_offset_in_bytes();
4476 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4477 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4478
4479 // Compare the target method with the expected method (e.g., Object.hashCode).
4480 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4481
4482 Node* native_call = makecon(native_call_addr);
4483 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4484 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4485
4486 return generate_slow_guard(test_native, slow_region);
4487 }
4488
4489 //-----------------------generate_method_call----------------------------
4490 // Use generate_method_call to make a slow-call to the real
4491 // method if the fast path fails. An alternative would be to
4492 // use a stub like OptoRuntime::slow_arraycopy_Java.
4493 // This only works for expanding the current library call,
4494 // not another intrinsic. (E.g., don't use this for making an
4495 // arraycopy call inside of the copyOf intrinsic.)
4496 CallJavaNode*
4497 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4498 // When compiling the intrinsic method itself, do not use this technique.
4499 guarantee(callee() != C->method(), "cannot make slow-call to self");
4500
4501 ciMethod* method = callee();
4502 // ensure the JVMS we have will be correct for this call
4503 guarantee(method_id == method->intrinsic_id(), "must match");
4504
4505 const TypeFunc* tf = TypeFunc::make(method);
4506 CallJavaNode* slow_call;
4507 if (is_static) {
4508 assert(!is_virtual, "");
4509 slow_call = new CallStaticJavaNode(C, tf,
4510 SharedRuntime::get_resolve_static_call_stub(), method);
4511 } else if (is_virtual) {
4512 null_check_receiver();
4513 int vtable_index = Method::invalid_vtable_index;
4514 if (UseInlineCaches) {
4515 // Suppress the vtable call
4516 } else {
4517 // hashCode and clone are not a miranda methods,
4518 // so the vtable index is fixed.
4519 // No need to use the linkResolver to get it.
4520 vtable_index = method->vtable_index();
4521 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4522 "bad index %d", vtable_index);
4523 }
4524 slow_call = new CallDynamicJavaNode(tf,
4525 SharedRuntime::get_resolve_virtual_call_stub(),
4526 method, vtable_index);
4527 } else { // neither virtual nor static: opt_virtual
4528 null_check_receiver();
4529 slow_call = new CallStaticJavaNode(C, tf,
4530 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4531 slow_call->set_optimized_virtual(true);
4532 }
4533 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4534 // To be able to issue a direct call (optimized virtual or virtual)
4535 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4536 // about the method being invoked should be attached to the call site to
4537 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4538 slow_call->set_override_symbolic_info(true);
4539 }
4540 set_arguments_for_java_call(slow_call);
4541 set_edges_for_java_call(slow_call);
4542 return slow_call;
4543 }
4544
4545
4546 /**
4547 * Build special case code for calls to hashCode on an object. This call may
4548 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4549 * slightly different code.
4550 */
4551 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4552 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4553 assert(!(is_virtual && is_static), "either virtual, special, or static");
4554
4555 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4556
4557 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4558 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4559 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4560 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4561 Node* obj = argument(0);
4562
4563 if (gvn().type(obj)->is_inlinetypeptr()) {
4564 return false;
4565 }
4566
4567 if (!is_static) {
4568 // Check for hashing null object
4569 obj = null_check_receiver();
4570 if (stopped()) return true; // unconditionally null
4571 result_reg->init_req(_null_path, top());
4572 result_val->init_req(_null_path, top());
4573 } else {
4574 // Do a null check, and return zero if null.
4575 // System.identityHashCode(null) == 0
4576 Node* null_ctl = top();
4577 obj = null_check_oop(obj, &null_ctl);
4578 result_reg->init_req(_null_path, null_ctl);
4579 result_val->init_req(_null_path, _gvn.intcon(0));
4580 }
4581
4582 // Unconditionally null? Then return right away.
4583 if (stopped()) {
4584 set_control( result_reg->in(_null_path));
4585 if (!stopped())
4586 set_result(result_val->in(_null_path));
4587 return true;
4588 }
4589
4590 // We only go to the fast case code if we pass a number of guards. The
4591 // paths which do not pass are accumulated in the slow_region.
4592 RegionNode* slow_region = new RegionNode(1);
4593 record_for_igvn(slow_region);
4594
4595 // If this is a virtual call, we generate a funny guard. We pull out
4596 // the vtable entry corresponding to hashCode() from the target object.
4597 // If the target method which we are calling happens to be the native
4598 // Object hashCode() method, we pass the guard. We do not need this
4599 // guard for non-virtual calls -- the caller is known to be the native
4600 // Object hashCode().
4601 if (is_virtual) {
4602 // After null check, get the object's klass.
4603 Node* obj_klass = load_object_klass(obj);
4604 generate_virtual_guard(obj_klass, slow_region);
4605 }
4606
4607 // Get the header out of the object, use LoadMarkNode when available
4608 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4609 // The control of the load must be NULL. Otherwise, the load can move before
4610 // the null check after castPP removal.
4611 Node* no_ctrl = NULL;
4612 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4613
4614 // Test the header to see if it is unlocked.
4615 // This also serves as guard against inline types
4616 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4617 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4618 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4619 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4620 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4621
4622 generate_slow_guard(test_unlocked, slow_region);
4623
4624 // Get the hash value and check to see that it has been properly assigned.
4625 // We depend on hash_mask being at most 32 bits and avoid the use of
4626 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4627 // vm: see markWord.hpp.
4628 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4629 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4630 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4631 // This hack lets the hash bits live anywhere in the mark object now, as long
4632 // as the shift drops the relevant bits into the low 32 bits. Note that
4633 // Java spec says that HashCode is an int so there's no point in capturing
4634 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4635 hshifted_header = ConvX2I(hshifted_header);
4636 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4637
4638 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4639 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4640 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4641
4642 generate_slow_guard(test_assigned, slow_region);
4643
4644 Node* init_mem = reset_memory();
4645 // fill in the rest of the null path:
4646 result_io ->init_req(_null_path, i_o());
4647 result_mem->init_req(_null_path, init_mem);
4648
4649 result_val->init_req(_fast_path, hash_val);
4650 result_reg->init_req(_fast_path, control());
4651 result_io ->init_req(_fast_path, i_o());
4652 result_mem->init_req(_fast_path, init_mem);
4653
4654 // Generate code for the slow case. We make a call to hashCode().
4655 set_control(_gvn.transform(slow_region));
4656 if (!stopped()) {
4657 // No need for PreserveJVMState, because we're using up the present state.
4658 set_all_memory(init_mem);
4659 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4660 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4661 Node* slow_result = set_results_for_java_call(slow_call);
4662 // this->control() comes from set_results_for_java_call
4663 result_reg->init_req(_slow_path, control());
4664 result_val->init_req(_slow_path, slow_result);
4665 result_io ->set_req(_slow_path, i_o());
4666 result_mem ->set_req(_slow_path, reset_memory());
4667 }
4668
4669 // Return the combined state.
4670 set_i_o( _gvn.transform(result_io) );
4671 set_all_memory( _gvn.transform(result_mem));
4672
4673 set_result(result_reg, result_val);
4674 return true;
4675 }
4676
4677 //---------------------------inline_native_getClass----------------------------
4678 // public final native Class<?> java.lang.Object.getClass();
4679 //
4680 // Build special case code for calls to getClass on an object.
4681 bool LibraryCallKit::inline_native_getClass() {
4682 Node* obj = argument(0);
4683 if (obj->is_InlineType()) {
4684 const Type* t = _gvn.type(obj);
4685 if (t->maybe_null()) {
4686 null_check(obj);
4687 }
4688 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4689 return true;
4690 }
4691 obj = null_check_receiver();
4692 if (stopped()) return true;
4693 set_result(load_mirror_from_klass(load_object_klass(obj)));
4694 return true;
4695 }
4696
4697 //-----------------inline_native_Reflection_getCallerClass---------------------
4698 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4699 //
4700 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4701 //
4702 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4703 // in that it must skip particular security frames and checks for
4704 // caller sensitive methods.
4705 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4706 #ifndef PRODUCT
4707 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4708 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4709 }
4710 #endif
4711
4712 if (!jvms()->has_method()) {
4713 #ifndef PRODUCT
4714 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4715 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4716 }
4717 #endif
4718 return false;
4719 }
4720
4721 // Walk back up the JVM state to find the caller at the required
4722 // depth.
4723 JVMState* caller_jvms = jvms();
4724
4725 // Cf. JVM_GetCallerClass
4726 // NOTE: Start the loop at depth 1 because the current JVM state does
4727 // not include the Reflection.getCallerClass() frame.
4728 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4729 ciMethod* m = caller_jvms->method();
4730 switch (n) {
4731 case 0:
4732 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4733 break;
4734 case 1:
4735 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4736 if (!m->caller_sensitive()) {
4737 #ifndef PRODUCT
4738 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4739 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4740 }
4741 #endif
4742 return false; // bail-out; let JVM_GetCallerClass do the work
4743 }
4744 break;
4745 default:
4746 if (!m->is_ignored_by_security_stack_walk()) {
4747 // We have reached the desired frame; return the holder class.
4748 // Acquire method holder as java.lang.Class and push as constant.
4749 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4750 ciInstance* caller_mirror = caller_klass->java_mirror();
4751 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4752
4753 #ifndef PRODUCT
4754 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4755 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());
4756 tty->print_cr(" JVM state at this point:");
4757 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4758 ciMethod* m = jvms()->of_depth(i)->method();
4759 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4760 }
4761 }
4762 #endif
4763 return true;
4764 }
4765 break;
4766 }
4767 }
4768
4769 #ifndef PRODUCT
4770 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4771 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4772 tty->print_cr(" JVM state at this point:");
4773 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4774 ciMethod* m = jvms()->of_depth(i)->method();
4775 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4776 }
4777 }
4778 #endif
4779
4780 return false; // bail-out; let JVM_GetCallerClass do the work
4781 }
4782
4783 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4784 Node* arg = argument(0);
4785 Node* result = NULL;
4786
4787 switch (id) {
4788 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4789 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4790 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4791 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4792 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4793 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4794
4795 case vmIntrinsics::_doubleToLongBits: {
4796 // two paths (plus control) merge in a wood
4797 RegionNode *r = new RegionNode(3);
4798 Node *phi = new PhiNode(r, TypeLong::LONG);
4799
4800 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4801 // Build the boolean node
4802 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4803
4804 // Branch either way.
4805 // NaN case is less traveled, which makes all the difference.
4806 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4807 Node *opt_isnan = _gvn.transform(ifisnan);
4808 assert( opt_isnan->is_If(), "Expect an IfNode");
4809 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4810 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4811
4812 set_control(iftrue);
4813
4814 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4815 Node *slow_result = longcon(nan_bits); // return NaN
4816 phi->init_req(1, _gvn.transform( slow_result ));
4817 r->init_req(1, iftrue);
4818
4819 // Else fall through
4820 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4821 set_control(iffalse);
4822
4823 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4824 r->init_req(2, iffalse);
4825
4826 // Post merge
4827 set_control(_gvn.transform(r));
4828 record_for_igvn(r);
4829
4830 C->set_has_split_ifs(true); // Has chance for split-if optimization
4831 result = phi;
4832 assert(result->bottom_type()->isa_long(), "must be");
4833 break;
4834 }
4835
4836 case vmIntrinsics::_floatToIntBits: {
4837 // two paths (plus control) merge in a wood
4838 RegionNode *r = new RegionNode(3);
4839 Node *phi = new PhiNode(r, TypeInt::INT);
4840
4841 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4842 // Build the boolean node
4843 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4844
4845 // Branch either way.
4846 // NaN case is less traveled, which makes all the difference.
4847 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4848 Node *opt_isnan = _gvn.transform(ifisnan);
4849 assert( opt_isnan->is_If(), "Expect an IfNode");
4850 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4851 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4852
4853 set_control(iftrue);
4854
4855 static const jint nan_bits = 0x7fc00000;
4856 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4857 phi->init_req(1, _gvn.transform( slow_result ));
4858 r->init_req(1, iftrue);
4859
4860 // Else fall through
4861 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4862 set_control(iffalse);
4863
4864 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4865 r->init_req(2, iffalse);
4866
4867 // Post merge
4868 set_control(_gvn.transform(r));
4869 record_for_igvn(r);
4870
4871 C->set_has_split_ifs(true); // Has chance for split-if optimization
4872 result = phi;
4873 assert(result->bottom_type()->isa_int(), "must be");
4874 break;
4875 }
4876
4877 default:
4878 fatal_unexpected_iid(id);
4879 break;
4880 }
4881 set_result(_gvn.transform(result));
4882 return true;
4883 }
4884
4885 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4886 Node* arg = argument(0);
4887 Node* result = NULL;
4888
4889 switch (id) {
4890 case vmIntrinsics::_floatIsInfinite:
4891 result = new IsInfiniteFNode(arg);
4892 break;
4893 case vmIntrinsics::_floatIsFinite:
4894 result = new IsFiniteFNode(arg);
4895 break;
4896 case vmIntrinsics::_doubleIsInfinite:
4897 result = new IsInfiniteDNode(arg);
4898 break;
4899 case vmIntrinsics::_doubleIsFinite:
4900 result = new IsFiniteDNode(arg);
4901 break;
4902 default:
4903 fatal_unexpected_iid(id);
4904 break;
4905 }
4906 set_result(_gvn.transform(result));
4907 return true;
4908 }
4909
4910 //----------------------inline_unsafe_copyMemory-------------------------
4911 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4912
4913 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4914 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4915 const Type* base_t = gvn.type(base);
4916
4917 bool in_native = (base_t == TypePtr::NULL_PTR);
4918 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4919 bool is_mixed = !in_heap && !in_native;
4920
4921 if (is_mixed) {
4922 return true; // mixed accesses can touch both on-heap and off-heap memory
4923 }
4924 if (in_heap) {
4925 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4926 if (!is_prim_array) {
4927 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4928 // there's not enough type information available to determine proper memory slice for it.
4929 return true;
4930 }
4931 }
4932 return false;
4933 }
4934
4935 bool LibraryCallKit::inline_unsafe_copyMemory() {
4936 if (callee()->is_static()) return false; // caller must have the capability!
4937 null_check_receiver(); // null-check receiver
4938 if (stopped()) return true;
4939
4940 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4941
4942 Node* src_base = argument(1); // type: oop
4943 Node* src_off = ConvL2X(argument(2)); // type: long
4944 Node* dst_base = argument(4); // type: oop
4945 Node* dst_off = ConvL2X(argument(5)); // type: long
4946 Node* size = ConvL2X(argument(7)); // type: long
4947
4948 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4949 "fieldOffset must be byte-scaled");
4950
4951 Node* src_addr = make_unsafe_address(src_base, src_off);
4952 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4953
4954 Node* thread = _gvn.transform(new ThreadLocalNode());
4955 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4956 BasicType doing_unsafe_access_bt = T_BYTE;
4957 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4958
4959 // update volatile field
4960 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4961
4962 int flags = RC_LEAF | RC_NO_FP;
4963
4964 const TypePtr* dst_type = TypePtr::BOTTOM;
4965
4966 // Adjust memory effects of the runtime call based on input values.
4967 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4968 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4969 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4970
4971 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4972 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4973 flags |= RC_NARROW_MEM; // narrow in memory
4974 }
4975 }
4976
4977 // Call it. Note that the length argument is not scaled.
4978 make_runtime_call(flags,
4979 OptoRuntime::fast_arraycopy_Type(),
4980 StubRoutines::unsafe_arraycopy(),
4981 "unsafe_arraycopy",
4982 dst_type,
4983 src_addr, dst_addr, size XTOP);
4984
4985 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4986
4987 return true;
4988 }
4989
4990 #undef XTOP
4991
4992 //------------------------clone_coping-----------------------------------
4993 // Helper function for inline_native_clone.
4994 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4995 assert(obj_size != NULL, "");
4996 Node* raw_obj = alloc_obj->in(1);
4997 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4998
4999 AllocateNode* alloc = NULL;
5000 if (ReduceBulkZeroing) {
5001 // We will be completely responsible for initializing this object -
5002 // mark Initialize node as complete.
5003 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
5004 // The object was just allocated - there should be no any stores!
5005 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
5006 // Mark as complete_with_arraycopy so that on AllocateNode
5007 // expansion, we know this AllocateNode is initialized by an array
5008 // copy and a StoreStore barrier exists after the array copy.
5009 alloc->initialization()->set_complete_with_arraycopy();
5010 }
5011
5012 Node* size = _gvn.transform(obj_size);
5013 access_clone(obj, alloc_obj, size, is_array);
5014
5015 // Do not let reads from the cloned object float above the arraycopy.
5016 if (alloc != NULL) {
5017 // Do not let stores that initialize this object be reordered with
5018 // a subsequent store that would make this object accessible by
5019 // other threads.
5020 // Record what AllocateNode this StoreStore protects so that
5021 // escape analysis can go from the MemBarStoreStoreNode to the
5022 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5023 // based on the escape status of the AllocateNode.
5024 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5025 } else {
5026 insert_mem_bar(Op_MemBarCPUOrder);
5027 }
5028 }
5029
5030 //------------------------inline_native_clone----------------------------
5031 // protected native Object java.lang.Object.clone();
5032 //
5033 // Here are the simple edge cases:
5034 // null receiver => normal trap
5035 // virtual and clone was overridden => slow path to out-of-line clone
5036 // not cloneable or finalizer => slow path to out-of-line Object.clone
5037 //
5038 // The general case has two steps, allocation and copying.
5039 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5040 //
5041 // Copying also has two cases, oop arrays and everything else.
5042 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5043 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5044 //
5045 // These steps fold up nicely if and when the cloned object's klass
5046 // can be sharply typed as an object array, a type array, or an instance.
5047 //
5048 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5049 PhiNode* result_val;
5050
5051 // Set the reexecute bit for the interpreter to reexecute
5052 // the bytecode that invokes Object.clone if deoptimization happens.
5053 { PreserveReexecuteState preexecs(this);
5054 jvms()->set_should_reexecute(true);
5055
5056 Node* obj = argument(0);
5057 obj = null_check_receiver();
5058 if (stopped()) return true;
5059
5060 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5061
5062 // If we are going to clone an instance, we need its exact type to
5063 // know the number and types of fields to convert the clone to
5064 // loads/stores. Maybe a speculative type can help us.
5065 if (!obj_type->klass_is_exact() &&
5066 obj_type->speculative_type() != NULL &&
5067 obj_type->speculative_type()->is_instance_klass() &&
5068 !obj_type->speculative_type()->is_inlinetype()) {
5069 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5070 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5071 !spec_ik->has_injected_fields()) {
5072 if (!obj_type->isa_instptr() ||
5073 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5074 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5075 }
5076 }
5077 }
5078
5079 // Conservatively insert a memory barrier on all memory slices.
5080 // Do not let writes into the original float below the clone.
5081 insert_mem_bar(Op_MemBarCPUOrder);
5082
5083 // paths into result_reg:
5084 enum {
5085 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5086 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5087 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5088 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5089 PATH_LIMIT
5090 };
5091 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5092 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5093 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5094 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5095 record_for_igvn(result_reg);
5096
5097 Node* obj_klass = load_object_klass(obj);
5098 // We only go to the fast case code if we pass a number of guards.
5099 // The paths which do not pass are accumulated in the slow_region.
5100 RegionNode* slow_region = new RegionNode(1);
5101 record_for_igvn(slow_region);
5102
5103 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
5104 if (array_ctl != NULL) {
5105 // It's an array.
5106 PreserveJVMState pjvms(this);
5107 set_control(array_ctl);
5108
5109 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5110 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5111 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5112 obj_type->can_be_inline_array() &&
5113 (ary_ptr == NULL || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5114 // Flattened inline type array may have object field that would require a
5115 // write barrier. Conservatively, go to slow path.
5116 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5117 }
5118
5119 if (!stopped()) {
5120 Node* obj_length = load_array_length(obj);
5121 Node* obj_size = NULL;
5122 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
5123
5124 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5125 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5126 // If it is an oop array, it requires very special treatment,
5127 // because gc barriers are required when accessing the array.
5128 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
5129 if (is_obja != NULL) {
5130 PreserveJVMState pjvms2(this);
5131 set_control(is_obja);
5132 // Generate a direct call to the right arraycopy function(s).
5133 // Clones are always tightly coupled.
5134 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5135 ac->set_clone_oop_array();
5136 Node* n = _gvn.transform(ac);
5137 assert(n == ac, "cannot disappear");
5138 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5139
5140 result_reg->init_req(_objArray_path, control());
5141 result_val->init_req(_objArray_path, alloc_obj);
5142 result_i_o ->set_req(_objArray_path, i_o());
5143 result_mem ->set_req(_objArray_path, reset_memory());
5144 }
5145 }
5146 // Otherwise, there are no barriers to worry about.
5147 // (We can dispense with card marks if we know the allocation
5148 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5149 // causes the non-eden paths to take compensating steps to
5150 // simulate a fresh allocation, so that no further
5151 // card marks are required in compiled code to initialize
5152 // the object.)
5153
5154 if (!stopped()) {
5155 copy_to_clone(obj, alloc_obj, obj_size, true);
5156
5157 // Present the results of the copy.
5158 result_reg->init_req(_array_path, control());
5159 result_val->init_req(_array_path, alloc_obj);
5160 result_i_o ->set_req(_array_path, i_o());
5161 result_mem ->set_req(_array_path, reset_memory());
5162 }
5163 }
5164 }
5165
5166 if (!stopped()) {
5167 // It's an instance (we did array above). Make the slow-path tests.
5168 // If this is a virtual call, we generate a funny guard. We grab
5169 // the vtable entry corresponding to clone() from the target object.
5170 // If the target method which we are calling happens to be the
5171 // Object clone() method, we pass the guard. We do not need this
5172 // guard for non-virtual calls; the caller is known to be the native
5173 // Object clone().
5174 if (is_virtual) {
5175 generate_virtual_guard(obj_klass, slow_region);
5176 }
5177
5178 // The object must be easily cloneable and must not have a finalizer.
5179 // Both of these conditions may be checked in a single test.
5180 // We could optimize the test further, but we don't care.
5181 generate_access_flags_guard(obj_klass,
5182 // Test both conditions:
5183 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5184 // Must be cloneable but not finalizer:
5185 JVM_ACC_IS_CLONEABLE_FAST,
5186 slow_region);
5187 }
5188
5189 if (!stopped()) {
5190 // It's an instance, and it passed the slow-path tests.
5191 PreserveJVMState pjvms(this);
5192 Node* obj_size = NULL;
5193 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5194 // is reexecuted if deoptimization occurs and there could be problems when merging
5195 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5196 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
5197
5198 copy_to_clone(obj, alloc_obj, obj_size, false);
5199
5200 // Present the results of the slow call.
5201 result_reg->init_req(_instance_path, control());
5202 result_val->init_req(_instance_path, alloc_obj);
5203 result_i_o ->set_req(_instance_path, i_o());
5204 result_mem ->set_req(_instance_path, reset_memory());
5205 }
5206
5207 // Generate code for the slow case. We make a call to clone().
5208 set_control(_gvn.transform(slow_region));
5209 if (!stopped()) {
5210 PreserveJVMState pjvms(this);
5211 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
5212 // We need to deoptimize on exception (see comment above)
5213 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5214 // this->control() comes from set_results_for_java_call
5215 result_reg->init_req(_slow_path, control());
5216 result_val->init_req(_slow_path, slow_result);
5217 result_i_o ->set_req(_slow_path, i_o());
5218 result_mem ->set_req(_slow_path, reset_memory());
5219 }
5220
5221 // Return the combined state.
5222 set_control( _gvn.transform(result_reg));
5223 set_i_o( _gvn.transform(result_i_o));
5224 set_all_memory( _gvn.transform(result_mem));
5225 } // original reexecute is set back here
5226
5227 set_result(_gvn.transform(result_val));
5228 return true;
5229 }
5230
5231 // If we have a tightly coupled allocation, the arraycopy may take care
5232 // of the array initialization. If one of the guards we insert between
5233 // the allocation and the arraycopy causes a deoptimization, an
5234 // uninitialized array will escape the compiled method. To prevent that
5235 // we set the JVM state for uncommon traps between the allocation and
5236 // the arraycopy to the state before the allocation so, in case of
5237 // deoptimization, we'll reexecute the allocation and the
5238 // initialization.
5239 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5240 if (alloc != NULL) {
5241 ciMethod* trap_method = alloc->jvms()->method();
5242 int trap_bci = alloc->jvms()->bci();
5243
5244 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5245 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5246 // Make sure there's no store between the allocation and the
5247 // arraycopy otherwise visible side effects could be rexecuted
5248 // in case of deoptimization and cause incorrect execution.
5249 bool no_interfering_store = true;
5250 Node* mem = alloc->in(TypeFunc::Memory);
5251 if (mem->is_MergeMem()) {
5252 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5253 Node* n = mms.memory();
5254 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5255 assert(n->is_Store(), "what else?");
5256 no_interfering_store = false;
5257 break;
5258 }
5259 }
5260 } else {
5261 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5262 Node* n = mms.memory();
5263 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5264 assert(n->is_Store(), "what else?");
5265 no_interfering_store = false;
5266 break;
5267 }
5268 }
5269 }
5270
5271 if (no_interfering_store) {
5272 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5273 uint size = alloc->req();
5274 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5275 old_jvms->set_map(sfpt);
5276 for (uint i = 0; i < size; i++) {
5277 sfpt->init_req(i, alloc->in(i));
5278 }
5279 // re-push array length for deoptimization
5280 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5281 old_jvms->set_sp(old_jvms->sp()+1);
5282 old_jvms->set_monoff(old_jvms->monoff()+1);
5283 old_jvms->set_scloff(old_jvms->scloff()+1);
5284 old_jvms->set_endoff(old_jvms->endoff()+1);
5285 old_jvms->set_should_reexecute(true);
5286
5287 sfpt->set_i_o(map()->i_o());
5288 sfpt->set_memory(map()->memory());
5289 sfpt->set_control(map()->control());
5290
5291 JVMState* saved_jvms = jvms();
5292 saved_reexecute_sp = _reexecute_sp;
5293
5294 set_jvms(sfpt->jvms());
5295 _reexecute_sp = jvms()->sp();
5296
5297 return saved_jvms;
5298 }
5299 }
5300 }
5301 return NULL;
5302 }
5303
5304 // In case of a deoptimization, we restart execution at the
5305 // allocation, allocating a new array. We would leave an uninitialized
5306 // array in the heap that GCs wouldn't expect. Move the allocation
5307 // after the traps so we don't allocate the array if we
5308 // deoptimize. This is possible because tightly_coupled_allocation()
5309 // guarantees there's no observer of the allocated array at this point
5310 // and the control flow is simple enough.
5311 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
5312 int saved_reexecute_sp, uint new_idx) {
5313 if (saved_jvms != NULL && !stopped()) {
5314 assert(alloc != NULL, "only with a tightly coupled allocation");
5315 // restore JVM state to the state at the arraycopy
5316 saved_jvms->map()->set_control(map()->control());
5317 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
5318 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
5319 // If we've improved the types of some nodes (null check) while
5320 // emitting the guards, propagate them to the current state
5321 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
5322 set_jvms(saved_jvms);
5323 _reexecute_sp = saved_reexecute_sp;
5324
5325 // Remove the allocation from above the guards
5326 CallProjections* callprojs = alloc->extract_projections(true);
5327 InitializeNode* init = alloc->initialization();
5328 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5329 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5330 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5331
5332 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5333 // the allocation (i.e. is only valid if the allocation succeeds):
5334 // 1) replace CastIINode with AllocateArrayNode's length here
5335 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5336 //
5337 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5338 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5339 Node* init_control = init->proj_out(TypeFunc::Control);
5340 Node* alloc_length = alloc->Ideal_length();
5341 #ifdef ASSERT
5342 Node* prev_cast = NULL;
5343 #endif
5344 for (uint i = 0; i < init_control->outcnt(); i++) {
5345 Node* init_out = init_control->raw_out(i);
5346 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5347 #ifdef ASSERT
5348 if (prev_cast == NULL) {
5349 prev_cast = init_out;
5350 } else {
5351 if (prev_cast->cmp(*init_out) == false) {
5352 prev_cast->dump();
5353 init_out->dump();
5354 assert(false, "not equal CastIINode");
5355 }
5356 }
5357 #endif
5358 C->gvn_replace_by(init_out, alloc_length);
5359 }
5360 }
5361 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5362
5363 // move the allocation here (after the guards)
5364 _gvn.hash_delete(alloc);
5365 alloc->set_req(TypeFunc::Control, control());
5366 alloc->set_req(TypeFunc::I_O, i_o());
5367 Node *mem = reset_memory();
5368 set_all_memory(mem);
5369 alloc->set_req(TypeFunc::Memory, mem);
5370 set_control(init->proj_out_or_null(TypeFunc::Control));
5371 set_i_o(callprojs->fallthrough_ioproj);
5372
5373 // Update memory as done in GraphKit::set_output_for_allocation()
5374 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5375 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5376 if (ary_type->isa_aryptr() && length_type != NULL) {
5377 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5378 }
5379 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5380 int elemidx = C->get_alias_index(telemref);
5381 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5382 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5383
5384 Node* allocx = _gvn.transform(alloc);
5385 assert(allocx == alloc, "where has the allocation gone?");
5386 assert(dest->is_CheckCastPP(), "not an allocation result?");
5387
5388 _gvn.hash_delete(dest);
5389 dest->set_req(0, control());
5390 Node* destx = _gvn.transform(dest);
5391 assert(destx == dest, "where has the allocation result gone?");
5392
5393 array_ideal_length(alloc, ary_type, true);
5394 }
5395 }
5396
5397
5398 //------------------------------inline_arraycopy-----------------------
5399 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5400 // Object dest, int destPos,
5401 // int length);
5402 bool LibraryCallKit::inline_arraycopy() {
5403 // Get the arguments.
5404 Node* src = argument(0); // type: oop
5405 Node* src_offset = argument(1); // type: int
5406 Node* dest = argument(2); // type: oop
5407 Node* dest_offset = argument(3); // type: int
5408 Node* length = argument(4); // type: int
5409
5410 uint new_idx = C->unique();
5411
5412 // Check for allocation before we add nodes that would confuse
5413 // tightly_coupled_allocation()
5414 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5415
5416 int saved_reexecute_sp = -1;
5417 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5418 // See arraycopy_restore_alloc_state() comment
5419 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5420 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5421 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5422 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5423
5424 // The following tests must be performed
5425 // (1) src and dest are arrays.
5426 // (2) src and dest arrays must have elements of the same BasicType
5427 // (3) src and dest must not be null.
5428 // (4) src_offset must not be negative.
5429 // (5) dest_offset must not be negative.
5430 // (6) length must not be negative.
5431 // (7) src_offset + length must not exceed length of src.
5432 // (8) dest_offset + length must not exceed length of dest.
5433 // (9) each element of an oop array must be assignable
5434
5435 // (3) src and dest must not be null.
5436 // always do this here because we need the JVM state for uncommon traps
5437 Node* null_ctl = top();
5438 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5439 assert(null_ctl->is_top(), "no null control here");
5440 dest = null_check(dest, T_ARRAY);
5441
5442 if (!can_emit_guards) {
5443 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5444 // guards but the arraycopy node could still take advantage of a
5445 // tightly allocated allocation. tightly_coupled_allocation() is
5446 // called again to make sure it takes the null check above into
5447 // account: the null check is mandatory and if it caused an
5448 // uncommon trap to be emitted then the allocation can't be
5449 // considered tightly coupled in this context.
5450 alloc = tightly_coupled_allocation(dest);
5451 }
5452
5453 bool validated = false;
5454
5455 const Type* src_type = _gvn.type(src);
5456 const Type* dest_type = _gvn.type(dest);
5457 const TypeAryPtr* top_src = src_type->isa_aryptr();
5458 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5459
5460 // Do we have the type of src?
5461 bool has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5462 // Do we have the type of dest?
5463 bool has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5464 // Is the type for src from speculation?
5465 bool src_spec = false;
5466 // Is the type for dest from speculation?
5467 bool dest_spec = false;
5468
5469 if ((!has_src || !has_dest) && can_emit_guards) {
5470 // We don't have sufficient type information, let's see if
5471 // speculative types can help. We need to have types for both src
5472 // and dest so that it pays off.
5473
5474 // Do we already have or could we have type information for src
5475 bool could_have_src = has_src;
5476 // Do we already have or could we have type information for dest
5477 bool could_have_dest = has_dest;
5478
5479 ciKlass* src_k = NULL;
5480 if (!has_src) {
5481 src_k = src_type->speculative_type_not_null();
5482 if (src_k != NULL && src_k->is_array_klass()) {
5483 could_have_src = true;
5484 }
5485 }
5486
5487 ciKlass* dest_k = NULL;
5488 if (!has_dest) {
5489 dest_k = dest_type->speculative_type_not_null();
5490 if (dest_k != NULL && dest_k->is_array_klass()) {
5491 could_have_dest = true;
5492 }
5493 }
5494
5495 if (could_have_src && could_have_dest) {
5496 // This is going to pay off so emit the required guards
5497 if (!has_src) {
5498 src = maybe_cast_profiled_obj(src, src_k, true);
5499 src_type = _gvn.type(src);
5500 top_src = src_type->isa_aryptr();
5501 has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5502 src_spec = true;
5503 }
5504 if (!has_dest) {
5505 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5506 dest_type = _gvn.type(dest);
5507 top_dest = dest_type->isa_aryptr();
5508 has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5509 dest_spec = true;
5510 }
5511 }
5512 }
5513
5514 if (has_src && has_dest && can_emit_guards) {
5515 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5516 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5517 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5518 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5519
5520 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5521 // If both arrays are object arrays then having the exact types
5522 // for both will remove the need for a subtype check at runtime
5523 // before the call and may make it possible to pick a faster copy
5524 // routine (without a subtype check on every element)
5525 // Do we have the exact type of src?
5526 bool could_have_src = src_spec;
5527 // Do we have the exact type of dest?
5528 bool could_have_dest = dest_spec;
5529 ciKlass* src_k = NULL;
5530 ciKlass* dest_k = NULL;
5531 if (!src_spec) {
5532 src_k = src_type->speculative_type_not_null();
5533 if (src_k != NULL && src_k->is_array_klass()) {
5534 could_have_src = true;
5535 }
5536 }
5537 if (!dest_spec) {
5538 dest_k = dest_type->speculative_type_not_null();
5539 if (dest_k != NULL && dest_k->is_array_klass()) {
5540 could_have_dest = true;
5541 }
5542 }
5543 if (could_have_src && could_have_dest) {
5544 // If we can have both exact types, emit the missing guards
5545 if (could_have_src && !src_spec) {
5546 src = maybe_cast_profiled_obj(src, src_k, true);
5547 src_type = _gvn.type(src);
5548 top_src = src_type->isa_aryptr();
5549 }
5550 if (could_have_dest && !dest_spec) {
5551 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5552 dest_type = _gvn.type(dest);
5553 top_dest = dest_type->isa_aryptr();
5554 }
5555 }
5556 }
5557 }
5558
5559 ciMethod* trap_method = method();
5560 int trap_bci = bci();
5561 if (saved_jvms != NULL) {
5562 trap_method = alloc->jvms()->method();
5563 trap_bci = alloc->jvms()->bci();
5564 }
5565
5566 bool negative_length_guard_generated = false;
5567
5568 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5569 can_emit_guards && !src->is_top() && !dest->is_top()) {
5570 // validate arguments: enables transformation the ArrayCopyNode
5571 validated = true;
5572
5573 RegionNode* slow_region = new RegionNode(1);
5574 record_for_igvn(slow_region);
5575
5576 // (1) src and dest are arrays.
5577 generate_non_array_guard(load_object_klass(src), slow_region);
5578 generate_non_array_guard(load_object_klass(dest), slow_region);
5579
5580 // (2) src and dest arrays must have elements of the same BasicType
5581 // done at macro expansion or at Ideal transformation time
5582
5583 // (4) src_offset must not be negative.
5584 generate_negative_guard(src_offset, slow_region);
5585
5586 // (5) dest_offset must not be negative.
5587 generate_negative_guard(dest_offset, slow_region);
5588
5589 // (7) src_offset + length must not exceed length of src.
5590 generate_limit_guard(src_offset, length,
5591 load_array_length(src),
5592 slow_region);
5593
5594 // (8) dest_offset + length must not exceed length of dest.
5595 generate_limit_guard(dest_offset, length,
5596 load_array_length(dest),
5597 slow_region);
5598
5599 // (6) length must not be negative.
5600 // This is also checked in generate_arraycopy() during macro expansion, but
5601 // we also have to check it here for the case where the ArrayCopyNode will
5602 // be eliminated by Escape Analysis.
5603 if (EliminateAllocations) {
5604 generate_negative_guard(length, slow_region);
5605 negative_length_guard_generated = true;
5606 }
5607
5608 // (9) each element of an oop array must be assignable
5609 Node* dest_klass = load_object_klass(dest);
5610 if (src != dest) {
5611 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5612 slow_region->add_req(not_subtype_ctrl);
5613 }
5614
5615 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5616 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5617 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5618 src_type = _gvn.type(src);
5619 top_src = src_type->isa_aryptr();
5620
5621 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
5622 if (!stopped() && UseFlatArray) {
5623 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
5624 assert(top_dest == NULL || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
5625 if (top_src != NULL && top_src->is_flat()) {
5626 // Src is flat, check that dest is flat as well
5627 if (top_dest != NULL && !top_dest->is_flat()) {
5628 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
5629 // Since dest is flat and src <: dest, dest must have the same type as src.
5630 top_dest = top_src->cast_to_exactness(false);
5631 assert(top_dest->is_flat(), "dest must be flat");
5632 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
5633 }
5634 } else if (top_src == NULL || !top_src->is_not_flat()) {
5635 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5636 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5637 assert(top_dest == NULL || !top_dest->is_flat(), "dest array must not be flat");
5638 generate_fair_guard(flat_array_test(src), slow_region);
5639 if (top_src != NULL) {
5640 top_src = top_src->cast_to_not_flat();
5641 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
5642 }
5643 }
5644 }
5645
5646 {
5647 PreserveJVMState pjvms(this);
5648 set_control(_gvn.transform(slow_region));
5649 uncommon_trap(Deoptimization::Reason_intrinsic,
5650 Deoptimization::Action_make_not_entrant);
5651 assert(stopped(), "Should be stopped");
5652 }
5653 }
5654
5655 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5656
5657 if (stopped()) {
5658 return true;
5659 }
5660
5661 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5662 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5663 // so the compiler has a chance to eliminate them: during macro expansion,
5664 // we have to set their control (CastPP nodes are eliminated).
5665 load_object_klass(src), load_object_klass(dest),
5666 load_array_length(src), load_array_length(dest));
5667
5668 ac->set_arraycopy(validated);
5669
5670 Node* n = _gvn.transform(ac);
5671 if (n == ac) {
5672 ac->connect_outputs(this);
5673 } else {
5674 assert(validated, "shouldn't transform if all arguments not validated");
5675 set_all_memory(n);
5676 }
5677 clear_upper_avx();
5678
5679
5680 return true;
5681 }
5682
5683
5684 // Helper function which determines if an arraycopy immediately follows
5685 // an allocation, with no intervening tests or other escapes for the object.
5686 AllocateArrayNode*
5687 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5688 if (stopped()) return NULL; // no fast path
5689 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5690
5691 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5692 if (alloc == NULL) return NULL;
5693
5694 Node* rawmem = memory(Compile::AliasIdxRaw);
5695 // Is the allocation's memory state untouched?
5696 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5697 // Bail out if there have been raw-memory effects since the allocation.
5698 // (Example: There might have been a call or safepoint.)
5699 return NULL;
5700 }
5701 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5702 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5703 return NULL;
5704 }
5705
5706 // There must be no unexpected observers of this allocation.
5707 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5708 Node* obs = ptr->fast_out(i);
5709 if (obs != this->map()) {
5710 return NULL;
5711 }
5712 }
5713
5714 // This arraycopy must unconditionally follow the allocation of the ptr.
5715 Node* alloc_ctl = ptr->in(0);
5716 Node* ctl = control();
5717 while (ctl != alloc_ctl) {
5718 // There may be guards which feed into the slow_region.
5719 // Any other control flow means that we might not get a chance
5720 // to finish initializing the allocated object.
5721 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5722 IfNode* iff = ctl->in(0)->as_If();
5723 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5724 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5725 // One more try: Various low-level checks bottom out in
5726 // uncommon traps. If the debug-info of the trap omits
5727 // any reference to the allocation, as we've already
5728 // observed, then there can be no objection to the trap.
5729 bool found_trap = false;
5730 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5731 Node* obs = not_ctl->fast_out(j);
5732 if (obs->in(0) == not_ctl && obs->is_Call() &&
5733 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5734 found_trap = true; break;
5735 }
5736 }
5737 if (found_trap) {
5738 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5739 continue;
5740 }
5741 }
5742 return NULL;
5743 }
5744
5745 // If we get this far, we have an allocation which immediately
5746 // precedes the arraycopy, and we can take over zeroing the new object.
5747 // The arraycopy will finish the initialization, and provide
5748 // a new control state to which we will anchor the destination pointer.
5749
5750 return alloc;
5751 }
5752
5753 //-------------inline_encodeISOArray-----------------------------------
5754 // encode char[] to byte[] in ISO_8859_1 or ASCII
5755 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5756 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5757 // no receiver since it is static method
5758 Node *src = argument(0);
5759 Node *src_offset = argument(1);
5760 Node *dst = argument(2);
5761 Node *dst_offset = argument(3);
5762 Node *length = argument(4);
5763
5764 src = must_be_not_null(src, true);
5765 dst = must_be_not_null(dst, true);
5766
5767 const Type* src_type = src->Value(&_gvn);
5768 const Type* dst_type = dst->Value(&_gvn);
5769 const TypeAryPtr* top_src = src_type->isa_aryptr();
5770 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5771 if (top_src == NULL || top_src->elem() == Type::BOTTOM ||
5772 top_dest == NULL || top_dest->elem() == Type::BOTTOM) {
5773 // failed array check
5774 return false;
5775 }
5776
5777 // Figure out the size and type of the elements we will be copying.
5778 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5779 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
5780 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5781 return false;
5782 }
5783
5784 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5785 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5786 // 'src_start' points to src array + scaled offset
5787 // 'dst_start' points to dst array + scaled offset
5788
5789 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5790 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5791 enc = _gvn.transform(enc);
5792 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5793 set_memory(res_mem, mtype);
5794 set_result(enc);
5795 clear_upper_avx();
5796
5797 return true;
5798 }
5799
5800 //-------------inline_multiplyToLen-----------------------------------
5801 bool LibraryCallKit::inline_multiplyToLen() {
5802 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5803
5804 address stubAddr = StubRoutines::multiplyToLen();
5805 if (stubAddr == NULL) {
5806 return false; // Intrinsic's stub is not implemented on this platform
5807 }
5808 const char* stubName = "multiplyToLen";
5809
5810 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5811
5812 // no receiver because it is a static method
5813 Node* x = argument(0);
5814 Node* xlen = argument(1);
5815 Node* y = argument(2);
5816 Node* ylen = argument(3);
5817 Node* z = argument(4);
5818
5819 x = must_be_not_null(x, true);
5820 y = must_be_not_null(y, true);
5821
5822 const Type* x_type = x->Value(&_gvn);
5823 const Type* y_type = y->Value(&_gvn);
5824 const TypeAryPtr* top_x = x_type->isa_aryptr();
5825 const TypeAryPtr* top_y = y_type->isa_aryptr();
5826 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5827 top_y == NULL || top_y->elem() == Type::BOTTOM) {
5828 // failed array check
5829 return false;
5830 }
5831
5832 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5833 BasicType y_elem = y_type->isa_aryptr()->elem()->array_element_basic_type();
5834 if (x_elem != T_INT || y_elem != T_INT) {
5835 return false;
5836 }
5837
5838 // Set the original stack and the reexecute bit for the interpreter to reexecute
5839 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5840 // on the return from z array allocation in runtime.
5841 { PreserveReexecuteState preexecs(this);
5842 jvms()->set_should_reexecute(true);
5843
5844 Node* x_start = array_element_address(x, intcon(0), x_elem);
5845 Node* y_start = array_element_address(y, intcon(0), y_elem);
5846 // 'x_start' points to x array + scaled xlen
5847 // 'y_start' points to y array + scaled ylen
5848
5849 // Allocate the result array
5850 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5851 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5852 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5853
5854 IdealKit ideal(this);
5855
5856 #define __ ideal.
5857 Node* one = __ ConI(1);
5858 Node* zero = __ ConI(0);
5859 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5860 __ set(need_alloc, zero);
5861 __ set(z_alloc, z);
5862 __ if_then(z, BoolTest::eq, null()); {
5863 __ increment (need_alloc, one);
5864 } __ else_(); {
5865 // Update graphKit memory and control from IdealKit.
5866 sync_kit(ideal);
5867 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5868 cast->init_req(0, control());
5869 _gvn.set_type(cast, cast->bottom_type());
5870 C->record_for_igvn(cast);
5871
5872 Node* zlen_arg = load_array_length(cast);
5873 // Update IdealKit memory and control from graphKit.
5874 __ sync_kit(this);
5875 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5876 __ increment (need_alloc, one);
5877 } __ end_if();
5878 } __ end_if();
5879
5880 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5881 // Update graphKit memory and control from IdealKit.
5882 sync_kit(ideal);
5883 Node * narr = new_array(klass_node, zlen, 1);
5884 // Update IdealKit memory and control from graphKit.
5885 __ sync_kit(this);
5886 __ set(z_alloc, narr);
5887 } __ end_if();
5888
5889 sync_kit(ideal);
5890 z = __ value(z_alloc);
5891 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5892 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5893 // Final sync IdealKit and GraphKit.
5894 final_sync(ideal);
5895 #undef __
5896
5897 Node* z_start = array_element_address(z, intcon(0), T_INT);
5898
5899 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5900 OptoRuntime::multiplyToLen_Type(),
5901 stubAddr, stubName, TypePtr::BOTTOM,
5902 x_start, xlen, y_start, ylen, z_start, zlen);
5903 } // original reexecute is set back here
5904
5905 C->set_has_split_ifs(true); // Has chance for split-if optimization
5906 set_result(z);
5907 return true;
5908 }
5909
5910 //-------------inline_squareToLen------------------------------------
5911 bool LibraryCallKit::inline_squareToLen() {
5912 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5913
5914 address stubAddr = StubRoutines::squareToLen();
5915 if (stubAddr == NULL) {
5916 return false; // Intrinsic's stub is not implemented on this platform
5917 }
5918 const char* stubName = "squareToLen";
5919
5920 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5921
5922 Node* x = argument(0);
5923 Node* len = argument(1);
5924 Node* z = argument(2);
5925 Node* zlen = argument(3);
5926
5927 x = must_be_not_null(x, true);
5928 z = must_be_not_null(z, true);
5929
5930 const Type* x_type = x->Value(&_gvn);
5931 const Type* z_type = z->Value(&_gvn);
5932 const TypeAryPtr* top_x = x_type->isa_aryptr();
5933 const TypeAryPtr* top_z = z_type->isa_aryptr();
5934 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5935 top_z == NULL || top_z->elem() == Type::BOTTOM) {
5936 // failed array check
5937 return false;
5938 }
5939
5940 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5941 BasicType z_elem = z_type->isa_aryptr()->elem()->array_element_basic_type();
5942 if (x_elem != T_INT || z_elem != T_INT) {
5943 return false;
5944 }
5945
5946
5947 Node* x_start = array_element_address(x, intcon(0), x_elem);
5948 Node* z_start = array_element_address(z, intcon(0), z_elem);
5949
5950 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5951 OptoRuntime::squareToLen_Type(),
5952 stubAddr, stubName, TypePtr::BOTTOM,
5953 x_start, len, z_start, zlen);
5954
5955 set_result(z);
5956 return true;
5957 }
5958
5959 //-------------inline_mulAdd------------------------------------------
5960 bool LibraryCallKit::inline_mulAdd() {
5961 assert(UseMulAddIntrinsic, "not implemented on this platform");
5962
5963 address stubAddr = StubRoutines::mulAdd();
5964 if (stubAddr == NULL) {
5965 return false; // Intrinsic's stub is not implemented on this platform
5966 }
5967 const char* stubName = "mulAdd";
5968
5969 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5970
5971 Node* out = argument(0);
5972 Node* in = argument(1);
5973 Node* offset = argument(2);
5974 Node* len = argument(3);
5975 Node* k = argument(4);
5976
5977 out = must_be_not_null(out, true);
5978
5979 const Type* out_type = out->Value(&_gvn);
5980 const Type* in_type = in->Value(&_gvn);
5981 const TypeAryPtr* top_out = out_type->isa_aryptr();
5982 const TypeAryPtr* top_in = in_type->isa_aryptr();
5983 if (top_out == NULL || top_out->elem() == Type::BOTTOM ||
5984 top_in == NULL || top_in->elem() == Type::BOTTOM) {
5985 // failed array check
5986 return false;
5987 }
5988
5989 BasicType out_elem = out_type->isa_aryptr()->elem()->array_element_basic_type();
5990 BasicType in_elem = in_type->isa_aryptr()->elem()->array_element_basic_type();
5991 if (out_elem != T_INT || in_elem != T_INT) {
5992 return false;
5993 }
5994
5995 Node* outlen = load_array_length(out);
5996 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5997 Node* out_start = array_element_address(out, intcon(0), out_elem);
5998 Node* in_start = array_element_address(in, intcon(0), in_elem);
5999
6000 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6001 OptoRuntime::mulAdd_Type(),
6002 stubAddr, stubName, TypePtr::BOTTOM,
6003 out_start,in_start, new_offset, len, k);
6004 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6005 set_result(result);
6006 return true;
6007 }
6008
6009 //-------------inline_montgomeryMultiply-----------------------------------
6010 bool LibraryCallKit::inline_montgomeryMultiply() {
6011 address stubAddr = StubRoutines::montgomeryMultiply();
6012 if (stubAddr == NULL) {
6013 return false; // Intrinsic's stub is not implemented on this platform
6014 }
6015
6016 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6017 const char* stubName = "montgomery_multiply";
6018
6019 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6020
6021 Node* a = argument(0);
6022 Node* b = argument(1);
6023 Node* n = argument(2);
6024 Node* len = argument(3);
6025 Node* inv = argument(4);
6026 Node* m = argument(6);
6027
6028 const Type* a_type = a->Value(&_gvn);
6029 const TypeAryPtr* top_a = a_type->isa_aryptr();
6030 const Type* b_type = b->Value(&_gvn);
6031 const TypeAryPtr* top_b = b_type->isa_aryptr();
6032 const Type* n_type = a->Value(&_gvn);
6033 const TypeAryPtr* top_n = n_type->isa_aryptr();
6034 const Type* m_type = a->Value(&_gvn);
6035 const TypeAryPtr* top_m = m_type->isa_aryptr();
6036 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
6037 top_b == NULL || top_b->elem() == Type::BOTTOM ||
6038 top_n == NULL || top_n->elem() == Type::BOTTOM ||
6039 top_m == NULL || top_m->elem() == Type::BOTTOM) {
6040 // failed array check
6041 return false;
6042 }
6043
6044 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
6045 BasicType b_elem = b_type->isa_aryptr()->elem()->array_element_basic_type();
6046 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
6047 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
6048 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6049 return false;
6050 }
6051
6052 // Make the call
6053 {
6054 Node* a_start = array_element_address(a, intcon(0), a_elem);
6055 Node* b_start = array_element_address(b, intcon(0), b_elem);
6056 Node* n_start = array_element_address(n, intcon(0), n_elem);
6057 Node* m_start = array_element_address(m, intcon(0), m_elem);
6058
6059 Node* call = make_runtime_call(RC_LEAF,
6060 OptoRuntime::montgomeryMultiply_Type(),
6061 stubAddr, stubName, TypePtr::BOTTOM,
6062 a_start, b_start, n_start, len, inv, top(),
6063 m_start);
6064 set_result(m);
6065 }
6066
6067 return true;
6068 }
6069
6070 bool LibraryCallKit::inline_montgomerySquare() {
6071 address stubAddr = StubRoutines::montgomerySquare();
6072 if (stubAddr == NULL) {
6073 return false; // Intrinsic's stub is not implemented on this platform
6074 }
6075
6076 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6077 const char* stubName = "montgomery_square";
6078
6079 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6080
6081 Node* a = argument(0);
6082 Node* n = argument(1);
6083 Node* len = argument(2);
6084 Node* inv = argument(3);
6085 Node* m = argument(5);
6086
6087 const Type* a_type = a->Value(&_gvn);
6088 const TypeAryPtr* top_a = a_type->isa_aryptr();
6089 const Type* n_type = a->Value(&_gvn);
6090 const TypeAryPtr* top_n = n_type->isa_aryptr();
6091 const Type* m_type = a->Value(&_gvn);
6092 const TypeAryPtr* top_m = m_type->isa_aryptr();
6093 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
6094 top_n == NULL || top_n->elem() == Type::BOTTOM ||
6095 top_m == NULL || top_m->elem() == Type::BOTTOM) {
6096 // failed array check
6097 return false;
6098 }
6099
6100 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
6101 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
6102 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
6103 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6104 return false;
6105 }
6106
6107 // Make the call
6108 {
6109 Node* a_start = array_element_address(a, intcon(0), a_elem);
6110 Node* n_start = array_element_address(n, intcon(0), n_elem);
6111 Node* m_start = array_element_address(m, intcon(0), m_elem);
6112
6113 Node* call = make_runtime_call(RC_LEAF,
6114 OptoRuntime::montgomerySquare_Type(),
6115 stubAddr, stubName, TypePtr::BOTTOM,
6116 a_start, n_start, len, inv, top(),
6117 m_start);
6118 set_result(m);
6119 }
6120
6121 return true;
6122 }
6123
6124 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6125 address stubAddr = NULL;
6126 const char* stubName = NULL;
6127
6128 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6129 if (stubAddr == NULL) {
6130 return false; // Intrinsic's stub is not implemented on this platform
6131 }
6132
6133 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6134
6135 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6136
6137 Node* newArr = argument(0);
6138 Node* oldArr = argument(1);
6139 Node* newIdx = argument(2);
6140 Node* shiftCount = argument(3);
6141 Node* numIter = argument(4);
6142
6143 const Type* newArr_type = newArr->Value(&_gvn);
6144 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
6145 const Type* oldArr_type = oldArr->Value(&_gvn);
6146 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
6147 if (top_newArr == NULL || top_newArr->elem() == Type::BOTTOM || top_oldArr == NULL
6148 || top_oldArr->elem() == Type::BOTTOM) {
6149 return false;
6150 }
6151
6152 BasicType newArr_elem = newArr_type->isa_aryptr()->elem()->array_element_basic_type();
6153 BasicType oldArr_elem = oldArr_type->isa_aryptr()->elem()->array_element_basic_type();
6154 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6155 return false;
6156 }
6157
6158 // Make the call
6159 {
6160 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6161 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6162
6163 Node* call = make_runtime_call(RC_LEAF,
6164 OptoRuntime::bigIntegerShift_Type(),
6165 stubAddr,
6166 stubName,
6167 TypePtr::BOTTOM,
6168 newArr_start,
6169 oldArr_start,
6170 newIdx,
6171 shiftCount,
6172 numIter);
6173 }
6174
6175 return true;
6176 }
6177
6178 //-------------inline_vectorizedMismatch------------------------------
6179 bool LibraryCallKit::inline_vectorizedMismatch() {
6180 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6181
6182 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6183 Node* obja = argument(0); // Object
6184 Node* aoffset = argument(1); // long
6185 Node* objb = argument(3); // Object
6186 Node* boffset = argument(4); // long
6187 Node* length = argument(6); // int
6188 Node* scale = argument(7); // int
6189
6190 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6191 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6192 if (obja_t == NULL || obja_t->elem() == Type::BOTTOM ||
6193 objb_t == NULL || objb_t->elem() == Type::BOTTOM ||
6194 scale == top()) {
6195 return false; // failed input validation
6196 }
6197
6198 Node* obja_adr = make_unsafe_address(obja, aoffset);
6199 Node* objb_adr = make_unsafe_address(objb, boffset);
6200
6201 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6202 //
6203 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6204 // if (length <= inline_limit) {
6205 // inline_path:
6206 // vmask = VectorMaskGen length
6207 // vload1 = LoadVectorMasked obja, vmask
6208 // vload2 = LoadVectorMasked objb, vmask
6209 // result1 = VectorCmpMasked vload1, vload2, vmask
6210 // } else {
6211 // call_stub_path:
6212 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6213 // }
6214 // exit_block:
6215 // return Phi(result1, result2);
6216 //
6217 enum { inline_path = 1, // input is small enough to process it all at once
6218 stub_path = 2, // input is too large; call into the VM
6219 PATH_LIMIT = 3
6220 };
6221
6222 Node* exit_block = new RegionNode(PATH_LIMIT);
6223 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6224 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6225
6226 Node* call_stub_path = control();
6227
6228 BasicType elem_bt = T_ILLEGAL;
6229
6230 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6231 if (scale_t->is_con()) {
6232 switch (scale_t->get_con()) {
6233 case 0: elem_bt = T_BYTE; break;
6234 case 1: elem_bt = T_SHORT; break;
6235 case 2: elem_bt = T_INT; break;
6236 case 3: elem_bt = T_LONG; break;
6237
6238 default: elem_bt = T_ILLEGAL; break; // not supported
6239 }
6240 }
6241
6242 int inline_limit = 0;
6243 bool do_partial_inline = false;
6244
6245 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6246 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6247 do_partial_inline = inline_limit >= 16;
6248 }
6249
6250 if (do_partial_inline) {
6251 assert(elem_bt != T_ILLEGAL, "sanity");
6252
6253 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6254 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6255 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6256
6257 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6258 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6259 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6260
6261 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
6262
6263 if (!stopped()) {
6264 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6265
6266 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6267 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6268 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6269 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6270
6271 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6272 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6273 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6274 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6275
6276 exit_block->init_req(inline_path, control());
6277 memory_phi->init_req(inline_path, map()->memory());
6278 result_phi->init_req(inline_path, result);
6279
6280 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6281 clear_upper_avx();
6282 }
6283 }
6284 }
6285
6286 if (call_stub_path != NULL) {
6287 set_control(call_stub_path);
6288
6289 Node* call = make_runtime_call(RC_LEAF,
6290 OptoRuntime::vectorizedMismatch_Type(),
6291 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6292 obja_adr, objb_adr, length, scale);
6293
6294 exit_block->init_req(stub_path, control());
6295 memory_phi->init_req(stub_path, map()->memory());
6296 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6297 }
6298
6299 exit_block = _gvn.transform(exit_block);
6300 memory_phi = _gvn.transform(memory_phi);
6301 result_phi = _gvn.transform(result_phi);
6302
6303 set_control(exit_block);
6304 set_all_memory(memory_phi);
6305 set_result(result_phi);
6306
6307 return true;
6308 }
6309
6310 /**
6311 * Calculate CRC32 for byte.
6312 * int java.util.zip.CRC32.update(int crc, int b)
6313 */
6314 bool LibraryCallKit::inline_updateCRC32() {
6315 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6316 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6317 // no receiver since it is static method
6318 Node* crc = argument(0); // type: int
6319 Node* b = argument(1); // type: int
6320
6321 /*
6322 * int c = ~ crc;
6323 * b = timesXtoThe32[(b ^ c) & 0xFF];
6324 * b = b ^ (c >>> 8);
6325 * crc = ~b;
6326 */
6327
6328 Node* M1 = intcon(-1);
6329 crc = _gvn.transform(new XorINode(crc, M1));
6330 Node* result = _gvn.transform(new XorINode(crc, b));
6331 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6332
6333 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6334 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6335 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6336 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6337
6338 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6339 result = _gvn.transform(new XorINode(crc, result));
6340 result = _gvn.transform(new XorINode(result, M1));
6341 set_result(result);
6342 return true;
6343 }
6344
6345 /**
6346 * Calculate CRC32 for byte[] array.
6347 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6348 */
6349 bool LibraryCallKit::inline_updateBytesCRC32() {
6350 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6351 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6352 // no receiver since it is static method
6353 Node* crc = argument(0); // type: int
6354 Node* src = argument(1); // type: oop
6355 Node* offset = argument(2); // type: int
6356 Node* length = argument(3); // type: int
6357
6358 const Type* src_type = src->Value(&_gvn);
6359 const TypeAryPtr* top_src = src_type->isa_aryptr();
6360 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6361 // failed array check
6362 return false;
6363 }
6364
6365 // Figure out the size and type of the elements we will be copying.
6366 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6367 if (src_elem != T_BYTE) {
6368 return false;
6369 }
6370
6371 // 'src_start' points to src array + scaled offset
6372 src = must_be_not_null(src, true);
6373 Node* src_start = array_element_address(src, offset, src_elem);
6374
6375 // We assume that range check is done by caller.
6376 // TODO: generate range check (offset+length < src.length) in debug VM.
6377
6378 // Call the stub.
6379 address stubAddr = StubRoutines::updateBytesCRC32();
6380 const char *stubName = "updateBytesCRC32";
6381
6382 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6383 stubAddr, stubName, TypePtr::BOTTOM,
6384 crc, src_start, length);
6385 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6386 set_result(result);
6387 return true;
6388 }
6389
6390 /**
6391 * Calculate CRC32 for ByteBuffer.
6392 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6393 */
6394 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6395 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6396 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6397 // no receiver since it is static method
6398 Node* crc = argument(0); // type: int
6399 Node* src = argument(1); // type: long
6400 Node* offset = argument(3); // type: int
6401 Node* length = argument(4); // type: int
6402
6403 src = ConvL2X(src); // adjust Java long to machine word
6404 Node* base = _gvn.transform(new CastX2PNode(src));
6405 offset = ConvI2X(offset);
6406
6407 // 'src_start' points to src array + scaled offset
6408 Node* src_start = basic_plus_adr(top(), base, offset);
6409
6410 // Call the stub.
6411 address stubAddr = StubRoutines::updateBytesCRC32();
6412 const char *stubName = "updateBytesCRC32";
6413
6414 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6415 stubAddr, stubName, TypePtr::BOTTOM,
6416 crc, src_start, length);
6417 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6418 set_result(result);
6419 return true;
6420 }
6421
6422 //------------------------------get_table_from_crc32c_class-----------------------
6423 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6424 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6425 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
6426
6427 return table;
6428 }
6429
6430 //------------------------------inline_updateBytesCRC32C-----------------------
6431 //
6432 // Calculate CRC32C for byte[] array.
6433 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6434 //
6435 bool LibraryCallKit::inline_updateBytesCRC32C() {
6436 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6437 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6438 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6439 // no receiver since it is a static method
6440 Node* crc = argument(0); // type: int
6441 Node* src = argument(1); // type: oop
6442 Node* offset = argument(2); // type: int
6443 Node* end = argument(3); // type: int
6444
6445 Node* length = _gvn.transform(new SubINode(end, offset));
6446
6447 const Type* src_type = src->Value(&_gvn);
6448 const TypeAryPtr* top_src = src_type->isa_aryptr();
6449 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6450 // failed array check
6451 return false;
6452 }
6453
6454 // Figure out the size and type of the elements we will be copying.
6455 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6456 if (src_elem != T_BYTE) {
6457 return false;
6458 }
6459
6460 // 'src_start' points to src array + scaled offset
6461 src = must_be_not_null(src, true);
6462 Node* src_start = array_element_address(src, offset, src_elem);
6463
6464 // static final int[] byteTable in class CRC32C
6465 Node* table = get_table_from_crc32c_class(callee()->holder());
6466 table = must_be_not_null(table, true);
6467 Node* table_start = array_element_address(table, intcon(0), T_INT);
6468
6469 // We assume that range check is done by caller.
6470 // TODO: generate range check (offset+length < src.length) in debug VM.
6471
6472 // Call the stub.
6473 address stubAddr = StubRoutines::updateBytesCRC32C();
6474 const char *stubName = "updateBytesCRC32C";
6475
6476 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6477 stubAddr, stubName, TypePtr::BOTTOM,
6478 crc, src_start, length, table_start);
6479 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6480 set_result(result);
6481 return true;
6482 }
6483
6484 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6485 //
6486 // Calculate CRC32C for DirectByteBuffer.
6487 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6488 //
6489 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6490 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6491 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6492 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6493 // no receiver since it is a static method
6494 Node* crc = argument(0); // type: int
6495 Node* src = argument(1); // type: long
6496 Node* offset = argument(3); // type: int
6497 Node* end = argument(4); // type: int
6498
6499 Node* length = _gvn.transform(new SubINode(end, offset));
6500
6501 src = ConvL2X(src); // adjust Java long to machine word
6502 Node* base = _gvn.transform(new CastX2PNode(src));
6503 offset = ConvI2X(offset);
6504
6505 // 'src_start' points to src array + scaled offset
6506 Node* src_start = basic_plus_adr(top(), base, offset);
6507
6508 // static final int[] byteTable in class CRC32C
6509 Node* table = get_table_from_crc32c_class(callee()->holder());
6510 table = must_be_not_null(table, true);
6511 Node* table_start = array_element_address(table, intcon(0), T_INT);
6512
6513 // Call the stub.
6514 address stubAddr = StubRoutines::updateBytesCRC32C();
6515 const char *stubName = "updateBytesCRC32C";
6516
6517 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6518 stubAddr, stubName, TypePtr::BOTTOM,
6519 crc, src_start, length, table_start);
6520 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6521 set_result(result);
6522 return true;
6523 }
6524
6525 //------------------------------inline_updateBytesAdler32----------------------
6526 //
6527 // Calculate Adler32 checksum for byte[] array.
6528 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6529 //
6530 bool LibraryCallKit::inline_updateBytesAdler32() {
6531 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6532 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6533 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6534 // no receiver since it is static method
6535 Node* crc = argument(0); // type: int
6536 Node* src = argument(1); // type: oop
6537 Node* offset = argument(2); // type: int
6538 Node* length = argument(3); // type: int
6539
6540 const Type* src_type = src->Value(&_gvn);
6541 const TypeAryPtr* top_src = src_type->isa_aryptr();
6542 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6543 // failed array check
6544 return false;
6545 }
6546
6547 // Figure out the size and type of the elements we will be copying.
6548 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6549 if (src_elem != T_BYTE) {
6550 return false;
6551 }
6552
6553 // 'src_start' points to src array + scaled offset
6554 Node* src_start = array_element_address(src, offset, src_elem);
6555
6556 // We assume that range check is done by caller.
6557 // TODO: generate range check (offset+length < src.length) in debug VM.
6558
6559 // Call the stub.
6560 address stubAddr = StubRoutines::updateBytesAdler32();
6561 const char *stubName = "updateBytesAdler32";
6562
6563 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6564 stubAddr, stubName, TypePtr::BOTTOM,
6565 crc, src_start, length);
6566 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6567 set_result(result);
6568 return true;
6569 }
6570
6571 //------------------------------inline_updateByteBufferAdler32---------------
6572 //
6573 // Calculate Adler32 checksum for DirectByteBuffer.
6574 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6575 //
6576 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6577 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6578 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6579 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6580 // no receiver since it is static method
6581 Node* crc = argument(0); // type: int
6582 Node* src = argument(1); // type: long
6583 Node* offset = argument(3); // type: int
6584 Node* length = argument(4); // type: int
6585
6586 src = ConvL2X(src); // adjust Java long to machine word
6587 Node* base = _gvn.transform(new CastX2PNode(src));
6588 offset = ConvI2X(offset);
6589
6590 // 'src_start' points to src array + scaled offset
6591 Node* src_start = basic_plus_adr(top(), base, offset);
6592
6593 // Call the stub.
6594 address stubAddr = StubRoutines::updateBytesAdler32();
6595 const char *stubName = "updateBytesAdler32";
6596
6597 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6598 stubAddr, stubName, TypePtr::BOTTOM,
6599 crc, src_start, length);
6600
6601 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6602 set_result(result);
6603 return true;
6604 }
6605
6606 //----------------------------inline_reference_get----------------------------
6607 // public T java.lang.ref.Reference.get();
6608 bool LibraryCallKit::inline_reference_get() {
6609 const int referent_offset = java_lang_ref_Reference::referent_offset();
6610
6611 // Get the argument:
6612 Node* reference_obj = null_check_receiver();
6613 if (stopped()) return true;
6614
6615 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6616 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6617 decorators, /*is_static*/ false, NULL);
6618 if (result == NULL) return false;
6619
6620 // Add memory barrier to prevent commoning reads from this field
6621 // across safepoint since GC can change its value.
6622 insert_mem_bar(Op_MemBarCPUOrder);
6623
6624 set_result(result);
6625 return true;
6626 }
6627
6628 //----------------------------inline_reference_refersTo0----------------------------
6629 // bool java.lang.ref.Reference.refersTo0();
6630 // bool java.lang.ref.PhantomReference.refersTo0();
6631 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6632 // Get arguments:
6633 Node* reference_obj = null_check_receiver();
6634 Node* other_obj = argument(1);
6635 if (stopped()) return true;
6636
6637 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6638 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6639 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6640 decorators, /*is_static*/ false, NULL);
6641 if (referent == NULL) return false;
6642
6643 // Add memory barrier to prevent commoning reads from this field
6644 // across safepoint since GC can change its value.
6645 insert_mem_bar(Op_MemBarCPUOrder);
6646
6647 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6648 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6649 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6650
6651 RegionNode* region = new RegionNode(3);
6652 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6653
6654 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6655 region->init_req(1, if_true);
6656 phi->init_req(1, intcon(1));
6657
6658 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6659 region->init_req(2, if_false);
6660 phi->init_req(2, intcon(0));
6661
6662 set_control(_gvn.transform(region));
6663 record_for_igvn(region);
6664 set_result(_gvn.transform(phi));
6665 return true;
6666 }
6667
6668
6669 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6670 DecoratorSet decorators, bool is_static,
6671 ciInstanceKlass* fromKls) {
6672 if (fromKls == NULL) {
6673 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6674 assert(tinst != NULL, "obj is null");
6675 assert(tinst->is_loaded(), "obj is not loaded");
6676 fromKls = tinst->instance_klass();
6677 } else {
6678 assert(is_static, "only for static field access");
6679 }
6680 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6681 ciSymbol::make(fieldTypeString),
6682 is_static);
6683
6684 assert(field != NULL, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6685 if (field == NULL) return (Node *) NULL;
6686
6687 if (is_static) {
6688 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6689 fromObj = makecon(tip);
6690 }
6691
6692 // Next code copied from Parse::do_get_xxx():
6693
6694 // Compute address and memory type.
6695 int offset = field->offset_in_bytes();
6696 bool is_vol = field->is_volatile();
6697 ciType* field_klass = field->type();
6698 assert(field_klass->is_loaded(), "should be loaded");
6699 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6700 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6701 BasicType bt = field->layout_type();
6702
6703 // Build the resultant type of the load
6704 const Type *type;
6705 if (bt == T_OBJECT) {
6706 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6707 } else {
6708 type = Type::get_const_basic_type(bt);
6709 }
6710
6711 if (is_vol) {
6712 decorators |= MO_SEQ_CST;
6713 }
6714
6715 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6716 }
6717
6718 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6719 bool is_exact /* true */, bool is_static /* false */,
6720 ciInstanceKlass * fromKls /* NULL */) {
6721 if (fromKls == NULL) {
6722 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6723 assert(tinst != NULL, "obj is null");
6724 assert(tinst->is_loaded(), "obj is not loaded");
6725 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6726 fromKls = tinst->instance_klass();
6727 }
6728 else {
6729 assert(is_static, "only for static field access");
6730 }
6731 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6732 ciSymbol::make(fieldTypeString),
6733 is_static);
6734
6735 assert(field != NULL, "undefined field");
6736 assert(!field->is_volatile(), "not defined for volatile fields");
6737
6738 if (is_static) {
6739 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6740 fromObj = makecon(tip);
6741 }
6742
6743 // Next code copied from Parse::do_get_xxx():
6744
6745 // Compute address and memory type.
6746 int offset = field->offset_in_bytes();
6747 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6748
6749 return adr;
6750 }
6751
6752 //------------------------------inline_aescrypt_Block-----------------------
6753 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6754 address stubAddr = NULL;
6755 const char *stubName;
6756 assert(UseAES, "need AES instruction support");
6757
6758 switch(id) {
6759 case vmIntrinsics::_aescrypt_encryptBlock:
6760 stubAddr = StubRoutines::aescrypt_encryptBlock();
6761 stubName = "aescrypt_encryptBlock";
6762 break;
6763 case vmIntrinsics::_aescrypt_decryptBlock:
6764 stubAddr = StubRoutines::aescrypt_decryptBlock();
6765 stubName = "aescrypt_decryptBlock";
6766 break;
6767 default:
6768 break;
6769 }
6770 if (stubAddr == NULL) return false;
6771
6772 Node* aescrypt_object = argument(0);
6773 Node* src = argument(1);
6774 Node* src_offset = argument(2);
6775 Node* dest = argument(3);
6776 Node* dest_offset = argument(4);
6777
6778 src = must_be_not_null(src, true);
6779 dest = must_be_not_null(dest, true);
6780
6781 // (1) src and dest are arrays.
6782 const Type* src_type = src->Value(&_gvn);
6783 const Type* dest_type = dest->Value(&_gvn);
6784 const TypeAryPtr* top_src = src_type->isa_aryptr();
6785 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6786 assert (top_src != NULL && top_src->elem() != Type::BOTTOM && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6787
6788 // for the quick and dirty code we will skip all the checks.
6789 // we are just trying to get the call to be generated.
6790 Node* src_start = src;
6791 Node* dest_start = dest;
6792 if (src_offset != NULL || dest_offset != NULL) {
6793 assert(src_offset != NULL && dest_offset != NULL, "");
6794 src_start = array_element_address(src, src_offset, T_BYTE);
6795 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6796 }
6797
6798 // now need to get the start of its expanded key array
6799 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6800 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6801 if (k_start == NULL) return false;
6802
6803 // Call the stub.
6804 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6805 stubAddr, stubName, TypePtr::BOTTOM,
6806 src_start, dest_start, k_start);
6807
6808 return true;
6809 }
6810
6811 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6812 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6813 address stubAddr = NULL;
6814 const char *stubName = NULL;
6815
6816 assert(UseAES, "need AES instruction support");
6817
6818 switch(id) {
6819 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6820 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6821 stubName = "cipherBlockChaining_encryptAESCrypt";
6822 break;
6823 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6824 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6825 stubName = "cipherBlockChaining_decryptAESCrypt";
6826 break;
6827 default:
6828 break;
6829 }
6830 if (stubAddr == NULL) return false;
6831
6832 Node* cipherBlockChaining_object = argument(0);
6833 Node* src = argument(1);
6834 Node* src_offset = argument(2);
6835 Node* len = argument(3);
6836 Node* dest = argument(4);
6837 Node* dest_offset = argument(5);
6838
6839 src = must_be_not_null(src, false);
6840 dest = must_be_not_null(dest, false);
6841
6842 // (1) src and dest are arrays.
6843 const Type* src_type = src->Value(&_gvn);
6844 const Type* dest_type = dest->Value(&_gvn);
6845 const TypeAryPtr* top_src = src_type->isa_aryptr();
6846 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6847 assert (top_src != NULL && top_src->elem() != Type::BOTTOM
6848 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6849
6850 // checks are the responsibility of the caller
6851 Node* src_start = src;
6852 Node* dest_start = dest;
6853 if (src_offset != NULL || dest_offset != NULL) {
6854 assert(src_offset != NULL && dest_offset != NULL, "");
6855 src_start = array_element_address(src, src_offset, T_BYTE);
6856 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6857 }
6858
6859 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6860 // (because of the predicated logic executed earlier).
6861 // so we cast it here safely.
6862 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6863
6864 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6865 if (embeddedCipherObj == NULL) return false;
6866
6867 // cast it to what we know it will be at runtime
6868 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6869 assert(tinst != NULL, "CBC obj is null");
6870 assert(tinst->is_loaded(), "CBC obj is not loaded");
6871 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6872 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6873
6874 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6875 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6876 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6877 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6878 aescrypt_object = _gvn.transform(aescrypt_object);
6879
6880 // we need to get the start of the aescrypt_object's expanded key array
6881 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6882 if (k_start == NULL) return false;
6883
6884 // similarly, get the start address of the r vector
6885 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6886 if (objRvec == NULL) return false;
6887 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6888
6889 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6890 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6891 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6892 stubAddr, stubName, TypePtr::BOTTOM,
6893 src_start, dest_start, k_start, r_start, len);
6894
6895 // return cipher length (int)
6896 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6897 set_result(retvalue);
6898 return true;
6899 }
6900
6901 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6902 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6903 address stubAddr = NULL;
6904 const char *stubName = NULL;
6905
6906 assert(UseAES, "need AES instruction support");
6907
6908 switch (id) {
6909 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6910 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6911 stubName = "electronicCodeBook_encryptAESCrypt";
6912 break;
6913 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6914 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6915 stubName = "electronicCodeBook_decryptAESCrypt";
6916 break;
6917 default:
6918 break;
6919 }
6920
6921 if (stubAddr == NULL) return false;
6922
6923 Node* electronicCodeBook_object = argument(0);
6924 Node* src = argument(1);
6925 Node* src_offset = argument(2);
6926 Node* len = argument(3);
6927 Node* dest = argument(4);
6928 Node* dest_offset = argument(5);
6929
6930 // (1) src and dest are arrays.
6931 const Type* src_type = src->Value(&_gvn);
6932 const Type* dest_type = dest->Value(&_gvn);
6933 const TypeAryPtr* top_src = src_type->isa_aryptr();
6934 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6935 assert(top_src != NULL && top_src->elem() != Type::BOTTOM
6936 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6937
6938 // checks are the responsibility of the caller
6939 Node* src_start = src;
6940 Node* dest_start = dest;
6941 if (src_offset != NULL || dest_offset != NULL) {
6942 assert(src_offset != NULL && dest_offset != NULL, "");
6943 src_start = array_element_address(src, src_offset, T_BYTE);
6944 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6945 }
6946
6947 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6948 // (because of the predicated logic executed earlier).
6949 // so we cast it here safely.
6950 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6951
6952 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6953 if (embeddedCipherObj == NULL) return false;
6954
6955 // cast it to what we know it will be at runtime
6956 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6957 assert(tinst != NULL, "ECB obj is null");
6958 assert(tinst->is_loaded(), "ECB obj is not loaded");
6959 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6960 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6961
6962 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6963 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6964 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6965 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6966 aescrypt_object = _gvn.transform(aescrypt_object);
6967
6968 // we need to get the start of the aescrypt_object's expanded key array
6969 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6970 if (k_start == NULL) return false;
6971
6972 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6973 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6974 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6975 stubAddr, stubName, TypePtr::BOTTOM,
6976 src_start, dest_start, k_start, len);
6977
6978 // return cipher length (int)
6979 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6980 set_result(retvalue);
6981 return true;
6982 }
6983
6984 //------------------------------inline_counterMode_AESCrypt-----------------------
6985 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6986 assert(UseAES, "need AES instruction support");
6987 if (!UseAESCTRIntrinsics) return false;
6988
6989 address stubAddr = NULL;
6990 const char *stubName = NULL;
6991 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6992 stubAddr = StubRoutines::counterMode_AESCrypt();
6993 stubName = "counterMode_AESCrypt";
6994 }
6995 if (stubAddr == NULL) return false;
6996
6997 Node* counterMode_object = argument(0);
6998 Node* src = argument(1);
6999 Node* src_offset = argument(2);
7000 Node* len = argument(3);
7001 Node* dest = argument(4);
7002 Node* dest_offset = argument(5);
7003
7004 // (1) src and dest are arrays.
7005 const Type* src_type = src->Value(&_gvn);
7006 const Type* dest_type = dest->Value(&_gvn);
7007 const TypeAryPtr* top_src = src_type->isa_aryptr();
7008 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
7009 assert(top_src != NULL && top_src->elem() != Type::BOTTOM &&
7010 top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
7011
7012 // checks are the responsibility of the caller
7013 Node* src_start = src;
7014 Node* dest_start = dest;
7015 if (src_offset != NULL || dest_offset != NULL) {
7016 assert(src_offset != NULL && dest_offset != NULL, "");
7017 src_start = array_element_address(src, src_offset, T_BYTE);
7018 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7019 }
7020
7021 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7022 // (because of the predicated logic executed earlier).
7023 // so we cast it here safely.
7024 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7025 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7026 if (embeddedCipherObj == NULL) return false;
7027 // cast it to what we know it will be at runtime
7028 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7029 assert(tinst != NULL, "CTR obj is null");
7030 assert(tinst->is_loaded(), "CTR obj is not loaded");
7031 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7032 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7033 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7034 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7035 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7036 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7037 aescrypt_object = _gvn.transform(aescrypt_object);
7038 // we need to get the start of the aescrypt_object's expanded key array
7039 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7040 if (k_start == NULL) return false;
7041 // similarly, get the start address of the r vector
7042 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7043 if (obj_counter == NULL) return false;
7044 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7045
7046 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7047 if (saved_encCounter == NULL) return false;
7048 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7049 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7050
7051 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7052 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7053 OptoRuntime::counterMode_aescrypt_Type(),
7054 stubAddr, stubName, TypePtr::BOTTOM,
7055 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7056
7057 // return cipher length (int)
7058 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7059 set_result(retvalue);
7060 return true;
7061 }
7062
7063 //------------------------------get_key_start_from_aescrypt_object-----------------------
7064 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7065 #if defined(PPC64) || defined(S390)
7066 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7067 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7068 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7069 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7070 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7071 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
7072 if (objSessionK == NULL) {
7073 return (Node *) NULL;
7074 }
7075 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7076 #else
7077 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7078 #endif // PPC64
7079 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
7080 if (objAESCryptKey == NULL) return (Node *) NULL;
7081
7082 // now have the array, need to get the start address of the K array
7083 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7084 return k_start;
7085 }
7086
7087 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7088 // Return node representing slow path of predicate check.
7089 // the pseudo code we want to emulate with this predicate is:
7090 // for encryption:
7091 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7092 // for decryption:
7093 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7094 // note cipher==plain is more conservative than the original java code but that's OK
7095 //
7096 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7097 // The receiver was checked for NULL already.
7098 Node* objCBC = argument(0);
7099
7100 Node* src = argument(1);
7101 Node* dest = argument(4);
7102
7103 // Load embeddedCipher field of CipherBlockChaining object.
7104 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7105
7106 // get AESCrypt klass for instanceOf check
7107 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7108 // will have same classloader as CipherBlockChaining object
7109 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7110 assert(tinst != NULL, "CBCobj is null");
7111 assert(tinst->is_loaded(), "CBCobj is not loaded");
7112
7113 // we want to do an instanceof comparison against the AESCrypt class
7114 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7115 if (!klass_AESCrypt->is_loaded()) {
7116 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7117 Node* ctrl = control();
7118 set_control(top()); // no regular fast path
7119 return ctrl;
7120 }
7121
7122 src = must_be_not_null(src, true);
7123 dest = must_be_not_null(dest, true);
7124
7125 // Resolve oops to stable for CmpP below.
7126 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7127
7128 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7129 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7130 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7131
7132 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7133
7134 // for encryption, we are done
7135 if (!decrypting)
7136 return instof_false; // even if it is NULL
7137
7138 // for decryption, we need to add a further check to avoid
7139 // taking the intrinsic path when cipher and plain are the same
7140 // see the original java code for why.
7141 RegionNode* region = new RegionNode(3);
7142 region->init_req(1, instof_false);
7143
7144 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7145 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7146 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
7147 region->init_req(2, src_dest_conjoint);
7148
7149 record_for_igvn(region);
7150 return _gvn.transform(region);
7151 }
7152
7153 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7154 // Return node representing slow path of predicate check.
7155 // the pseudo code we want to emulate with this predicate is:
7156 // for encryption:
7157 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7158 // for decryption:
7159 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7160 // note cipher==plain is more conservative than the original java code but that's OK
7161 //
7162 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7163 // The receiver was checked for NULL already.
7164 Node* objECB = argument(0);
7165
7166 // Load embeddedCipher field of ElectronicCodeBook object.
7167 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7168
7169 // get AESCrypt klass for instanceOf check
7170 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7171 // will have same classloader as ElectronicCodeBook object
7172 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7173 assert(tinst != NULL, "ECBobj is null");
7174 assert(tinst->is_loaded(), "ECBobj is not loaded");
7175
7176 // we want to do an instanceof comparison against the AESCrypt class
7177 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7178 if (!klass_AESCrypt->is_loaded()) {
7179 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7180 Node* ctrl = control();
7181 set_control(top()); // no regular fast path
7182 return ctrl;
7183 }
7184 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7185
7186 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7187 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7188 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7189
7190 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7191
7192 // for encryption, we are done
7193 if (!decrypting)
7194 return instof_false; // even if it is NULL
7195
7196 // for decryption, we need to add a further check to avoid
7197 // taking the intrinsic path when cipher and plain are the same
7198 // see the original java code for why.
7199 RegionNode* region = new RegionNode(3);
7200 region->init_req(1, instof_false);
7201 Node* src = argument(1);
7202 Node* dest = argument(4);
7203 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7204 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7205 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
7206 region->init_req(2, src_dest_conjoint);
7207
7208 record_for_igvn(region);
7209 return _gvn.transform(region);
7210 }
7211
7212 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7213 // Return node representing slow path of predicate check.
7214 // the pseudo code we want to emulate with this predicate is:
7215 // for encryption:
7216 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7217 // for decryption:
7218 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7219 // note cipher==plain is more conservative than the original java code but that's OK
7220 //
7221
7222 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7223 // The receiver was checked for NULL already.
7224 Node* objCTR = argument(0);
7225
7226 // Load embeddedCipher field of CipherBlockChaining object.
7227 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7228
7229 // get AESCrypt klass for instanceOf check
7230 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7231 // will have same classloader as CipherBlockChaining object
7232 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7233 assert(tinst != NULL, "CTRobj is null");
7234 assert(tinst->is_loaded(), "CTRobj is not loaded");
7235
7236 // we want to do an instanceof comparison against the AESCrypt class
7237 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7238 if (!klass_AESCrypt->is_loaded()) {
7239 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7240 Node* ctrl = control();
7241 set_control(top()); // no regular fast path
7242 return ctrl;
7243 }
7244
7245 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7246 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7247 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7248 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7249 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7250
7251 return instof_false; // even if it is NULL
7252 }
7253
7254 //------------------------------inline_ghash_processBlocks
7255 bool LibraryCallKit::inline_ghash_processBlocks() {
7256 address stubAddr;
7257 const char *stubName;
7258 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7259
7260 stubAddr = StubRoutines::ghash_processBlocks();
7261 stubName = "ghash_processBlocks";
7262
7263 Node* data = argument(0);
7264 Node* offset = argument(1);
7265 Node* len = argument(2);
7266 Node* state = argument(3);
7267 Node* subkeyH = argument(4);
7268
7269 state = must_be_not_null(state, true);
7270 subkeyH = must_be_not_null(subkeyH, true);
7271 data = must_be_not_null(data, true);
7272
7273 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7274 assert(state_start, "state is NULL");
7275 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7276 assert(subkeyH_start, "subkeyH is NULL");
7277 Node* data_start = array_element_address(data, offset, T_BYTE);
7278 assert(data_start, "data is NULL");
7279
7280 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7281 OptoRuntime::ghash_processBlocks_Type(),
7282 stubAddr, stubName, TypePtr::BOTTOM,
7283 state_start, subkeyH_start, data_start, len);
7284 return true;
7285 }
7286
7287 bool LibraryCallKit::inline_base64_encodeBlock() {
7288 address stubAddr;
7289 const char *stubName;
7290 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7291 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7292 stubAddr = StubRoutines::base64_encodeBlock();
7293 stubName = "encodeBlock";
7294
7295 if (!stubAddr) return false;
7296 Node* base64obj = argument(0);
7297 Node* src = argument(1);
7298 Node* offset = argument(2);
7299 Node* len = argument(3);
7300 Node* dest = argument(4);
7301 Node* dp = argument(5);
7302 Node* isURL = argument(6);
7303
7304 src = must_be_not_null(src, true);
7305 dest = must_be_not_null(dest, true);
7306
7307 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7308 assert(src_start, "source array is NULL");
7309 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7310 assert(dest_start, "destination array is NULL");
7311
7312 Node* base64 = make_runtime_call(RC_LEAF,
7313 OptoRuntime::base64_encodeBlock_Type(),
7314 stubAddr, stubName, TypePtr::BOTTOM,
7315 src_start, offset, len, dest_start, dp, isURL);
7316 return true;
7317 }
7318
7319 bool LibraryCallKit::inline_base64_decodeBlock() {
7320 address stubAddr;
7321 const char *stubName;
7322 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7323 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7324 stubAddr = StubRoutines::base64_decodeBlock();
7325 stubName = "decodeBlock";
7326
7327 if (!stubAddr) return false;
7328 Node* base64obj = argument(0);
7329 Node* src = argument(1);
7330 Node* src_offset = argument(2);
7331 Node* len = argument(3);
7332 Node* dest = argument(4);
7333 Node* dest_offset = argument(5);
7334 Node* isURL = argument(6);
7335 Node* isMIME = argument(7);
7336
7337 src = must_be_not_null(src, true);
7338 dest = must_be_not_null(dest, true);
7339
7340 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7341 assert(src_start, "source array is NULL");
7342 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7343 assert(dest_start, "destination array is NULL");
7344
7345 Node* call = make_runtime_call(RC_LEAF,
7346 OptoRuntime::base64_decodeBlock_Type(),
7347 stubAddr, stubName, TypePtr::BOTTOM,
7348 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7349 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7350 set_result(result);
7351 return true;
7352 }
7353
7354 //------------------------------inline_digestBase_implCompress-----------------------
7355 //
7356 // Calculate MD5 for single-block byte[] array.
7357 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7358 //
7359 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7360 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7361 //
7362 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7363 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7364 //
7365 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7366 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7367 //
7368 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7369 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7370 //
7371 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7372 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7373
7374 Node* digestBase_obj = argument(0);
7375 Node* src = argument(1); // type oop
7376 Node* ofs = argument(2); // type int
7377
7378 const Type* src_type = src->Value(&_gvn);
7379 const TypeAryPtr* top_src = src_type->isa_aryptr();
7380 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7381 // failed array check
7382 return false;
7383 }
7384 // Figure out the size and type of the elements we will be copying.
7385 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7386 if (src_elem != T_BYTE) {
7387 return false;
7388 }
7389 // 'src_start' points to src array + offset
7390 src = must_be_not_null(src, true);
7391 Node* src_start = array_element_address(src, ofs, src_elem);
7392 Node* state = NULL;
7393 Node* digest_length = NULL;
7394 address stubAddr;
7395 const char *stubName;
7396
7397 switch(id) {
7398 case vmIntrinsics::_md5_implCompress:
7399 assert(UseMD5Intrinsics, "need MD5 instruction support");
7400 state = get_state_from_digest_object(digestBase_obj, T_INT);
7401 stubAddr = StubRoutines::md5_implCompress();
7402 stubName = "md5_implCompress";
7403 break;
7404 case vmIntrinsics::_sha_implCompress:
7405 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7406 state = get_state_from_digest_object(digestBase_obj, T_INT);
7407 stubAddr = StubRoutines::sha1_implCompress();
7408 stubName = "sha1_implCompress";
7409 break;
7410 case vmIntrinsics::_sha2_implCompress:
7411 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7412 state = get_state_from_digest_object(digestBase_obj, T_INT);
7413 stubAddr = StubRoutines::sha256_implCompress();
7414 stubName = "sha256_implCompress";
7415 break;
7416 case vmIntrinsics::_sha5_implCompress:
7417 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7418 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7419 stubAddr = StubRoutines::sha512_implCompress();
7420 stubName = "sha512_implCompress";
7421 break;
7422 case vmIntrinsics::_sha3_implCompress:
7423 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7424 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7425 stubAddr = StubRoutines::sha3_implCompress();
7426 stubName = "sha3_implCompress";
7427 digest_length = get_digest_length_from_digest_object(digestBase_obj);
7428 if (digest_length == NULL) return false;
7429 break;
7430 default:
7431 fatal_unexpected_iid(id);
7432 return false;
7433 }
7434 if (state == NULL) return false;
7435
7436 assert(stubAddr != NULL, "Stub is generated");
7437 if (stubAddr == NULL) return false;
7438
7439 // Call the stub.
7440 Node* call;
7441 if (digest_length == NULL) {
7442 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7443 stubAddr, stubName, TypePtr::BOTTOM,
7444 src_start, state);
7445 } else {
7446 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7447 stubAddr, stubName, TypePtr::BOTTOM,
7448 src_start, state, digest_length);
7449 }
7450
7451 return true;
7452 }
7453
7454 //------------------------------inline_digestBase_implCompressMB-----------------------
7455 //
7456 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7457 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7458 //
7459 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7460 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7461 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7462 assert((uint)predicate < 5, "sanity");
7463 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7464
7465 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7466 Node* src = argument(1); // byte[] array
7467 Node* ofs = argument(2); // type int
7468 Node* limit = argument(3); // type int
7469
7470 const Type* src_type = src->Value(&_gvn);
7471 const TypeAryPtr* top_src = src_type->isa_aryptr();
7472 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7473 // failed array check
7474 return false;
7475 }
7476 // Figure out the size and type of the elements we will be copying.
7477 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7478 if (src_elem != T_BYTE) {
7479 return false;
7480 }
7481 // 'src_start' points to src array + offset
7482 src = must_be_not_null(src, false);
7483 Node* src_start = array_element_address(src, ofs, src_elem);
7484
7485 const char* klass_digestBase_name = NULL;
7486 const char* stub_name = NULL;
7487 address stub_addr = NULL;
7488 BasicType elem_type = T_INT;
7489
7490 switch (predicate) {
7491 case 0:
7492 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7493 klass_digestBase_name = "sun/security/provider/MD5";
7494 stub_name = "md5_implCompressMB";
7495 stub_addr = StubRoutines::md5_implCompressMB();
7496 }
7497 break;
7498 case 1:
7499 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7500 klass_digestBase_name = "sun/security/provider/SHA";
7501 stub_name = "sha1_implCompressMB";
7502 stub_addr = StubRoutines::sha1_implCompressMB();
7503 }
7504 break;
7505 case 2:
7506 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7507 klass_digestBase_name = "sun/security/provider/SHA2";
7508 stub_name = "sha256_implCompressMB";
7509 stub_addr = StubRoutines::sha256_implCompressMB();
7510 }
7511 break;
7512 case 3:
7513 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7514 klass_digestBase_name = "sun/security/provider/SHA5";
7515 stub_name = "sha512_implCompressMB";
7516 stub_addr = StubRoutines::sha512_implCompressMB();
7517 elem_type = T_LONG;
7518 }
7519 break;
7520 case 4:
7521 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7522 klass_digestBase_name = "sun/security/provider/SHA3";
7523 stub_name = "sha3_implCompressMB";
7524 stub_addr = StubRoutines::sha3_implCompressMB();
7525 elem_type = T_BYTE;
7526 }
7527 break;
7528 default:
7529 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7530 }
7531 if (klass_digestBase_name != NULL) {
7532 assert(stub_addr != NULL, "Stub is generated");
7533 if (stub_addr == NULL) return false;
7534
7535 // get DigestBase klass to lookup for SHA klass
7536 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7537 assert(tinst != NULL, "digestBase_obj is not instance???");
7538 assert(tinst->is_loaded(), "DigestBase is not loaded");
7539
7540 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7541 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7542 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7543 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7544 }
7545 return false;
7546 }
7547
7548 //------------------------------inline_digestBase_implCompressMB-----------------------
7549 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7550 BasicType elem_type, address stubAddr, const char *stubName,
7551 Node* src_start, Node* ofs, Node* limit) {
7552 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7553 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7554 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7555 digest_obj = _gvn.transform(digest_obj);
7556
7557 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7558 if (state == NULL) return false;
7559
7560 Node* digest_length = NULL;
7561 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7562 digest_length = get_digest_length_from_digest_object(digest_obj);
7563 if (digest_length == NULL) return false;
7564 }
7565
7566 // Call the stub.
7567 Node* call;
7568 if (digest_length == NULL) {
7569 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7570 OptoRuntime::digestBase_implCompressMB_Type(false),
7571 stubAddr, stubName, TypePtr::BOTTOM,
7572 src_start, state, ofs, limit);
7573 } else {
7574 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7575 OptoRuntime::digestBase_implCompressMB_Type(true),
7576 stubAddr, stubName, TypePtr::BOTTOM,
7577 src_start, state, digest_length, ofs, limit);
7578 }
7579
7580 // return ofs (int)
7581 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7582 set_result(result);
7583
7584 return true;
7585 }
7586
7587 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7588 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7589 assert(UseAES, "need AES instruction support");
7590 address stubAddr = NULL;
7591 const char *stubName = NULL;
7592 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7593 stubName = "galoisCounterMode_AESCrypt";
7594
7595 if (stubAddr == NULL) return false;
7596
7597 Node* in = argument(0);
7598 Node* inOfs = argument(1);
7599 Node* len = argument(2);
7600 Node* ct = argument(3);
7601 Node* ctOfs = argument(4);
7602 Node* out = argument(5);
7603 Node* outOfs = argument(6);
7604 Node* gctr_object = argument(7);
7605 Node* ghash_object = argument(8);
7606
7607 // (1) in, ct and out are arrays.
7608 const Type* in_type = in->Value(&_gvn);
7609 const Type* ct_type = ct->Value(&_gvn);
7610 const Type* out_type = out->Value(&_gvn);
7611 const TypeAryPtr* top_in = in_type->isa_aryptr();
7612 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7613 const TypeAryPtr* top_out = out_type->isa_aryptr();
7614 assert(top_in != NULL && top_in->elem() != Type::BOTTOM &&
7615 top_ct != NULL && top_ct->elem() != Type::BOTTOM &&
7616 top_out != NULL && top_out->elem() != Type::BOTTOM, "args are strange");
7617
7618 // checks are the responsibility of the caller
7619 Node* in_start = in;
7620 Node* ct_start = ct;
7621 Node* out_start = out;
7622 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7623 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7624 in_start = array_element_address(in, inOfs, T_BYTE);
7625 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7626 out_start = array_element_address(out, outOfs, T_BYTE);
7627 }
7628
7629 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7630 // (because of the predicated logic executed earlier).
7631 // so we cast it here safely.
7632 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7633 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7634 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7635 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7636 Node* state = load_field_from_object(ghash_object, "state", "[J");
7637
7638 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7639 return false;
7640 }
7641 // cast it to what we know it will be at runtime
7642 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7643 assert(tinst != NULL, "GCTR obj is null");
7644 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7645 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7646 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7647 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7648 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7649 const TypeOopPtr* xtype = aklass->as_instance_type();
7650 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7651 aescrypt_object = _gvn.transform(aescrypt_object);
7652 // we need to get the start of the aescrypt_object's expanded key array
7653 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7654 if (k_start == NULL) return false;
7655 // similarly, get the start address of the r vector
7656 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7657 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7658 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7659
7660
7661 // Call the stub, passing params
7662 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7663 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7664 stubAddr, stubName, TypePtr::BOTTOM,
7665 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7666
7667 // return cipher length (int)
7668 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7669 set_result(retvalue);
7670
7671 return true;
7672 }
7673
7674 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7675 // Return node representing slow path of predicate check.
7676 // the pseudo code we want to emulate with this predicate is:
7677 // for encryption:
7678 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7679 // for decryption:
7680 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7681 // note cipher==plain is more conservative than the original java code but that's OK
7682 //
7683
7684 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7685 // The receiver was checked for NULL already.
7686 Node* objGCTR = argument(7);
7687 // Load embeddedCipher field of GCTR object.
7688 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7689 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7690
7691 // get AESCrypt klass for instanceOf check
7692 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7693 // will have same classloader as CipherBlockChaining object
7694 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7695 assert(tinst != NULL, "GCTR obj is null");
7696 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7697
7698 // we want to do an instanceof comparison against the AESCrypt class
7699 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7700 if (!klass_AESCrypt->is_loaded()) {
7701 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7702 Node* ctrl = control();
7703 set_control(top()); // no regular fast path
7704 return ctrl;
7705 }
7706
7707 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7708 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7709 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7710 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7711 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7712
7713 return instof_false; // even if it is NULL
7714 }
7715
7716 //------------------------------get_state_from_digest_object-----------------------
7717 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7718 const char* state_type;
7719 switch (elem_type) {
7720 case T_BYTE: state_type = "[B"; break;
7721 case T_INT: state_type = "[I"; break;
7722 case T_LONG: state_type = "[J"; break;
7723 default: ShouldNotReachHere();
7724 }
7725 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7726 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7727 if (digest_state == NULL) return (Node *) NULL;
7728
7729 // now have the array, need to get the start address of the state array
7730 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7731 return state;
7732 }
7733
7734 //------------------------------get_digest_length_from_sha3_object----------------------------------
7735 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
7736 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
7737 assert (digest_length != NULL, "sanity");
7738 return digest_length;
7739 }
7740
7741 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7742 // Return node representing slow path of predicate check.
7743 // the pseudo code we want to emulate with this predicate is:
7744 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7745 //
7746 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7747 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7748 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7749 assert((uint)predicate < 5, "sanity");
7750
7751 // The receiver was checked for NULL already.
7752 Node* digestBaseObj = argument(0);
7753
7754 // get DigestBase klass for instanceOf check
7755 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7756 assert(tinst != NULL, "digestBaseObj is null");
7757 assert(tinst->is_loaded(), "DigestBase is not loaded");
7758
7759 const char* klass_name = NULL;
7760 switch (predicate) {
7761 case 0:
7762 if (UseMD5Intrinsics) {
7763 // we want to do an instanceof comparison against the MD5 class
7764 klass_name = "sun/security/provider/MD5";
7765 }
7766 break;
7767 case 1:
7768 if (UseSHA1Intrinsics) {
7769 // we want to do an instanceof comparison against the SHA class
7770 klass_name = "sun/security/provider/SHA";
7771 }
7772 break;
7773 case 2:
7774 if (UseSHA256Intrinsics) {
7775 // we want to do an instanceof comparison against the SHA2 class
7776 klass_name = "sun/security/provider/SHA2";
7777 }
7778 break;
7779 case 3:
7780 if (UseSHA512Intrinsics) {
7781 // we want to do an instanceof comparison against the SHA5 class
7782 klass_name = "sun/security/provider/SHA5";
7783 }
7784 break;
7785 case 4:
7786 if (UseSHA3Intrinsics) {
7787 // we want to do an instanceof comparison against the SHA3 class
7788 klass_name = "sun/security/provider/SHA3";
7789 }
7790 break;
7791 default:
7792 fatal("unknown SHA intrinsic predicate: %d", predicate);
7793 }
7794
7795 ciKlass* klass = NULL;
7796 if (klass_name != NULL) {
7797 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7798 }
7799 if ((klass == NULL) || !klass->is_loaded()) {
7800 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7801 Node* ctrl = control();
7802 set_control(top()); // no intrinsic path
7803 return ctrl;
7804 }
7805 ciInstanceKlass* instklass = klass->as_instance_klass();
7806
7807 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7808 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7809 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7810 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7811
7812 return instof_false; // even if it is NULL
7813 }
7814
7815 //-------------inline_fma-----------------------------------
7816 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7817 Node *a = NULL;
7818 Node *b = NULL;
7819 Node *c = NULL;
7820 Node* result = NULL;
7821 switch (id) {
7822 case vmIntrinsics::_fmaD:
7823 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7824 // no receiver since it is static method
7825 a = round_double_node(argument(0));
7826 b = round_double_node(argument(2));
7827 c = round_double_node(argument(4));
7828 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7829 break;
7830 case vmIntrinsics::_fmaF:
7831 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7832 a = argument(0);
7833 b = argument(1);
7834 c = argument(2);
7835 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7836 break;
7837 default:
7838 fatal_unexpected_iid(id); break;
7839 }
7840 set_result(result);
7841 return true;
7842 }
7843
7844 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7845 // argument(0) is receiver
7846 Node* codePoint = argument(1);
7847 Node* n = NULL;
7848
7849 switch (id) {
7850 case vmIntrinsics::_isDigit :
7851 n = new DigitNode(control(), codePoint);
7852 break;
7853 case vmIntrinsics::_isLowerCase :
7854 n = new LowerCaseNode(control(), codePoint);
7855 break;
7856 case vmIntrinsics::_isUpperCase :
7857 n = new UpperCaseNode(control(), codePoint);
7858 break;
7859 case vmIntrinsics::_isWhitespace :
7860 n = new WhitespaceNode(control(), codePoint);
7861 break;
7862 default:
7863 fatal_unexpected_iid(id);
7864 }
7865
7866 set_result(_gvn.transform(n));
7867 return true;
7868 }
7869
7870 //------------------------------inline_fp_min_max------------------------------
7871 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7872 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7873
7874 // The intrinsic should be used only when the API branches aren't predictable,
7875 // the last one performing the most important comparison. The following heuristic
7876 // uses the branch statistics to eventually bail out if necessary.
7877
7878 ciMethodData *md = callee()->method_data();
7879
7880 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7881 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7882
7883 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7884 // Bail out if the call-site didn't contribute enough to the statistics.
7885 return false;
7886 }
7887
7888 uint taken = 0, not_taken = 0;
7889
7890 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7891 if (p->is_BranchData()) {
7892 taken = ((ciBranchData*)p)->taken();
7893 not_taken = ((ciBranchData*)p)->not_taken();
7894 }
7895 }
7896
7897 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7898 balance = balance < 0 ? -balance : balance;
7899 if ( balance > 0.2 ) {
7900 // Bail out if the most important branch is predictable enough.
7901 return false;
7902 }
7903 }
7904 */
7905
7906 Node *a = NULL;
7907 Node *b = NULL;
7908 Node *n = NULL;
7909 switch (id) {
7910 case vmIntrinsics::_maxF:
7911 case vmIntrinsics::_minF:
7912 case vmIntrinsics::_maxF_strict:
7913 case vmIntrinsics::_minF_strict:
7914 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7915 a = argument(0);
7916 b = argument(1);
7917 break;
7918 case vmIntrinsics::_maxD:
7919 case vmIntrinsics::_minD:
7920 case vmIntrinsics::_maxD_strict:
7921 case vmIntrinsics::_minD_strict:
7922 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7923 a = round_double_node(argument(0));
7924 b = round_double_node(argument(2));
7925 break;
7926 default:
7927 fatal_unexpected_iid(id);
7928 break;
7929 }
7930 switch (id) {
7931 case vmIntrinsics::_maxF:
7932 case vmIntrinsics::_maxF_strict:
7933 n = new MaxFNode(a, b);
7934 break;
7935 case vmIntrinsics::_minF:
7936 case vmIntrinsics::_minF_strict:
7937 n = new MinFNode(a, b);
7938 break;
7939 case vmIntrinsics::_maxD:
7940 case vmIntrinsics::_maxD_strict:
7941 n = new MaxDNode(a, b);
7942 break;
7943 case vmIntrinsics::_minD:
7944 case vmIntrinsics::_minD_strict:
7945 n = new MinDNode(a, b);
7946 break;
7947 default:
7948 fatal_unexpected_iid(id);
7949 break;
7950 }
7951 set_result(_gvn.transform(n));
7952 return true;
7953 }
7954
7955 bool LibraryCallKit::inline_profileBoolean() {
7956 Node* counts = argument(1);
7957 const TypeAryPtr* ary = NULL;
7958 ciArray* aobj = NULL;
7959 if (counts->is_Con()
7960 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7961 && (aobj = ary->const_oop()->as_array()) != NULL
7962 && (aobj->length() == 2)) {
7963 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7964 jint false_cnt = aobj->element_value(0).as_int();
7965 jint true_cnt = aobj->element_value(1).as_int();
7966
7967 if (C->log() != NULL) {
7968 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7969 false_cnt, true_cnt);
7970 }
7971
7972 if (false_cnt + true_cnt == 0) {
7973 // According to profile, never executed.
7974 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7975 Deoptimization::Action_reinterpret);
7976 return true;
7977 }
7978
7979 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7980 // is a number of each value occurrences.
7981 Node* result = argument(0);
7982 if (false_cnt == 0 || true_cnt == 0) {
7983 // According to profile, one value has been never seen.
7984 int expected_val = (false_cnt == 0) ? 1 : 0;
7985
7986 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7987 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7988
7989 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7990 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7991 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7992
7993 { // Slow path: uncommon trap for never seen value and then reexecute
7994 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7995 // the value has been seen at least once.
7996 PreserveJVMState pjvms(this);
7997 PreserveReexecuteState preexecs(this);
7998 jvms()->set_should_reexecute(true);
7999
8000 set_control(slow_path);
8001 set_i_o(i_o());
8002
8003 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8004 Deoptimization::Action_reinterpret);
8005 }
8006 // The guard for never seen value enables sharpening of the result and
8007 // returning a constant. It allows to eliminate branches on the same value
8008 // later on.
8009 set_control(fast_path);
8010 result = intcon(expected_val);
8011 }
8012 // Stop profiling.
8013 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8014 // By replacing method body with profile data (represented as ProfileBooleanNode
8015 // on IR level) we effectively disable profiling.
8016 // It enables full speed execution once optimized code is generated.
8017 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8018 C->record_for_igvn(profile);
8019 set_result(profile);
8020 return true;
8021 } else {
8022 // Continue profiling.
8023 // Profile data isn't available at the moment. So, execute method's bytecode version.
8024 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8025 // is compiled and counters aren't available since corresponding MethodHandle
8026 // isn't a compile-time constant.
8027 return false;
8028 }
8029 }
8030
8031 bool LibraryCallKit::inline_isCompileConstant() {
8032 Node* n = argument(0);
8033 set_result(n->is_Con() ? intcon(1) : intcon(0));
8034 return true;
8035 }
8036
8037 //------------------------------- inline_getObjectSize --------------------------------------
8038 //
8039 // Calculate the runtime size of the object/array.
8040 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8041 //
8042 bool LibraryCallKit::inline_getObjectSize() {
8043 Node* obj = argument(3);
8044 Node* klass_node = load_object_klass(obj);
8045
8046 jint layout_con = Klass::_lh_neutral_value;
8047 Node* layout_val = get_layout_helper(klass_node, layout_con);
8048 int layout_is_con = (layout_val == NULL);
8049
8050 if (layout_is_con) {
8051 // Layout helper is constant, can figure out things at compile time.
8052
8053 if (Klass::layout_helper_is_instance(layout_con)) {
8054 // Instance case: layout_con contains the size itself.
8055 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8056 set_result(size);
8057 } else {
8058 // Array case: size is round(header + element_size*arraylength).
8059 // Since arraylength is different for every array instance, we have to
8060 // compute the whole thing at runtime.
8061
8062 Node* arr_length = load_array_length(obj);
8063
8064 int round_mask = MinObjAlignmentInBytes - 1;
8065 int hsize = Klass::layout_helper_header_size(layout_con);
8066 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8067
8068 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8069 round_mask = 0; // strength-reduce it if it goes away completely
8070 }
8071 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8072 Node* header_size = intcon(hsize + round_mask);
8073
8074 Node* lengthx = ConvI2X(arr_length);
8075 Node* headerx = ConvI2X(header_size);
8076
8077 Node* abody = lengthx;
8078 if (eshift != 0) {
8079 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8080 }
8081 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8082 if (round_mask != 0) {
8083 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8084 }
8085 size = ConvX2L(size);
8086 set_result(size);
8087 }
8088 } else {
8089 // Layout helper is not constant, need to test for array-ness at runtime.
8090
8091 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8092 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8093 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8094 record_for_igvn(result_reg);
8095
8096 Node* array_ctl = generate_array_guard(klass_node, NULL);
8097 if (array_ctl != NULL) {
8098 // Array case: size is round(header + element_size*arraylength).
8099 // Since arraylength is different for every array instance, we have to
8100 // compute the whole thing at runtime.
8101
8102 PreserveJVMState pjvms(this);
8103 set_control(array_ctl);
8104 Node* arr_length = load_array_length(obj);
8105
8106 int round_mask = MinObjAlignmentInBytes - 1;
8107 Node* mask = intcon(round_mask);
8108
8109 Node* hss = intcon(Klass::_lh_header_size_shift);
8110 Node* hsm = intcon(Klass::_lh_header_size_mask);
8111 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8112 header_size = _gvn.transform(new AndINode(header_size, hsm));
8113 header_size = _gvn.transform(new AddINode(header_size, mask));
8114
8115 // There is no need to mask or shift this value.
8116 // The semantics of LShiftINode include an implicit mask to 0x1F.
8117 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8118 Node* elem_shift = layout_val;
8119
8120 Node* lengthx = ConvI2X(arr_length);
8121 Node* headerx = ConvI2X(header_size);
8122
8123 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8124 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8125 if (round_mask != 0) {
8126 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8127 }
8128 size = ConvX2L(size);
8129
8130 result_reg->init_req(_array_path, control());
8131 result_val->init_req(_array_path, size);
8132 }
8133
8134 if (!stopped()) {
8135 // Instance case: the layout helper gives us instance size almost directly,
8136 // but we need to mask out the _lh_instance_slow_path_bit.
8137 Node* size = ConvI2X(layout_val);
8138 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8139 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8140 size = _gvn.transform(new AndXNode(size, mask));
8141 size = ConvX2L(size);
8142
8143 result_reg->init_req(_instance_path, control());
8144 result_val->init_req(_instance_path, size);
8145 }
8146
8147 set_result(result_reg, result_val);
8148 }
8149
8150 return true;
8151 }
8152
8153 //------------------------------- inline_blackhole --------------------------------------
8154 //
8155 // Make sure all arguments to this node are alive.
8156 // This matches methods that were requested to be blackholed through compile commands.
8157 //
8158 bool LibraryCallKit::inline_blackhole() {
8159 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8160 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8161 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8162
8163 // Bind call arguments as blackhole arguments to keep them alive
8164 Node* bh = insert_mem_bar(Op_Blackhole);
8165 uint nargs = callee()->arg_size();
8166 for (uint i = 0; i < nargs; i++) {
8167 bh->add_req(argument(i));
8168 }
8169
8170 return true;
8171 }