1 /*
2 * Copyright (c) 1999, 2023, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dabs:
258 case vmIntrinsics::_fabs:
259 case vmIntrinsics::_iabs:
260 case vmIntrinsics::_labs:
261 case vmIntrinsics::_datan2:
262 case vmIntrinsics::_dsqrt:
263 case vmIntrinsics::_dsqrt_strict:
264 case vmIntrinsics::_dexp:
265 case vmIntrinsics::_dlog:
266 case vmIntrinsics::_dlog10:
267 case vmIntrinsics::_dpow:
268 case vmIntrinsics::_dcopySign:
269 case vmIntrinsics::_fcopySign:
270 case vmIntrinsics::_dsignum:
271 case vmIntrinsics::_roundF:
272 case vmIntrinsics::_roundD:
273 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
274
275 case vmIntrinsics::_notify:
276 case vmIntrinsics::_notifyAll:
277 return inline_notify(intrinsic_id());
278
279 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
280 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
281 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
282 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
283 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
284 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
285 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
286 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
287 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
288 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
289 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
290 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
291 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
292 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
293
294 case vmIntrinsics::_arraycopy: return inline_arraycopy();
295
296 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
297 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
298 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
299 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
300
301 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
308 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
309
310 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
311 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
312
313 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
314
315 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
316 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
317 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
318 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
319
320 case vmIntrinsics::_compressStringC:
321 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
322 case vmIntrinsics::_inflateStringC:
323 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
324
325 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
326 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
327 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
328 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
329 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
330 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
331 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
332 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
333 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
334
335 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
336 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
337 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
338 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
339 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
340 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
341 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
342 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
343 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, 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::_scopedValueCache: return inline_native_scopedValueCache();
481 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
482
483 #if INCLUDE_JVMTI
484 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
485 "notifyJvmtiStart", true, false);
486 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
487 "notifyJvmtiEnd", false, true);
488 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
489 "notifyJvmtiMount", false, false);
490 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
491 "notifyJvmtiUnmount", false, false);
492 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_getLength: return inline_native_getLength();
508 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
509 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
510 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
511 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
512 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
513 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
514 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
515
516 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
517 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
518
519 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
520
521 case vmIntrinsics::_isInstance:
522 case vmIntrinsics::_getModifiers:
523 case vmIntrinsics::_isInterface:
524 case vmIntrinsics::_isArray:
525 case vmIntrinsics::_isPrimitive:
526 case vmIntrinsics::_isHidden:
527 case vmIntrinsics::_getSuperclass:
528 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
529
530 case vmIntrinsics::_floatToRawIntBits:
531 case vmIntrinsics::_floatToIntBits:
532 case vmIntrinsics::_intBitsToFloat:
533 case vmIntrinsics::_doubleToRawLongBits:
534 case vmIntrinsics::_doubleToLongBits:
535 case vmIntrinsics::_longBitsToDouble:
536 case vmIntrinsics::_floatToFloat16:
537 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
538
539 case vmIntrinsics::_floatIsFinite:
540 case vmIntrinsics::_floatIsInfinite:
541 case vmIntrinsics::_doubleIsFinite:
542 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
543
544 case vmIntrinsics::_numberOfLeadingZeros_i:
545 case vmIntrinsics::_numberOfLeadingZeros_l:
546 case vmIntrinsics::_numberOfTrailingZeros_i:
547 case vmIntrinsics::_numberOfTrailingZeros_l:
548 case vmIntrinsics::_bitCount_i:
549 case vmIntrinsics::_bitCount_l:
550 case vmIntrinsics::_reverse_i:
551 case vmIntrinsics::_reverse_l:
552 case vmIntrinsics::_reverseBytes_i:
553 case vmIntrinsics::_reverseBytes_l:
554 case vmIntrinsics::_reverseBytes_s:
555 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
556
557 case vmIntrinsics::_compress_i:
558 case vmIntrinsics::_compress_l:
559 case vmIntrinsics::_expand_i:
560 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
561
562 case vmIntrinsics::_compareUnsigned_i:
563 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
564
565 case vmIntrinsics::_divideUnsigned_i:
566 case vmIntrinsics::_divideUnsigned_l:
567 case vmIntrinsics::_remainderUnsigned_i:
568 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
569
570 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
571
572 case vmIntrinsics::_Reference_get: return inline_reference_get();
573 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
574 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
575
576 case vmIntrinsics::_Class_cast: return inline_Class_cast();
577
578 case vmIntrinsics::_aescrypt_encryptBlock:
579 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
580
581 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
582 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
583 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
584
585 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
586 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
587 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
588
589 case vmIntrinsics::_counterMode_AESCrypt:
590 return inline_counterMode_AESCrypt(intrinsic_id());
591
592 case vmIntrinsics::_galoisCounterMode_AESCrypt:
593 return inline_galoisCounterMode_AESCrypt();
594
595 case vmIntrinsics::_md5_implCompress:
596 case vmIntrinsics::_sha_implCompress:
597 case vmIntrinsics::_sha2_implCompress:
598 case vmIntrinsics::_sha5_implCompress:
599 case vmIntrinsics::_sha3_implCompress:
600 return inline_digestBase_implCompress(intrinsic_id());
601
602 case vmIntrinsics::_digestBase_implCompressMB:
603 return inline_digestBase_implCompressMB(predicate);
604
605 case vmIntrinsics::_multiplyToLen:
606 return inline_multiplyToLen();
607
608 case vmIntrinsics::_squareToLen:
609 return inline_squareToLen();
610
611 case vmIntrinsics::_mulAdd:
612 return inline_mulAdd();
613
614 case vmIntrinsics::_montgomeryMultiply:
615 return inline_montgomeryMultiply();
616 case vmIntrinsics::_montgomerySquare:
617 return inline_montgomerySquare();
618
619 case vmIntrinsics::_bigIntegerRightShiftWorker:
620 return inline_bigIntegerShift(true);
621 case vmIntrinsics::_bigIntegerLeftShiftWorker:
622 return inline_bigIntegerShift(false);
623
624 case vmIntrinsics::_vectorizedMismatch:
625 return inline_vectorizedMismatch();
626
627 case vmIntrinsics::_ghash_processBlocks:
628 return inline_ghash_processBlocks();
629 case vmIntrinsics::_chacha20Block:
630 return inline_chacha20Block();
631 case vmIntrinsics::_base64_encodeBlock:
632 return inline_base64_encodeBlock();
633 case vmIntrinsics::_base64_decodeBlock:
634 return inline_base64_decodeBlock();
635 case vmIntrinsics::_poly1305_processBlocks:
636 return inline_poly1305_processBlocks();
637
638 case vmIntrinsics::_encodeISOArray:
639 case vmIntrinsics::_encodeByteISOArray:
640 return inline_encodeISOArray(false);
641 case vmIntrinsics::_encodeAsciiArray:
642 return inline_encodeISOArray(true);
643
644 case vmIntrinsics::_updateCRC32:
645 return inline_updateCRC32();
646 case vmIntrinsics::_updateBytesCRC32:
647 return inline_updateBytesCRC32();
648 case vmIntrinsics::_updateByteBufferCRC32:
649 return inline_updateByteBufferCRC32();
650
651 case vmIntrinsics::_updateBytesCRC32C:
652 return inline_updateBytesCRC32C();
653 case vmIntrinsics::_updateDirectByteBufferCRC32C:
654 return inline_updateDirectByteBufferCRC32C();
655
656 case vmIntrinsics::_updateBytesAdler32:
657 return inline_updateBytesAdler32();
658 case vmIntrinsics::_updateByteBufferAdler32:
659 return inline_updateByteBufferAdler32();
660
661 case vmIntrinsics::_profileBoolean:
662 return inline_profileBoolean();
663 case vmIntrinsics::_isCompileConstant:
664 return inline_isCompileConstant();
665
666 case vmIntrinsics::_countPositives:
667 return inline_countPositives();
668
669 case vmIntrinsics::_fmaD:
670 case vmIntrinsics::_fmaF:
671 return inline_fma(intrinsic_id());
672
673 case vmIntrinsics::_isDigit:
674 case vmIntrinsics::_isLowerCase:
675 case vmIntrinsics::_isUpperCase:
676 case vmIntrinsics::_isWhitespace:
677 return inline_character_compare(intrinsic_id());
678
679 case vmIntrinsics::_min:
680 case vmIntrinsics::_max:
681 case vmIntrinsics::_min_strict:
682 case vmIntrinsics::_max_strict:
683 return inline_min_max(intrinsic_id());
684
685 case vmIntrinsics::_maxF:
686 case vmIntrinsics::_minF:
687 case vmIntrinsics::_maxD:
688 case vmIntrinsics::_minD:
689 case vmIntrinsics::_maxF_strict:
690 case vmIntrinsics::_minF_strict:
691 case vmIntrinsics::_maxD_strict:
692 case vmIntrinsics::_minD_strict:
693 return inline_fp_min_max(intrinsic_id());
694
695 case vmIntrinsics::_VectorUnaryOp:
696 return inline_vector_nary_operation(1);
697 case vmIntrinsics::_VectorBinaryOp:
698 return inline_vector_nary_operation(2);
699 case vmIntrinsics::_VectorTernaryOp:
700 return inline_vector_nary_operation(3);
701 case vmIntrinsics::_VectorFromBitsCoerced:
702 return inline_vector_frombits_coerced();
703 case vmIntrinsics::_VectorShuffleIota:
704 return inline_vector_shuffle_iota();
705 case vmIntrinsics::_VectorMaskOp:
706 return inline_vector_mask_operation();
707 case vmIntrinsics::_VectorShuffleToVector:
708 return inline_vector_shuffle_to_vector();
709 case vmIntrinsics::_VectorLoadOp:
710 return inline_vector_mem_operation(/*is_store=*/false);
711 case vmIntrinsics::_VectorLoadMaskedOp:
712 return inline_vector_mem_masked_operation(/*is_store*/false);
713 case vmIntrinsics::_VectorStoreOp:
714 return inline_vector_mem_operation(/*is_store=*/true);
715 case vmIntrinsics::_VectorStoreMaskedOp:
716 return inline_vector_mem_masked_operation(/*is_store=*/true);
717 case vmIntrinsics::_VectorGatherOp:
718 return inline_vector_gather_scatter(/*is_scatter*/ false);
719 case vmIntrinsics::_VectorScatterOp:
720 return inline_vector_gather_scatter(/*is_scatter*/ true);
721 case vmIntrinsics::_VectorReductionCoerced:
722 return inline_vector_reduction();
723 case vmIntrinsics::_VectorTest:
724 return inline_vector_test();
725 case vmIntrinsics::_VectorBlend:
726 return inline_vector_blend();
727 case vmIntrinsics::_VectorRearrange:
728 return inline_vector_rearrange();
729 case vmIntrinsics::_VectorCompare:
730 return inline_vector_compare();
731 case vmIntrinsics::_VectorBroadcastInt:
732 return inline_vector_broadcast_int();
733 case vmIntrinsics::_VectorConvert:
734 return inline_vector_convert();
735 case vmIntrinsics::_VectorInsert:
736 return inline_vector_insert();
737 case vmIntrinsics::_VectorExtract:
738 return inline_vector_extract();
739 case vmIntrinsics::_VectorCompressExpand:
740 return inline_vector_compress_expand();
741 case vmIntrinsics::_IndexVector:
742 return inline_index_vector();
743 case vmIntrinsics::_IndexPartiallyInUpperRange:
744 return inline_index_partially_in_upper_range();
745
746 case vmIntrinsics::_getObjectSize:
747 return inline_getObjectSize();
748
749 case vmIntrinsics::_blackhole:
750 return inline_blackhole();
751
752 default:
753 // If you get here, it may be that someone has added a new intrinsic
754 // to the list in vmIntrinsics.hpp without implementing it here.
755 #ifndef PRODUCT
756 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
757 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
758 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
759 }
760 #endif
761 return false;
762 }
763 }
764
765 Node* LibraryCallKit::try_to_predicate(int predicate) {
766 if (!jvms()->has_method()) {
767 // Root JVMState has a null method.
768 assert(map()->memory()->Opcode() == Op_Parm, "");
769 // Insert the memory aliasing node
770 set_all_memory(reset_memory());
771 }
772 assert(merged_memory(), "");
773
774 switch (intrinsic_id()) {
775 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
776 return inline_cipherBlockChaining_AESCrypt_predicate(false);
777 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
778 return inline_cipherBlockChaining_AESCrypt_predicate(true);
779 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
780 return inline_electronicCodeBook_AESCrypt_predicate(false);
781 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
782 return inline_electronicCodeBook_AESCrypt_predicate(true);
783 case vmIntrinsics::_counterMode_AESCrypt:
784 return inline_counterMode_AESCrypt_predicate();
785 case vmIntrinsics::_digestBase_implCompressMB:
786 return inline_digestBase_implCompressMB_predicate(predicate);
787 case vmIntrinsics::_galoisCounterMode_AESCrypt:
788 return inline_galoisCounterMode_AESCrypt_predicate();
789
790 default:
791 // If you get here, it may be that someone has added a new intrinsic
792 // to the list in vmIntrinsics.hpp without implementing it here.
793 #ifndef PRODUCT
794 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
795 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
796 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
797 }
798 #endif
799 Node* slow_ctl = control();
800 set_control(top()); // No fast path intrinsic
801 return slow_ctl;
802 }
803 }
804
805 //------------------------------set_result-------------------------------
806 // Helper function for finishing intrinsics.
807 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
808 record_for_igvn(region);
809 set_control(_gvn.transform(region));
810 set_result( _gvn.transform(value));
811 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
812 }
813
814 //------------------------------generate_guard---------------------------
815 // Helper function for generating guarded fast-slow graph structures.
816 // The given 'test', if true, guards a slow path. If the test fails
817 // then a fast path can be taken. (We generally hope it fails.)
818 // In all cases, GraphKit::control() is updated to the fast path.
819 // The returned value represents the control for the slow path.
820 // The return value is never 'top'; it is either a valid control
821 // or null if it is obvious that the slow path can never be taken.
822 // Also, if region and the slow control are not null, the slow edge
823 // is appended to the region.
824 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
825 if (stopped()) {
826 // Already short circuited.
827 return nullptr;
828 }
829
830 // Build an if node and its projections.
831 // If test is true we take the slow path, which we assume is uncommon.
832 if (_gvn.type(test) == TypeInt::ZERO) {
833 // The slow branch is never taken. No need to build this guard.
834 return nullptr;
835 }
836
837 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
838
839 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
840 if (if_slow == top()) {
841 // The slow branch is never taken. No need to build this guard.
842 return nullptr;
843 }
844
845 if (region != nullptr)
846 region->add_req(if_slow);
847
848 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
849 set_control(if_fast);
850
851 return if_slow;
852 }
853
854 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
855 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
856 }
857 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
858 return generate_guard(test, region, PROB_FAIR);
859 }
860
861 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
862 Node* *pos_index) {
863 if (stopped())
864 return nullptr; // already stopped
865 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
866 return nullptr; // index is already adequately typed
867 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
868 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
869 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
870 if (is_neg != nullptr && pos_index != nullptr) {
871 // Emulate effect of Parse::adjust_map_after_if.
872 Node* ccast = new CastIINode(index, TypeInt::POS);
873 ccast->set_req(0, control());
874 (*pos_index) = _gvn.transform(ccast);
875 }
876 return is_neg;
877 }
878
879 // Make sure that 'position' is a valid limit index, in [0..length].
880 // There are two equivalent plans for checking this:
881 // A. (offset + copyLength) unsigned<= arrayLength
882 // B. offset <= (arrayLength - copyLength)
883 // We require that all of the values above, except for the sum and
884 // difference, are already known to be non-negative.
885 // Plan A is robust in the face of overflow, if offset and copyLength
886 // are both hugely positive.
887 //
888 // Plan B is less direct and intuitive, but it does not overflow at
889 // all, since the difference of two non-negatives is always
890 // representable. Whenever Java methods must perform the equivalent
891 // check they generally use Plan B instead of Plan A.
892 // For the moment we use Plan A.
893 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
894 Node* subseq_length,
895 Node* array_length,
896 RegionNode* region) {
897 if (stopped())
898 return nullptr; // already stopped
899 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
900 if (zero_offset && subseq_length->eqv_uncast(array_length))
901 return nullptr; // common case of whole-array copy
902 Node* last = subseq_length;
903 if (!zero_offset) // last += offset
904 last = _gvn.transform(new AddINode(last, offset));
905 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
906 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
907 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
908 return is_over;
909 }
910
911 // Emit range checks for the given String.value byte array
912 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
913 if (stopped()) {
914 return; // already stopped
915 }
916 RegionNode* bailout = new RegionNode(1);
917 record_for_igvn(bailout);
918 if (char_count) {
919 // Convert char count to byte count
920 count = _gvn.transform(new LShiftINode(count, intcon(1)));
921 }
922
923 // Offset and count must not be negative
924 generate_negative_guard(offset, bailout);
925 generate_negative_guard(count, bailout);
926 // Offset + count must not exceed length of array
927 generate_limit_guard(offset, count, load_array_length(array), bailout);
928
929 if (bailout->req() > 1) {
930 PreserveJVMState pjvms(this);
931 set_control(_gvn.transform(bailout));
932 uncommon_trap(Deoptimization::Reason_intrinsic,
933 Deoptimization::Action_maybe_recompile);
934 }
935 }
936
937 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
938 bool is_immutable) {
939 ciKlass* thread_klass = env()->Thread_klass();
940 const Type* thread_type
941 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
942
943 Node* thread = _gvn.transform(new ThreadLocalNode());
944 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
945 tls_output = thread;
946
947 Node* thread_obj_handle
948 = (is_immutable
949 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
950 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
951 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
952 thread_obj_handle = _gvn.transform(thread_obj_handle);
953
954 DecoratorSet decorators = IN_NATIVE;
955 if (is_immutable) {
956 decorators |= C2_IMMUTABLE_MEMORY;
957 }
958 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
959 }
960
961 //--------------------------generate_current_thread--------------------
962 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
963 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
964 /*is_immutable*/false);
965 }
966
967 //--------------------------generate_virtual_thread--------------------
968 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
969 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
970 !C->method()->changes_current_thread());
971 }
972
973 //------------------------------make_string_method_node------------------------
974 // Helper method for String intrinsic functions. This version is called with
975 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
976 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
977 // containing the lengths of str1 and str2.
978 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
979 Node* result = nullptr;
980 switch (opcode) {
981 case Op_StrIndexOf:
982 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
983 str1_start, cnt1, str2_start, cnt2, ae);
984 break;
985 case Op_StrComp:
986 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
987 str1_start, cnt1, str2_start, cnt2, ae);
988 break;
989 case Op_StrEquals:
990 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
991 // Use the constant length if there is one because optimized match rule may exist.
992 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
993 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
994 break;
995 default:
996 ShouldNotReachHere();
997 return nullptr;
998 }
999
1000 // All these intrinsics have checks.
1001 C->set_has_split_ifs(true); // Has chance for split-if optimization
1002 clear_upper_avx();
1003
1004 return _gvn.transform(result);
1005 }
1006
1007 //------------------------------inline_string_compareTo------------------------
1008 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1009 Node* arg1 = argument(0);
1010 Node* arg2 = argument(1);
1011
1012 arg1 = must_be_not_null(arg1, true);
1013 arg2 = must_be_not_null(arg2, true);
1014
1015 // Get start addr and length of first argument
1016 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1017 Node* arg1_cnt = load_array_length(arg1);
1018
1019 // Get start addr and length of second argument
1020 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1021 Node* arg2_cnt = load_array_length(arg2);
1022
1023 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1024 set_result(result);
1025 return true;
1026 }
1027
1028 //------------------------------inline_string_equals------------------------
1029 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1030 Node* arg1 = argument(0);
1031 Node* arg2 = argument(1);
1032
1033 // paths (plus control) merge
1034 RegionNode* region = new RegionNode(3);
1035 Node* phi = new PhiNode(region, TypeInt::BOOL);
1036
1037 if (!stopped()) {
1038
1039 arg1 = must_be_not_null(arg1, true);
1040 arg2 = must_be_not_null(arg2, true);
1041
1042 // Get start addr and length of first argument
1043 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1044 Node* arg1_cnt = load_array_length(arg1);
1045
1046 // Get start addr and length of second argument
1047 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1048 Node* arg2_cnt = load_array_length(arg2);
1049
1050 // Check for arg1_cnt != arg2_cnt
1051 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1052 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1053 Node* if_ne = generate_slow_guard(bol, nullptr);
1054 if (if_ne != nullptr) {
1055 phi->init_req(2, intcon(0));
1056 region->init_req(2, if_ne);
1057 }
1058
1059 // Check for count == 0 is done by assembler code for StrEquals.
1060
1061 if (!stopped()) {
1062 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1063 phi->init_req(1, equals);
1064 region->init_req(1, control());
1065 }
1066 }
1067
1068 // post merge
1069 set_control(_gvn.transform(region));
1070 record_for_igvn(region);
1071
1072 set_result(_gvn.transform(phi));
1073 return true;
1074 }
1075
1076 //------------------------------inline_array_equals----------------------------
1077 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1078 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1079 Node* arg1 = argument(0);
1080 Node* arg2 = argument(1);
1081
1082 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1083 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1084 clear_upper_avx();
1085
1086 return true;
1087 }
1088
1089
1090 //------------------------------inline_countPositives------------------------------
1091 bool LibraryCallKit::inline_countPositives() {
1092 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1093 return false;
1094 }
1095
1096 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1097 // no receiver since it is static method
1098 Node* ba = argument(0);
1099 Node* offset = argument(1);
1100 Node* len = argument(2);
1101
1102 ba = must_be_not_null(ba, true);
1103
1104 // Range checks
1105 generate_string_range_check(ba, offset, len, false);
1106 if (stopped()) {
1107 return true;
1108 }
1109 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1110 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1111 set_result(_gvn.transform(result));
1112 return true;
1113 }
1114
1115 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1116 Node* index = argument(0);
1117 Node* length = bt == T_INT ? argument(1) : argument(2);
1118 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1119 return false;
1120 }
1121
1122 // check that length is positive
1123 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1124 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1125
1126 {
1127 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1128 uncommon_trap(Deoptimization::Reason_intrinsic,
1129 Deoptimization::Action_make_not_entrant);
1130 }
1131
1132 if (stopped()) {
1133 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1134 return true;
1135 }
1136
1137 // length is now known positive, add a cast node to make this explicit
1138 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1139 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1140 casted_length = _gvn.transform(casted_length);
1141 replace_in_map(length, casted_length);
1142 length = casted_length;
1143
1144 // Use an unsigned comparison for the range check itself
1145 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1146 BoolTest::mask btest = BoolTest::lt;
1147 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1148 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1149 _gvn.set_type(rc, rc->Value(&_gvn));
1150 if (!rc_bool->is_Con()) {
1151 record_for_igvn(rc);
1152 }
1153 set_control(_gvn.transform(new IfTrueNode(rc)));
1154 {
1155 PreserveJVMState pjvms(this);
1156 set_control(_gvn.transform(new IfFalseNode(rc)));
1157 uncommon_trap(Deoptimization::Reason_range_check,
1158 Deoptimization::Action_make_not_entrant);
1159 }
1160
1161 if (stopped()) {
1162 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1163 return true;
1164 }
1165
1166 // index is now known to be >= 0 and < length, cast it
1167 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1168 result = _gvn.transform(result);
1169 set_result(result);
1170 replace_in_map(index, result);
1171 return true;
1172 }
1173
1174 //------------------------------inline_string_indexOf------------------------
1175 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1176 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1177 return false;
1178 }
1179 Node* src = argument(0);
1180 Node* tgt = argument(1);
1181
1182 // Make the merge point
1183 RegionNode* result_rgn = new RegionNode(4);
1184 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1185
1186 src = must_be_not_null(src, true);
1187 tgt = must_be_not_null(tgt, true);
1188
1189 // Get start addr and length of source string
1190 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1191 Node* src_count = load_array_length(src);
1192
1193 // Get start addr and length of substring
1194 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1195 Node* tgt_count = load_array_length(tgt);
1196
1197 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1198 // Divide src size by 2 if String is UTF16 encoded
1199 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1200 }
1201 if (ae == StrIntrinsicNode::UU) {
1202 // Divide substring size by 2 if String is UTF16 encoded
1203 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1204 }
1205
1206 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1207 if (result != nullptr) {
1208 result_phi->init_req(3, result);
1209 result_rgn->init_req(3, control());
1210 }
1211 set_control(_gvn.transform(result_rgn));
1212 record_for_igvn(result_rgn);
1213 set_result(_gvn.transform(result_phi));
1214
1215 return true;
1216 }
1217
1218 //-----------------------------inline_string_indexOf-----------------------
1219 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1220 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1221 return false;
1222 }
1223 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1224 return false;
1225 }
1226 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1227 Node* src = argument(0); // byte[]
1228 Node* src_count = argument(1); // char count
1229 Node* tgt = argument(2); // byte[]
1230 Node* tgt_count = argument(3); // char count
1231 Node* from_index = argument(4); // char index
1232
1233 src = must_be_not_null(src, true);
1234 tgt = must_be_not_null(tgt, true);
1235
1236 // Multiply byte array index by 2 if String is UTF16 encoded
1237 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1238 src_count = _gvn.transform(new SubINode(src_count, from_index));
1239 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1240 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1241
1242 // Range checks
1243 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1244 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1245 if (stopped()) {
1246 return true;
1247 }
1248
1249 RegionNode* region = new RegionNode(5);
1250 Node* phi = new PhiNode(region, TypeInt::INT);
1251
1252 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1253 if (result != nullptr) {
1254 // The result is index relative to from_index if substring was found, -1 otherwise.
1255 // Generate code which will fold into cmove.
1256 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1257 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1258
1259 Node* if_lt = generate_slow_guard(bol, nullptr);
1260 if (if_lt != nullptr) {
1261 // result == -1
1262 phi->init_req(3, result);
1263 region->init_req(3, if_lt);
1264 }
1265 if (!stopped()) {
1266 result = _gvn.transform(new AddINode(result, from_index));
1267 phi->init_req(4, result);
1268 region->init_req(4, control());
1269 }
1270 }
1271
1272 set_control(_gvn.transform(region));
1273 record_for_igvn(region);
1274 set_result(_gvn.transform(phi));
1275 clear_upper_avx();
1276
1277 return true;
1278 }
1279
1280 // Create StrIndexOfNode with fast path checks
1281 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1282 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1283 // Check for substr count > string count
1284 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1285 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1286 Node* if_gt = generate_slow_guard(bol, nullptr);
1287 if (if_gt != nullptr) {
1288 phi->init_req(1, intcon(-1));
1289 region->init_req(1, if_gt);
1290 }
1291 if (!stopped()) {
1292 // Check for substr count == 0
1293 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1294 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1295 Node* if_zero = generate_slow_guard(bol, nullptr);
1296 if (if_zero != nullptr) {
1297 phi->init_req(2, intcon(0));
1298 region->init_req(2, if_zero);
1299 }
1300 }
1301 if (!stopped()) {
1302 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1303 }
1304 return nullptr;
1305 }
1306
1307 //-----------------------------inline_string_indexOfChar-----------------------
1308 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1309 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1310 return false;
1311 }
1312 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1313 return false;
1314 }
1315 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1316 Node* src = argument(0); // byte[]
1317 Node* int_ch = argument(1);
1318 Node* from_index = argument(2);
1319 Node* max = argument(3);
1320
1321 src = must_be_not_null(src, true);
1322
1323 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1324 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1325 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1326
1327 // Range checks
1328 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1329
1330 // Check for int_ch >= 0
1331 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1332 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1333 {
1334 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1335 uncommon_trap(Deoptimization::Reason_intrinsic,
1336 Deoptimization::Action_maybe_recompile);
1337 }
1338 if (stopped()) {
1339 return true;
1340 }
1341
1342 RegionNode* region = new RegionNode(3);
1343 Node* phi = new PhiNode(region, TypeInt::INT);
1344
1345 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1346 C->set_has_split_ifs(true); // Has chance for split-if optimization
1347 _gvn.transform(result);
1348
1349 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1350 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1351
1352 Node* if_lt = generate_slow_guard(bol, nullptr);
1353 if (if_lt != nullptr) {
1354 // result == -1
1355 phi->init_req(2, result);
1356 region->init_req(2, if_lt);
1357 }
1358 if (!stopped()) {
1359 result = _gvn.transform(new AddINode(result, from_index));
1360 phi->init_req(1, result);
1361 region->init_req(1, control());
1362 }
1363 set_control(_gvn.transform(region));
1364 record_for_igvn(region);
1365 set_result(_gvn.transform(phi));
1366
1367 return true;
1368 }
1369 //---------------------------inline_string_copy---------------------
1370 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1371 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1372 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1373 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1374 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1375 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1376 bool LibraryCallKit::inline_string_copy(bool compress) {
1377 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1378 return false;
1379 }
1380 int nargs = 5; // 2 oops, 3 ints
1381 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1382
1383 Node* src = argument(0);
1384 Node* src_offset = argument(1);
1385 Node* dst = argument(2);
1386 Node* dst_offset = argument(3);
1387 Node* length = argument(4);
1388
1389 // Check for allocation before we add nodes that would confuse
1390 // tightly_coupled_allocation()
1391 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1392
1393 // Figure out the size and type of the elements we will be copying.
1394 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1395 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1396 if (src_type == nullptr || dst_type == nullptr) {
1397 return false;
1398 }
1399 BasicType src_elem = src_type->elem()->array_element_basic_type();
1400 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1401 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1402 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1403 "Unsupported array types for inline_string_copy");
1404
1405 src = must_be_not_null(src, true);
1406 dst = must_be_not_null(dst, true);
1407
1408 // Convert char[] offsets to byte[] offsets
1409 bool convert_src = (compress && src_elem == T_BYTE);
1410 bool convert_dst = (!compress && dst_elem == T_BYTE);
1411 if (convert_src) {
1412 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1413 } else if (convert_dst) {
1414 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1415 }
1416
1417 // Range checks
1418 generate_string_range_check(src, src_offset, length, convert_src);
1419 generate_string_range_check(dst, dst_offset, length, convert_dst);
1420 if (stopped()) {
1421 return true;
1422 }
1423
1424 Node* src_start = array_element_address(src, src_offset, src_elem);
1425 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1426 // 'src_start' points to src array + scaled offset
1427 // 'dst_start' points to dst array + scaled offset
1428 Node* count = nullptr;
1429 if (compress) {
1430 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1431 } else {
1432 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1433 }
1434
1435 if (alloc != nullptr) {
1436 if (alloc->maybe_set_complete(&_gvn)) {
1437 // "You break it, you buy it."
1438 InitializeNode* init = alloc->initialization();
1439 assert(init->is_complete(), "we just did this");
1440 init->set_complete_with_arraycopy();
1441 assert(dst->is_CheckCastPP(), "sanity");
1442 assert(dst->in(0)->in(0) == init, "dest pinned");
1443 }
1444 // Do not let stores that initialize this object be reordered with
1445 // a subsequent store that would make this object accessible by
1446 // other threads.
1447 // Record what AllocateNode this StoreStore protects so that
1448 // escape analysis can go from the MemBarStoreStoreNode to the
1449 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1450 // based on the escape status of the AllocateNode.
1451 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1452 }
1453 if (compress) {
1454 set_result(_gvn.transform(count));
1455 }
1456 clear_upper_avx();
1457
1458 return true;
1459 }
1460
1461 #ifdef _LP64
1462 #define XTOP ,top() /*additional argument*/
1463 #else //_LP64
1464 #define XTOP /*no additional argument*/
1465 #endif //_LP64
1466
1467 //------------------------inline_string_toBytesU--------------------------
1468 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1469 bool LibraryCallKit::inline_string_toBytesU() {
1470 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1471 return false;
1472 }
1473 // Get the arguments.
1474 Node* value = argument(0);
1475 Node* offset = argument(1);
1476 Node* length = argument(2);
1477
1478 Node* newcopy = nullptr;
1479
1480 // Set the original stack and the reexecute bit for the interpreter to reexecute
1481 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1482 { PreserveReexecuteState preexecs(this);
1483 jvms()->set_should_reexecute(true);
1484
1485 // Check if a null path was taken unconditionally.
1486 value = null_check(value);
1487
1488 RegionNode* bailout = new RegionNode(1);
1489 record_for_igvn(bailout);
1490
1491 // Range checks
1492 generate_negative_guard(offset, bailout);
1493 generate_negative_guard(length, bailout);
1494 generate_limit_guard(offset, length, load_array_length(value), bailout);
1495 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1496 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1497
1498 if (bailout->req() > 1) {
1499 PreserveJVMState pjvms(this);
1500 set_control(_gvn.transform(bailout));
1501 uncommon_trap(Deoptimization::Reason_intrinsic,
1502 Deoptimization::Action_maybe_recompile);
1503 }
1504 if (stopped()) {
1505 return true;
1506 }
1507
1508 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1509 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1510 newcopy = new_array(klass_node, size, 0); // no arguments to push
1511 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1512 guarantee(alloc != nullptr, "created above");
1513
1514 // Calculate starting addresses.
1515 Node* src_start = array_element_address(value, offset, T_CHAR);
1516 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1517
1518 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1519 const TypeInt* toffset = gvn().type(offset)->is_int();
1520 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1521
1522 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1523 const char* copyfunc_name = "arraycopy";
1524 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1525 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1526 OptoRuntime::fast_arraycopy_Type(),
1527 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1528 src_start, dst_start, ConvI2X(length) XTOP);
1529 // Do not let reads from the cloned object float above the arraycopy.
1530 if (alloc->maybe_set_complete(&_gvn)) {
1531 // "You break it, you buy it."
1532 InitializeNode* init = alloc->initialization();
1533 assert(init->is_complete(), "we just did this");
1534 init->set_complete_with_arraycopy();
1535 assert(newcopy->is_CheckCastPP(), "sanity");
1536 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1537 }
1538 // Do not let stores that initialize this object be reordered with
1539 // a subsequent store that would make this object accessible by
1540 // other threads.
1541 // Record what AllocateNode this StoreStore protects so that
1542 // escape analysis can go from the MemBarStoreStoreNode to the
1543 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1544 // based on the escape status of the AllocateNode.
1545 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1546 } // original reexecute is set back here
1547
1548 C->set_has_split_ifs(true); // Has chance for split-if optimization
1549 if (!stopped()) {
1550 set_result(newcopy);
1551 }
1552 clear_upper_avx();
1553
1554 return true;
1555 }
1556
1557 //------------------------inline_string_getCharsU--------------------------
1558 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1559 bool LibraryCallKit::inline_string_getCharsU() {
1560 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1561 return false;
1562 }
1563
1564 // Get the arguments.
1565 Node* src = argument(0);
1566 Node* src_begin = argument(1);
1567 Node* src_end = argument(2); // exclusive offset (i < src_end)
1568 Node* dst = argument(3);
1569 Node* dst_begin = argument(4);
1570
1571 // Check for allocation before we add nodes that would confuse
1572 // tightly_coupled_allocation()
1573 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1574
1575 // Check if a null path was taken unconditionally.
1576 src = null_check(src);
1577 dst = null_check(dst);
1578 if (stopped()) {
1579 return true;
1580 }
1581
1582 // Get length and convert char[] offset to byte[] offset
1583 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1584 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1585
1586 // Range checks
1587 generate_string_range_check(src, src_begin, length, true);
1588 generate_string_range_check(dst, dst_begin, length, false);
1589 if (stopped()) {
1590 return true;
1591 }
1592
1593 if (!stopped()) {
1594 // Calculate starting addresses.
1595 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1596 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1597
1598 // Check if array addresses are aligned to HeapWordSize
1599 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1600 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1601 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1602 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1603
1604 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1605 const char* copyfunc_name = "arraycopy";
1606 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1607 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1608 OptoRuntime::fast_arraycopy_Type(),
1609 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1610 src_start, dst_start, ConvI2X(length) XTOP);
1611 // Do not let reads from the cloned object float above the arraycopy.
1612 if (alloc != nullptr) {
1613 if (alloc->maybe_set_complete(&_gvn)) {
1614 // "You break it, you buy it."
1615 InitializeNode* init = alloc->initialization();
1616 assert(init->is_complete(), "we just did this");
1617 init->set_complete_with_arraycopy();
1618 assert(dst->is_CheckCastPP(), "sanity");
1619 assert(dst->in(0)->in(0) == init, "dest pinned");
1620 }
1621 // Do not let stores that initialize this object be reordered with
1622 // a subsequent store that would make this object accessible by
1623 // other threads.
1624 // Record what AllocateNode this StoreStore protects so that
1625 // escape analysis can go from the MemBarStoreStoreNode to the
1626 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1627 // based on the escape status of the AllocateNode.
1628 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1629 } else {
1630 insert_mem_bar(Op_MemBarCPUOrder);
1631 }
1632 }
1633
1634 C->set_has_split_ifs(true); // Has chance for split-if optimization
1635 return true;
1636 }
1637
1638 //----------------------inline_string_char_access----------------------------
1639 // Store/Load char to/from byte[] array.
1640 // static void StringUTF16.putChar(byte[] val, int index, int c)
1641 // static char StringUTF16.getChar(byte[] val, int index)
1642 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1643 Node* value = argument(0);
1644 Node* index = argument(1);
1645 Node* ch = is_store ? argument(2) : nullptr;
1646
1647 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1648 // correctly requires matched array shapes.
1649 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1650 "sanity: byte[] and char[] bases agree");
1651 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1652 "sanity: byte[] and char[] scales agree");
1653
1654 // Bail when getChar over constants is requested: constant folding would
1655 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1656 // Java method would constant fold nicely instead.
1657 if (!is_store && value->is_Con() && index->is_Con()) {
1658 return false;
1659 }
1660
1661 // Save state and restore on bailout
1662 uint old_sp = sp();
1663 SafePointNode* old_map = clone_map();
1664
1665 value = must_be_not_null(value, true);
1666
1667 Node* adr = array_element_address(value, index, T_CHAR);
1668 if (adr->is_top()) {
1669 set_map(old_map);
1670 set_sp(old_sp);
1671 return false;
1672 }
1673 destruct_map_clone(old_map);
1674 if (is_store) {
1675 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1676 } else {
1677 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);
1678 set_result(ch);
1679 }
1680 return true;
1681 }
1682
1683 //--------------------------round_double_node--------------------------------
1684 // Round a double node if necessary.
1685 Node* LibraryCallKit::round_double_node(Node* n) {
1686 if (Matcher::strict_fp_requires_explicit_rounding) {
1687 #ifdef IA32
1688 if (UseSSE < 2) {
1689 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1690 }
1691 #else
1692 Unimplemented();
1693 #endif // IA32
1694 }
1695 return n;
1696 }
1697
1698 //------------------------------inline_math-----------------------------------
1699 // public static double Math.abs(double)
1700 // public static double Math.sqrt(double)
1701 // public static double Math.log(double)
1702 // public static double Math.log10(double)
1703 // public static double Math.round(double)
1704 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1705 Node* arg = round_double_node(argument(0));
1706 Node* n = nullptr;
1707 switch (id) {
1708 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1709 case vmIntrinsics::_dsqrt:
1710 case vmIntrinsics::_dsqrt_strict:
1711 n = new SqrtDNode(C, control(), arg); break;
1712 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1713 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1714 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1715 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1716 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1717 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1718 default: fatal_unexpected_iid(id); break;
1719 }
1720 set_result(_gvn.transform(n));
1721 return true;
1722 }
1723
1724 //------------------------------inline_math-----------------------------------
1725 // public static float Math.abs(float)
1726 // public static int Math.abs(int)
1727 // public static long Math.abs(long)
1728 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1729 Node* arg = argument(0);
1730 Node* n = nullptr;
1731 switch (id) {
1732 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1733 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1734 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1735 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1736 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1737 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1738 default: fatal_unexpected_iid(id); break;
1739 }
1740 set_result(_gvn.transform(n));
1741 return true;
1742 }
1743
1744 //------------------------------runtime_math-----------------------------
1745 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1746 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1747 "must be (DD)D or (D)D type");
1748
1749 // Inputs
1750 Node* a = round_double_node(argument(0));
1751 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1752
1753 const TypePtr* no_memory_effects = nullptr;
1754 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1755 no_memory_effects,
1756 a, top(), b, b ? top() : nullptr);
1757 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1758 #ifdef ASSERT
1759 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1760 assert(value_top == top(), "second value must be top");
1761 #endif
1762
1763 set_result(value);
1764 return true;
1765 }
1766
1767 //------------------------------inline_math_pow-----------------------------
1768 bool LibraryCallKit::inline_math_pow() {
1769 Node* exp = round_double_node(argument(2));
1770 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1771 if (d != nullptr) {
1772 if (d->getd() == 2.0) {
1773 // Special case: pow(x, 2.0) => x * x
1774 Node* base = round_double_node(argument(0));
1775 set_result(_gvn.transform(new MulDNode(base, base)));
1776 return true;
1777 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1778 // Special case: pow(x, 0.5) => sqrt(x)
1779 Node* base = round_double_node(argument(0));
1780 Node* zero = _gvn.zerocon(T_DOUBLE);
1781
1782 RegionNode* region = new RegionNode(3);
1783 Node* phi = new PhiNode(region, Type::DOUBLE);
1784
1785 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1786 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1787 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1788 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1789 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1790
1791 Node* if_pow = generate_slow_guard(test, nullptr);
1792 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1793 phi->init_req(1, value_sqrt);
1794 region->init_req(1, control());
1795
1796 if (if_pow != nullptr) {
1797 set_control(if_pow);
1798 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1799 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1800 const TypePtr* no_memory_effects = nullptr;
1801 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1802 no_memory_effects, base, top(), exp, top());
1803 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1804 #ifdef ASSERT
1805 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1806 assert(value_top == top(), "second value must be top");
1807 #endif
1808 phi->init_req(2, value_pow);
1809 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1810 }
1811
1812 C->set_has_split_ifs(true); // Has chance for split-if optimization
1813 set_control(_gvn.transform(region));
1814 record_for_igvn(region);
1815 set_result(_gvn.transform(phi));
1816
1817 return true;
1818 }
1819 }
1820
1821 return StubRoutines::dpow() != nullptr ?
1822 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1823 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1824 }
1825
1826 //------------------------------inline_math_native-----------------------------
1827 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1828 switch (id) {
1829 case vmIntrinsics::_dsin:
1830 return StubRoutines::dsin() != nullptr ?
1831 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1832 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1833 case vmIntrinsics::_dcos:
1834 return StubRoutines::dcos() != nullptr ?
1835 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1836 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1837 case vmIntrinsics::_dtan:
1838 return StubRoutines::dtan() != nullptr ?
1839 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1840 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1841 case vmIntrinsics::_dexp:
1842 return StubRoutines::dexp() != nullptr ?
1843 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1844 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1845 case vmIntrinsics::_dlog:
1846 return StubRoutines::dlog() != nullptr ?
1847 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1848 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1849 case vmIntrinsics::_dlog10:
1850 return StubRoutines::dlog10() != nullptr ?
1851 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1852 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1853
1854 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1855 case vmIntrinsics::_ceil:
1856 case vmIntrinsics::_floor:
1857 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1858
1859 case vmIntrinsics::_dsqrt:
1860 case vmIntrinsics::_dsqrt_strict:
1861 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1862 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1863 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1864 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1865 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1866
1867 case vmIntrinsics::_dpow: return inline_math_pow();
1868 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1869 case vmIntrinsics::_fcopySign: return inline_math(id);
1870 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1871 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1872 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1873
1874 // These intrinsics are not yet correctly implemented
1875 case vmIntrinsics::_datan2:
1876 return false;
1877
1878 default:
1879 fatal_unexpected_iid(id);
1880 return false;
1881 }
1882 }
1883
1884 //----------------------------inline_notify-----------------------------------*
1885 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1886 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1887 address func;
1888 if (id == vmIntrinsics::_notify) {
1889 func = OptoRuntime::monitor_notify_Java();
1890 } else {
1891 func = OptoRuntime::monitor_notifyAll_Java();
1892 }
1893 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1894 make_slow_call_ex(call, env()->Throwable_klass(), false);
1895 return true;
1896 }
1897
1898
1899 //----------------------------inline_min_max-----------------------------------
1900 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1901 set_result(generate_min_max(id, argument(0), argument(1)));
1902 return true;
1903 }
1904
1905 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1906 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1907 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1908 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1909 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1910
1911 {
1912 PreserveJVMState pjvms(this);
1913 PreserveReexecuteState preexecs(this);
1914 jvms()->set_should_reexecute(true);
1915
1916 set_control(slow_path);
1917 set_i_o(i_o());
1918
1919 uncommon_trap(Deoptimization::Reason_intrinsic,
1920 Deoptimization::Action_none);
1921 }
1922
1923 set_control(fast_path);
1924 set_result(math);
1925 }
1926
1927 template <typename OverflowOp>
1928 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1929 typedef typename OverflowOp::MathOp MathOp;
1930
1931 MathOp* mathOp = new MathOp(arg1, arg2);
1932 Node* operation = _gvn.transform( mathOp );
1933 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1934 inline_math_mathExact(operation, ofcheck);
1935 return true;
1936 }
1937
1938 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1939 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1940 }
1941
1942 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1943 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1944 }
1945
1946 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1947 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1948 }
1949
1950 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1951 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1952 }
1953
1954 bool LibraryCallKit::inline_math_negateExactI() {
1955 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1956 }
1957
1958 bool LibraryCallKit::inline_math_negateExactL() {
1959 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1960 }
1961
1962 bool LibraryCallKit::inline_math_multiplyExactI() {
1963 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1964 }
1965
1966 bool LibraryCallKit::inline_math_multiplyExactL() {
1967 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1968 }
1969
1970 bool LibraryCallKit::inline_math_multiplyHigh() {
1971 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1972 return true;
1973 }
1974
1975 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1976 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1977 return true;
1978 }
1979
1980 Node*
1981 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1982 Node* result_val = nullptr;
1983 switch (id) {
1984 case vmIntrinsics::_min:
1985 case vmIntrinsics::_min_strict:
1986 result_val = _gvn.transform(new MinINode(x0, y0));
1987 break;
1988 case vmIntrinsics::_max:
1989 case vmIntrinsics::_max_strict:
1990 result_val = _gvn.transform(new MaxINode(x0, y0));
1991 break;
1992 default:
1993 fatal_unexpected_iid(id);
1994 break;
1995 }
1996 return result_val;
1997 }
1998
1999 inline int
2000 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2001 const TypePtr* base_type = TypePtr::NULL_PTR;
2002 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2003 if (base_type == nullptr) {
2004 // Unknown type.
2005 return Type::AnyPtr;
2006 } else if (base_type == TypePtr::NULL_PTR) {
2007 // Since this is a null+long form, we have to switch to a rawptr.
2008 base = _gvn.transform(new CastX2PNode(offset));
2009 offset = MakeConX(0);
2010 return Type::RawPtr;
2011 } else if (base_type->base() == Type::RawPtr) {
2012 return Type::RawPtr;
2013 } else if (base_type->isa_oopptr()) {
2014 // Base is never null => always a heap address.
2015 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2016 return Type::OopPtr;
2017 }
2018 // Offset is small => always a heap address.
2019 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2020 if (offset_type != nullptr &&
2021 base_type->offset() == 0 && // (should always be?)
2022 offset_type->_lo >= 0 &&
2023 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2024 return Type::OopPtr;
2025 } else if (type == T_OBJECT) {
2026 // off heap access to an oop doesn't make any sense. Has to be on
2027 // heap.
2028 return Type::OopPtr;
2029 }
2030 // Otherwise, it might either be oop+off or null+addr.
2031 return Type::AnyPtr;
2032 } else {
2033 // No information:
2034 return Type::AnyPtr;
2035 }
2036 }
2037
2038 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2039 Node* uncasted_base = base;
2040 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2041 if (kind == Type::RawPtr) {
2042 return basic_plus_adr(top(), uncasted_base, offset);
2043 } else if (kind == Type::AnyPtr) {
2044 assert(base == uncasted_base, "unexpected base change");
2045 if (can_cast) {
2046 if (!_gvn.type(base)->speculative_maybe_null() &&
2047 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2048 // According to profiling, this access is always on
2049 // heap. Casting the base to not null and thus avoiding membars
2050 // around the access should allow better optimizations
2051 Node* null_ctl = top();
2052 base = null_check_oop(base, &null_ctl, true, true, true);
2053 assert(null_ctl->is_top(), "no null control here");
2054 return basic_plus_adr(base, offset);
2055 } else if (_gvn.type(base)->speculative_always_null() &&
2056 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2057 // According to profiling, this access is always off
2058 // heap.
2059 base = null_assert(base);
2060 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2061 offset = MakeConX(0);
2062 return basic_plus_adr(top(), raw_base, offset);
2063 }
2064 }
2065 // We don't know if it's an on heap or off heap access. Fall back
2066 // to raw memory access.
2067 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2068 return basic_plus_adr(top(), raw, offset);
2069 } else {
2070 assert(base == uncasted_base, "unexpected base change");
2071 // We know it's an on heap access so base can't be null
2072 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2073 base = must_be_not_null(base, true);
2074 }
2075 return basic_plus_adr(base, offset);
2076 }
2077 }
2078
2079 //--------------------------inline_number_methods-----------------------------
2080 // inline int Integer.numberOfLeadingZeros(int)
2081 // inline int Long.numberOfLeadingZeros(long)
2082 //
2083 // inline int Integer.numberOfTrailingZeros(int)
2084 // inline int Long.numberOfTrailingZeros(long)
2085 //
2086 // inline int Integer.bitCount(int)
2087 // inline int Long.bitCount(long)
2088 //
2089 // inline char Character.reverseBytes(char)
2090 // inline short Short.reverseBytes(short)
2091 // inline int Integer.reverseBytes(int)
2092 // inline long Long.reverseBytes(long)
2093 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2094 Node* arg = argument(0);
2095 Node* n = nullptr;
2096 switch (id) {
2097 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2098 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2099 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2100 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2101 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2102 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2103 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2104 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2105 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2106 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2107 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2108 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2109 default: fatal_unexpected_iid(id); break;
2110 }
2111 set_result(_gvn.transform(n));
2112 return true;
2113 }
2114
2115 //--------------------------inline_bitshuffle_methods-----------------------------
2116 // inline int Integer.compress(int, int)
2117 // inline int Integer.expand(int, int)
2118 // inline long Long.compress(long, long)
2119 // inline long Long.expand(long, long)
2120 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2121 Node* n = nullptr;
2122 switch (id) {
2123 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2124 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2125 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2126 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2127 default: fatal_unexpected_iid(id); break;
2128 }
2129 set_result(_gvn.transform(n));
2130 return true;
2131 }
2132
2133 //--------------------------inline_number_methods-----------------------------
2134 // inline int Integer.compareUnsigned(int, int)
2135 // inline int Long.compareUnsigned(long, long)
2136 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2137 Node* arg1 = argument(0);
2138 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2139 Node* n = nullptr;
2140 switch (id) {
2141 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2142 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2143 default: fatal_unexpected_iid(id); break;
2144 }
2145 set_result(_gvn.transform(n));
2146 return true;
2147 }
2148
2149 //--------------------------inline_unsigned_divmod_methods-----------------------------
2150 // inline int Integer.divideUnsigned(int, int)
2151 // inline int Integer.remainderUnsigned(int, int)
2152 // inline long Long.divideUnsigned(long, long)
2153 // inline long Long.remainderUnsigned(long, long)
2154 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2155 Node* n = nullptr;
2156 switch (id) {
2157 case vmIntrinsics::_divideUnsigned_i: {
2158 zero_check_int(argument(1));
2159 // Compile-time detect of null-exception
2160 if (stopped()) {
2161 return true; // keep the graph constructed so far
2162 }
2163 n = new UDivINode(control(), argument(0), argument(1));
2164 break;
2165 }
2166 case vmIntrinsics::_divideUnsigned_l: {
2167 zero_check_long(argument(2));
2168 // Compile-time detect of null-exception
2169 if (stopped()) {
2170 return true; // keep the graph constructed so far
2171 }
2172 n = new UDivLNode(control(), argument(0), argument(2));
2173 break;
2174 }
2175 case vmIntrinsics::_remainderUnsigned_i: {
2176 zero_check_int(argument(1));
2177 // Compile-time detect of null-exception
2178 if (stopped()) {
2179 return true; // keep the graph constructed so far
2180 }
2181 n = new UModINode(control(), argument(0), argument(1));
2182 break;
2183 }
2184 case vmIntrinsics::_remainderUnsigned_l: {
2185 zero_check_long(argument(2));
2186 // Compile-time detect of null-exception
2187 if (stopped()) {
2188 return true; // keep the graph constructed so far
2189 }
2190 n = new UModLNode(control(), argument(0), argument(2));
2191 break;
2192 }
2193 default: fatal_unexpected_iid(id); break;
2194 }
2195 set_result(_gvn.transform(n));
2196 return true;
2197 }
2198
2199 //----------------------------inline_unsafe_access----------------------------
2200
2201 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2202 // Attempt to infer a sharper value type from the offset and base type.
2203 ciKlass* sharpened_klass = nullptr;
2204
2205 // See if it is an instance field, with an object type.
2206 if (alias_type->field() != nullptr) {
2207 if (alias_type->field()->type()->is_klass()) {
2208 sharpened_klass = alias_type->field()->type()->as_klass();
2209 }
2210 }
2211
2212 const TypeOopPtr* result = nullptr;
2213 // See if it is a narrow oop array.
2214 if (adr_type->isa_aryptr()) {
2215 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2216 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2217 if (elem_type != nullptr && elem_type->is_loaded()) {
2218 // Sharpen the value type.
2219 result = elem_type;
2220 }
2221 }
2222 }
2223
2224 // The sharpened class might be unloaded if there is no class loader
2225 // contraint in place.
2226 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2227 // Sharpen the value type.
2228 result = TypeOopPtr::make_from_klass(sharpened_klass);
2229 }
2230 if (result != nullptr) {
2231 #ifndef PRODUCT
2232 if (C->print_intrinsics() || C->print_inlining()) {
2233 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2234 tty->print(" sharpened value: "); result->dump(); tty->cr();
2235 }
2236 #endif
2237 }
2238 return result;
2239 }
2240
2241 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2242 switch (kind) {
2243 case Relaxed:
2244 return MO_UNORDERED;
2245 case Opaque:
2246 return MO_RELAXED;
2247 case Acquire:
2248 return MO_ACQUIRE;
2249 case Release:
2250 return MO_RELEASE;
2251 case Volatile:
2252 return MO_SEQ_CST;
2253 default:
2254 ShouldNotReachHere();
2255 return 0;
2256 }
2257 }
2258
2259 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2260 if (callee()->is_static()) return false; // caller must have the capability!
2261 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2262 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2263 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2264 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2265
2266 if (is_reference_type(type)) {
2267 decorators |= ON_UNKNOWN_OOP_REF;
2268 }
2269
2270 if (unaligned) {
2271 decorators |= C2_UNALIGNED;
2272 }
2273
2274 #ifndef PRODUCT
2275 {
2276 ResourceMark rm;
2277 // Check the signatures.
2278 ciSignature* sig = callee()->signature();
2279 #ifdef ASSERT
2280 if (!is_store) {
2281 // Object getReference(Object base, int/long offset), etc.
2282 BasicType rtype = sig->return_type()->basic_type();
2283 assert(rtype == type, "getter must return the expected value");
2284 assert(sig->count() == 2, "oop getter has 2 arguments");
2285 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2286 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2287 } else {
2288 // void putReference(Object base, int/long offset, Object x), etc.
2289 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2290 assert(sig->count() == 3, "oop putter has 3 arguments");
2291 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2292 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2293 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2294 assert(vtype == type, "putter must accept the expected value");
2295 }
2296 #endif // ASSERT
2297 }
2298 #endif //PRODUCT
2299
2300 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2301
2302 Node* receiver = argument(0); // type: oop
2303
2304 // Build address expression.
2305 Node* heap_base_oop = top();
2306
2307 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2308 Node* base = argument(1); // type: oop
2309 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2310 Node* offset = argument(2); // type: long
2311 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2312 // to be plain byte offsets, which are also the same as those accepted
2313 // by oopDesc::field_addr.
2314 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2315 "fieldOffset must be byte-scaled");
2316 // 32-bit machines ignore the high half!
2317 offset = ConvL2X(offset);
2318
2319 // Save state and restore on bailout
2320 uint old_sp = sp();
2321 SafePointNode* old_map = clone_map();
2322
2323 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2324
2325 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2326 if (type != T_OBJECT) {
2327 decorators |= IN_NATIVE; // off-heap primitive access
2328 } else {
2329 set_map(old_map);
2330 set_sp(old_sp);
2331 return false; // off-heap oop accesses are not supported
2332 }
2333 } else {
2334 heap_base_oop = base; // on-heap or mixed access
2335 }
2336
2337 // Can base be null? Otherwise, always on-heap access.
2338 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2339
2340 if (!can_access_non_heap) {
2341 decorators |= IN_HEAP;
2342 }
2343
2344 Node* val = is_store ? argument(4) : nullptr;
2345
2346 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2347 if (adr_type == TypePtr::NULL_PTR) {
2348 set_map(old_map);
2349 set_sp(old_sp);
2350 return false; // off-heap access with zero address
2351 }
2352
2353 // Try to categorize the address.
2354 Compile::AliasType* alias_type = C->alias_type(adr_type);
2355 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2356
2357 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2358 alias_type->adr_type() == TypeAryPtr::RANGE) {
2359 set_map(old_map);
2360 set_sp(old_sp);
2361 return false; // not supported
2362 }
2363
2364 bool mismatched = false;
2365 BasicType bt = alias_type->basic_type();
2366 if (bt != T_ILLEGAL) {
2367 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2368 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2369 // Alias type doesn't differentiate between byte[] and boolean[]).
2370 // Use address type to get the element type.
2371 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2372 }
2373 if (is_reference_type(bt, true)) {
2374 // accessing an array field with getReference is not a mismatch
2375 bt = T_OBJECT;
2376 }
2377 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2378 // Don't intrinsify mismatched object accesses
2379 set_map(old_map);
2380 set_sp(old_sp);
2381 return false;
2382 }
2383 mismatched = (bt != type);
2384 } else if (alias_type->adr_type()->isa_oopptr()) {
2385 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2386 }
2387
2388 destruct_map_clone(old_map);
2389 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2390
2391 if (mismatched) {
2392 decorators |= C2_MISMATCHED;
2393 }
2394
2395 // First guess at the value type.
2396 const Type *value_type = Type::get_const_basic_type(type);
2397
2398 // Figure out the memory ordering.
2399 decorators |= mo_decorator_for_access_kind(kind);
2400
2401 if (!is_store && type == T_OBJECT) {
2402 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2403 if (tjp != nullptr) {
2404 value_type = tjp;
2405 }
2406 }
2407
2408 receiver = null_check(receiver);
2409 if (stopped()) {
2410 return true;
2411 }
2412 // Heap pointers get a null-check from the interpreter,
2413 // as a courtesy. However, this is not guaranteed by Unsafe,
2414 // and it is not possible to fully distinguish unintended nulls
2415 // from intended ones in this API.
2416
2417 if (!is_store) {
2418 Node* p = nullptr;
2419 // Try to constant fold a load from a constant field
2420 ciField* field = alias_type->field();
2421 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2422 // final or stable field
2423 p = make_constant_from_field(field, heap_base_oop);
2424 }
2425
2426 if (p == nullptr) { // Could not constant fold the load
2427 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2428 // Normalize the value returned by getBoolean in the following cases
2429 if (type == T_BOOLEAN &&
2430 (mismatched ||
2431 heap_base_oop == top() || // - heap_base_oop is null or
2432 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2433 // and the unsafe access is made to large offset
2434 // (i.e., larger than the maximum offset necessary for any
2435 // field access)
2436 ) {
2437 IdealKit ideal = IdealKit(this);
2438 #define __ ideal.
2439 IdealVariable normalized_result(ideal);
2440 __ declarations_done();
2441 __ set(normalized_result, p);
2442 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2443 __ set(normalized_result, ideal.ConI(1));
2444 ideal.end_if();
2445 final_sync(ideal);
2446 p = __ value(normalized_result);
2447 #undef __
2448 }
2449 }
2450 if (type == T_ADDRESS) {
2451 p = gvn().transform(new CastP2XNode(nullptr, p));
2452 p = ConvX2UL(p);
2453 }
2454 // The load node has the control of the preceding MemBarCPUOrder. All
2455 // following nodes will have the control of the MemBarCPUOrder inserted at
2456 // the end of this method. So, pushing the load onto the stack at a later
2457 // point is fine.
2458 set_result(p);
2459 } else {
2460 if (bt == T_ADDRESS) {
2461 // Repackage the long as a pointer.
2462 val = ConvL2X(val);
2463 val = gvn().transform(new CastX2PNode(val));
2464 }
2465 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2466 }
2467
2468 return true;
2469 }
2470
2471 //----------------------------inline_unsafe_load_store----------------------------
2472 // This method serves a couple of different customers (depending on LoadStoreKind):
2473 //
2474 // LS_cmp_swap:
2475 //
2476 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2477 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2478 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2479 //
2480 // LS_cmp_swap_weak:
2481 //
2482 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2483 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2484 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2485 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2486 //
2487 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2488 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2489 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2490 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2491 //
2492 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2493 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2494 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2495 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2496 //
2497 // LS_cmp_exchange:
2498 //
2499 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2500 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2501 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2502 //
2503 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2504 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2505 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2506 //
2507 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2508 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2509 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2510 //
2511 // LS_get_add:
2512 //
2513 // int getAndAddInt( Object o, long offset, int delta)
2514 // long getAndAddLong(Object o, long offset, long delta)
2515 //
2516 // LS_get_set:
2517 //
2518 // int getAndSet(Object o, long offset, int newValue)
2519 // long getAndSet(Object o, long offset, long newValue)
2520 // Object getAndSet(Object o, long offset, Object newValue)
2521 //
2522 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2523 // This basic scheme here is the same as inline_unsafe_access, but
2524 // differs in enough details that combining them would make the code
2525 // overly confusing. (This is a true fact! I originally combined
2526 // them, but even I was confused by it!) As much code/comments as
2527 // possible are retained from inline_unsafe_access though to make
2528 // the correspondences clearer. - dl
2529
2530 if (callee()->is_static()) return false; // caller must have the capability!
2531
2532 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2533 decorators |= mo_decorator_for_access_kind(access_kind);
2534
2535 #ifndef PRODUCT
2536 BasicType rtype;
2537 {
2538 ResourceMark rm;
2539 // Check the signatures.
2540 ciSignature* sig = callee()->signature();
2541 rtype = sig->return_type()->basic_type();
2542 switch(kind) {
2543 case LS_get_add:
2544 case LS_get_set: {
2545 // Check the signatures.
2546 #ifdef ASSERT
2547 assert(rtype == type, "get and set must return the expected type");
2548 assert(sig->count() == 3, "get and set has 3 arguments");
2549 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2550 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2551 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2552 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2553 #endif // ASSERT
2554 break;
2555 }
2556 case LS_cmp_swap:
2557 case LS_cmp_swap_weak: {
2558 // Check the signatures.
2559 #ifdef ASSERT
2560 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2561 assert(sig->count() == 4, "CAS has 4 arguments");
2562 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2563 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2564 #endif // ASSERT
2565 break;
2566 }
2567 case LS_cmp_exchange: {
2568 // Check the signatures.
2569 #ifdef ASSERT
2570 assert(rtype == type, "CAS must return the expected type");
2571 assert(sig->count() == 4, "CAS has 4 arguments");
2572 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2573 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2574 #endif // ASSERT
2575 break;
2576 }
2577 default:
2578 ShouldNotReachHere();
2579 }
2580 }
2581 #endif //PRODUCT
2582
2583 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2584
2585 // Get arguments:
2586 Node* receiver = nullptr;
2587 Node* base = nullptr;
2588 Node* offset = nullptr;
2589 Node* oldval = nullptr;
2590 Node* newval = nullptr;
2591 switch(kind) {
2592 case LS_cmp_swap:
2593 case LS_cmp_swap_weak:
2594 case LS_cmp_exchange: {
2595 const bool two_slot_type = type2size[type] == 2;
2596 receiver = argument(0); // type: oop
2597 base = argument(1); // type: oop
2598 offset = argument(2); // type: long
2599 oldval = argument(4); // type: oop, int, or long
2600 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2601 break;
2602 }
2603 case LS_get_add:
2604 case LS_get_set: {
2605 receiver = argument(0); // type: oop
2606 base = argument(1); // type: oop
2607 offset = argument(2); // type: long
2608 oldval = nullptr;
2609 newval = argument(4); // type: oop, int, or long
2610 break;
2611 }
2612 default:
2613 ShouldNotReachHere();
2614 }
2615
2616 // Build field offset expression.
2617 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2618 // to be plain byte offsets, which are also the same as those accepted
2619 // by oopDesc::field_addr.
2620 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2621 // 32-bit machines ignore the high half of long offsets
2622 offset = ConvL2X(offset);
2623 // Save state and restore on bailout
2624 uint old_sp = sp();
2625 SafePointNode* old_map = clone_map();
2626 Node* adr = make_unsafe_address(base, offset,type, false);
2627 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2628
2629 Compile::AliasType* alias_type = C->alias_type(adr_type);
2630 BasicType bt = alias_type->basic_type();
2631 if (bt != T_ILLEGAL &&
2632 (is_reference_type(bt) != (type == T_OBJECT))) {
2633 // Don't intrinsify mismatched object accesses.
2634 set_map(old_map);
2635 set_sp(old_sp);
2636 return false;
2637 }
2638
2639 destruct_map_clone(old_map);
2640
2641 // For CAS, unlike inline_unsafe_access, there seems no point in
2642 // trying to refine types. Just use the coarse types here.
2643 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2644 const Type *value_type = Type::get_const_basic_type(type);
2645
2646 switch (kind) {
2647 case LS_get_set:
2648 case LS_cmp_exchange: {
2649 if (type == T_OBJECT) {
2650 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2651 if (tjp != nullptr) {
2652 value_type = tjp;
2653 }
2654 }
2655 break;
2656 }
2657 case LS_cmp_swap:
2658 case LS_cmp_swap_weak:
2659 case LS_get_add:
2660 break;
2661 default:
2662 ShouldNotReachHere();
2663 }
2664
2665 // Null check receiver.
2666 receiver = null_check(receiver);
2667 if (stopped()) {
2668 return true;
2669 }
2670
2671 int alias_idx = C->get_alias_index(adr_type);
2672
2673 if (is_reference_type(type)) {
2674 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2675
2676 // Transformation of a value which could be null pointer (CastPP #null)
2677 // could be delayed during Parse (for example, in adjust_map_after_if()).
2678 // Execute transformation here to avoid barrier generation in such case.
2679 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2680 newval = _gvn.makecon(TypePtr::NULL_PTR);
2681
2682 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2683 // Refine the value to a null constant, when it is known to be null
2684 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2685 }
2686 }
2687
2688 Node* result = nullptr;
2689 switch (kind) {
2690 case LS_cmp_exchange: {
2691 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2692 oldval, newval, value_type, type, decorators);
2693 break;
2694 }
2695 case LS_cmp_swap_weak:
2696 decorators |= C2_WEAK_CMPXCHG;
2697 case LS_cmp_swap: {
2698 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2699 oldval, newval, value_type, type, decorators);
2700 break;
2701 }
2702 case LS_get_set: {
2703 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2704 newval, value_type, type, decorators);
2705 break;
2706 }
2707 case LS_get_add: {
2708 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2709 newval, value_type, type, decorators);
2710 break;
2711 }
2712 default:
2713 ShouldNotReachHere();
2714 }
2715
2716 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2717 set_result(result);
2718 return true;
2719 }
2720
2721 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2722 // Regardless of form, don't allow previous ld/st to move down,
2723 // then issue acquire, release, or volatile mem_bar.
2724 insert_mem_bar(Op_MemBarCPUOrder);
2725 switch(id) {
2726 case vmIntrinsics::_loadFence:
2727 insert_mem_bar(Op_LoadFence);
2728 return true;
2729 case vmIntrinsics::_storeFence:
2730 insert_mem_bar(Op_StoreFence);
2731 return true;
2732 case vmIntrinsics::_storeStoreFence:
2733 insert_mem_bar(Op_StoreStoreFence);
2734 return true;
2735 case vmIntrinsics::_fullFence:
2736 insert_mem_bar(Op_MemBarVolatile);
2737 return true;
2738 default:
2739 fatal_unexpected_iid(id);
2740 return false;
2741 }
2742 }
2743
2744 bool LibraryCallKit::inline_onspinwait() {
2745 insert_mem_bar(Op_OnSpinWait);
2746 return true;
2747 }
2748
2749 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2750 if (!kls->is_Con()) {
2751 return true;
2752 }
2753 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2754 if (klsptr == nullptr) {
2755 return true;
2756 }
2757 ciInstanceKlass* ik = klsptr->instance_klass();
2758 // don't need a guard for a klass that is already initialized
2759 return !ik->is_initialized();
2760 }
2761
2762 //----------------------------inline_unsafe_writeback0-------------------------
2763 // public native void Unsafe.writeback0(long address)
2764 bool LibraryCallKit::inline_unsafe_writeback0() {
2765 if (!Matcher::has_match_rule(Op_CacheWB)) {
2766 return false;
2767 }
2768 #ifndef PRODUCT
2769 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2770 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2771 ciSignature* sig = callee()->signature();
2772 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2773 #endif
2774 null_check_receiver(); // null-check, then ignore
2775 Node *addr = argument(1);
2776 addr = new CastX2PNode(addr);
2777 addr = _gvn.transform(addr);
2778 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2779 flush = _gvn.transform(flush);
2780 set_memory(flush, TypeRawPtr::BOTTOM);
2781 return true;
2782 }
2783
2784 //----------------------------inline_unsafe_writeback0-------------------------
2785 // public native void Unsafe.writeback0(long address)
2786 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2787 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2788 return false;
2789 }
2790 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2791 return false;
2792 }
2793 #ifndef PRODUCT
2794 assert(Matcher::has_match_rule(Op_CacheWB),
2795 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2796 : "found match rule for CacheWBPostSync but not CacheWB"));
2797
2798 #endif
2799 null_check_receiver(); // null-check, then ignore
2800 Node *sync;
2801 if (is_pre) {
2802 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2803 } else {
2804 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2805 }
2806 sync = _gvn.transform(sync);
2807 set_memory(sync, TypeRawPtr::BOTTOM);
2808 return true;
2809 }
2810
2811 //----------------------------inline_unsafe_allocate---------------------------
2812 // public native Object Unsafe.allocateInstance(Class<?> cls);
2813 bool LibraryCallKit::inline_unsafe_allocate() {
2814
2815 #if INCLUDE_JVMTI
2816 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2817 return false;
2818 }
2819 #endif //INCLUDE_JVMTI
2820
2821 if (callee()->is_static()) return false; // caller must have the capability!
2822
2823 null_check_receiver(); // null-check, then ignore
2824 Node* cls = null_check(argument(1));
2825 if (stopped()) return true;
2826
2827 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2828 kls = null_check(kls);
2829 if (stopped()) return true; // argument was like int.class
2830
2831 #if INCLUDE_JVMTI
2832 // Don't try to access new allocated obj in the intrinsic.
2833 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2834 // Deoptimize and allocate in interpreter instead.
2835 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2836 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2837 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2838 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2839 {
2840 BuildCutout unless(this, tst, PROB_MAX);
2841 uncommon_trap(Deoptimization::Reason_intrinsic,
2842 Deoptimization::Action_make_not_entrant);
2843 }
2844 if (stopped()) {
2845 return true;
2846 }
2847 #endif //INCLUDE_JVMTI
2848
2849 Node* test = nullptr;
2850 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2851 // Note: The argument might still be an illegal value like
2852 // Serializable.class or Object[].class. The runtime will handle it.
2853 // But we must make an explicit check for initialization.
2854 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2855 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2856 // can generate code to load it as unsigned byte.
2857 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2858 Node* bits = intcon(InstanceKlass::fully_initialized);
2859 test = _gvn.transform(new SubINode(inst, bits));
2860 // The 'test' is non-zero if we need to take a slow path.
2861 }
2862
2863 Node* obj = new_instance(kls, test);
2864 set_result(obj);
2865 return true;
2866 }
2867
2868 //------------------------inline_native_time_funcs--------------
2869 // inline code for System.currentTimeMillis() and System.nanoTime()
2870 // these have the same type and signature
2871 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2872 const TypeFunc* tf = OptoRuntime::void_long_Type();
2873 const TypePtr* no_memory_effects = nullptr;
2874 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2875 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2876 #ifdef ASSERT
2877 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2878 assert(value_top == top(), "second value must be top");
2879 #endif
2880 set_result(value);
2881 return true;
2882 }
2883
2884
2885 #if INCLUDE_JVMTI
2886
2887 // When notifications are disabled then just update the VTMS transition bit and return.
2888 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2889 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2890 if (!DoJVMTIVirtualThreadTransitions) {
2891 return true;
2892 }
2893 IdealKit ideal(this);
2894
2895 Node* ONE = ideal.ConI(1);
2896 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2897 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2898 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2899
2900 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2901 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2902 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2903 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2904
2905 sync_kit(ideal);
2906 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2907 ideal.sync_kit(this);
2908 } ideal.else_(); {
2909 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2910 Node* vt_oop = _gvn.transform(argument(0)); // this argument - VirtualThread oop
2911 Node* thread = ideal.thread();
2912 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2913 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2914 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2915
2916 sync_kit(ideal);
2917 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2918 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2919
2920 ideal.sync_kit(this);
2921 } ideal.end_if();
2922 final_sync(ideal);
2923
2924 return true;
2925 }
2926
2927 // Always update the temporary VTMS transition bit.
2928 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
2929 if (!DoJVMTIVirtualThreadTransitions) {
2930 return true;
2931 }
2932 IdealKit ideal(this);
2933
2934 {
2935 // unconditionally update the temporary VTMS transition bit in current JavaThread
2936 Node* thread = ideal.thread();
2937 Node* hide = _gvn.transform(argument(1)); // hide argument for temporary VTMS transition notification
2938 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
2939 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2940
2941 sync_kit(ideal);
2942 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2943 ideal.sync_kit(this);
2944 }
2945 final_sync(ideal);
2946
2947 return true;
2948 }
2949
2950 #endif // INCLUDE_JVMTI
2951
2952 #ifdef JFR_HAVE_INTRINSICS
2953
2954 /**
2955 * if oop->klass != null
2956 * // normal class
2957 * epoch = _epoch_state ? 2 : 1
2958 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2959 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2960 * }
2961 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2962 * else
2963 * // primitive class
2964 * if oop->array_klass != null
2965 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2966 * else
2967 * id = LAST_TYPE_ID + 1 // void class path
2968 * if (!signaled)
2969 * signaled = true
2970 */
2971 bool LibraryCallKit::inline_native_classID() {
2972 Node* cls = argument(0);
2973
2974 IdealKit ideal(this);
2975 #define __ ideal.
2976 IdealVariable result(ideal); __ declarations_done();
2977 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
2978 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2979 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2980
2981
2982 __ if_then(kls, BoolTest::ne, null()); {
2983 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2984 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2985
2986 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
2987 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2988 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2989 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2990 mask = _gvn.transform(new OrLNode(mask, epoch));
2991 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2992
2993 float unlikely = PROB_UNLIKELY(0.999);
2994 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
2995 sync_kit(ideal);
2996 make_runtime_call(RC_LEAF,
2997 OptoRuntime::class_id_load_barrier_Type(),
2998 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
2999 "class id load barrier",
3000 TypePtr::BOTTOM,
3001 kls);
3002 ideal.sync_kit(this);
3003 } __ end_if();
3004
3005 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3006 } __ else_(); {
3007 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3008 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3009 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3010 __ if_then(array_kls, BoolTest::ne, null()); {
3011 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3012 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3013 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3014 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3015 } __ else_(); {
3016 // void class case
3017 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3018 } __ end_if();
3019
3020 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3021 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3022 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3023 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3024 } __ end_if();
3025 } __ end_if();
3026
3027 final_sync(ideal);
3028 set_result(ideal.value(result));
3029 #undef __
3030 return true;
3031 }
3032
3033 //------------------------inline_native_jvm_commit------------------
3034 bool LibraryCallKit::inline_native_jvm_commit() {
3035 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3036
3037 // Save input memory and i_o state.
3038 Node* input_memory_state = reset_memory();
3039 set_all_memory(input_memory_state);
3040 Node* input_io_state = i_o();
3041
3042 // TLS.
3043 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3044 // Jfr java buffer.
3045 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3046 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3047 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3048
3049 // Load the current value of the notified field in the JfrThreadLocal.
3050 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3051 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3052
3053 // Test for notification.
3054 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3055 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3056 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3057
3058 // True branch, is notified.
3059 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3060 set_control(is_notified);
3061
3062 // Reset notified state.
3063 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3064
3065 // Iff notified, the return address of the commit method is the current position of the backing java buffer. This is used to reset the event writer.
3066 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3067 // Convert the machine-word to a long.
3068 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3069
3070 // False branch, not notified.
3071 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3072 set_control(not_notified);
3073 set_all_memory(input_memory_state);
3074
3075 // Arg is the next position as a long.
3076 Node* arg = argument(0);
3077 // Convert long to machine-word.
3078 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3079
3080 // Store the next_position to the underlying jfr java buffer.
3081 Node* commit_memory;
3082 #ifdef _LP64
3083 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3084 #else
3085 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3086 #endif
3087
3088 // Now load the flags from off the java buffer and decide if the buffer is a lease. If so, it needs to be returned post-commit.
3089 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3090 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3091 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3092
3093 // And flags with lease constant.
3094 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3095
3096 // Branch on lease to conditionalize returning the leased java buffer.
3097 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3098 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3099 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3100
3101 // False branch, not a lease.
3102 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3103
3104 // True branch, is lease.
3105 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3106 set_control(is_lease);
3107
3108 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3109 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3110 OptoRuntime::void_void_Type(),
3111 StubRoutines::jfr_return_lease(),
3112 "return_lease", TypePtr::BOTTOM);
3113 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3114
3115 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3116 record_for_igvn(lease_compare_rgn);
3117 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3118 record_for_igvn(lease_compare_mem);
3119 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3120 record_for_igvn(lease_compare_io);
3121 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3122 record_for_igvn(lease_result_value);
3123
3124 // Update control and phi nodes.
3125 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3126 lease_compare_rgn->init_req(_false_path, not_lease);
3127
3128 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3129 lease_compare_mem->init_req(_false_path, commit_memory);
3130
3131 lease_compare_io->init_req(_true_path, i_o());
3132 lease_compare_io->init_req(_false_path, input_io_state);
3133
3134 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3135 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3136
3137 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3138 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3139 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3140 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3141
3142 // Update control and phi nodes.
3143 result_rgn->init_req(_true_path, is_notified);
3144 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3145
3146 result_mem->init_req(_true_path, notified_reset_memory);
3147 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3148
3149 result_io->init_req(_true_path, input_io_state);
3150 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3151
3152 result_value->init_req(_true_path, current_pos);
3153 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3154
3155 // Set output state.
3156 set_control(_gvn.transform(result_rgn));
3157 set_all_memory(_gvn.transform(result_mem));
3158 set_i_o(_gvn.transform(result_io));
3159 set_result(result_rgn, result_value);
3160 return true;
3161 }
3162
3163 /*
3164 * The intrinsic is a model of this pseudo-code:
3165 *
3166 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3167 * jobject h_event_writer = tl->java_event_writer();
3168 * if (h_event_writer == nullptr) {
3169 * return nullptr;
3170 * }
3171 * oop threadObj = Thread::threadObj();
3172 * oop vthread = java_lang_Thread::vthread(threadObj);
3173 * traceid tid;
3174 * bool excluded;
3175 * if (vthread != threadObj) { // i.e. current thread is virtual
3176 * tid = java_lang_Thread::tid(vthread);
3177 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3178 * excluded = vthread_epoch_raw & excluded_mask;
3179 * if (!excluded) {
3180 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3181 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3182 * if (vthread_epoch != current_epoch) {
3183 * write_checkpoint();
3184 * }
3185 * }
3186 * } else {
3187 * tid = java_lang_Thread::tid(threadObj);
3188 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3189 * excluded = thread_epoch_raw & excluded_mask;
3190 * }
3191 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3192 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3193 * if (tid_in_event_writer != tid) {
3194 * setField(event_writer, "threadID", tid);
3195 * setField(event_writer, "excluded", excluded);
3196 * }
3197 * return event_writer
3198 */
3199 bool LibraryCallKit::inline_native_getEventWriter() {
3200 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3201
3202 // Save input memory and i_o state.
3203 Node* input_memory_state = reset_memory();
3204 set_all_memory(input_memory_state);
3205 Node* input_io_state = i_o();
3206
3207 Node* excluded_mask = _gvn.intcon(32768);
3208 Node* epoch_mask = _gvn.intcon(32767);
3209
3210 // TLS
3211 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3212
3213 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3214 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3215
3216 // Load the eventwriter jobject handle.
3217 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3218
3219 // Null check the jobject handle.
3220 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3221 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3222 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3223
3224 // False path, jobj is null.
3225 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3226
3227 // True path, jobj is not null.
3228 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3229
3230 set_control(jobj_is_not_null);
3231
3232 // Load the threadObj for the CarrierThread.
3233 Node* threadObj = generate_current_thread(tls_ptr);
3234
3235 // Load the vthread.
3236 Node* vthread = generate_virtual_thread(tls_ptr);
3237
3238 // If vthread != threadObj, this is a virtual thread.
3239 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3240 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3241 IfNode* iff_vthread_not_equal_threadObj =
3242 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3243
3244 // False branch, fallback to threadObj.
3245 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3246 set_control(vthread_equal_threadObj);
3247
3248 // Load the tid field from the vthread object.
3249 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3250
3251 // Load the raw epoch value from the threadObj.
3252 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3253 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3254 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3255
3256 // Mask off the excluded information from the epoch.
3257 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3258
3259 // True branch, this is a virtual thread.
3260 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3261 set_control(vthread_not_equal_threadObj);
3262
3263 // Load the tid field from the vthread object.
3264 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3265
3266 // Load the raw epoch value from the vthread.
3267 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3268 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3269 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3270
3271 // Mask off the excluded information from the epoch.
3272 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3273
3274 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3275 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3276 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3277 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3278
3279 // False branch, vthread is excluded, no need to write epoch info.
3280 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3281
3282 // True branch, vthread is included, update epoch info.
3283 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3284 set_control(included);
3285
3286 // Get epoch value.
3287 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3288
3289 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3290 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3291 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3292
3293 // Compare the epoch in the vthread to the current epoch generation.
3294 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3295 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3296 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3297
3298 // False path, epoch is equal, checkpoint information is valid.
3299 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3300
3301 // True path, epoch is not equal, write a checkpoint for the vthread.
3302 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3303
3304 set_control(epoch_is_not_equal);
3305
3306 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3307 // The call also updates the native thread local thread id and the vthread with the current epoch.
3308 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3309 OptoRuntime::jfr_write_checkpoint_Type(),
3310 StubRoutines::jfr_write_checkpoint(),
3311 "write_checkpoint", TypePtr::BOTTOM);
3312 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3313
3314 // vthread epoch != current epoch
3315 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3316 record_for_igvn(epoch_compare_rgn);
3317 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3318 record_for_igvn(epoch_compare_mem);
3319 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3320 record_for_igvn(epoch_compare_io);
3321
3322 // Update control and phi nodes.
3323 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3324 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3325 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3326 epoch_compare_mem->init_req(_false_path, input_memory_state);
3327 epoch_compare_io->init_req(_true_path, i_o());
3328 epoch_compare_io->init_req(_false_path, input_io_state);
3329
3330 // excluded != true
3331 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3332 record_for_igvn(exclude_compare_rgn);
3333 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3334 record_for_igvn(exclude_compare_mem);
3335 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3336 record_for_igvn(exclude_compare_io);
3337
3338 // Update control and phi nodes.
3339 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3340 exclude_compare_rgn->init_req(_false_path, excluded);
3341 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3342 exclude_compare_mem->init_req(_false_path, input_memory_state);
3343 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3344 exclude_compare_io->init_req(_false_path, input_io_state);
3345
3346 // vthread != threadObj
3347 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3348 record_for_igvn(vthread_compare_rgn);
3349 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3350 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3351 record_for_igvn(vthread_compare_io);
3352 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3353 record_for_igvn(tid);
3354 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3355 record_for_igvn(exclusion);
3356
3357 // Update control and phi nodes.
3358 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3359 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3360 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3361 vthread_compare_mem->init_req(_false_path, input_memory_state);
3362 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3363 vthread_compare_io->init_req(_false_path, input_io_state);
3364 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3365 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3366 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3367 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3368
3369 // Update branch state.
3370 set_control(_gvn.transform(vthread_compare_rgn));
3371 set_all_memory(_gvn.transform(vthread_compare_mem));
3372 set_i_o(_gvn.transform(vthread_compare_io));
3373
3374 // Load the event writer oop by dereferencing the jobject handle.
3375 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3376 assert(klass_EventWriter->is_loaded(), "invariant");
3377 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3378 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3379 const TypeOopPtr* const xtype = aklass->as_instance_type();
3380 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3381 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3382
3383 // Load the current thread id from the event writer object.
3384 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3385 // Get the field offset to, conditionally, store an updated tid value later.
3386 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3387 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3388 // Get the field offset to, conditionally, store an updated exclusion value later.
3389 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3390 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3391
3392 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3393 record_for_igvn(event_writer_tid_compare_rgn);
3394 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3395 record_for_igvn(event_writer_tid_compare_mem);
3396 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3397 record_for_igvn(event_writer_tid_compare_io);
3398
3399 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3400 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3401 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3402 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3403
3404 // False path, tids are the same.
3405 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3406
3407 // True path, tid is not equal, need to update the tid in the event writer.
3408 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3409 record_for_igvn(tid_is_not_equal);
3410
3411 // Store the exclusion state to the event writer.
3412 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3413
3414 // Store the tid to the event writer.
3415 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3416
3417 // Update control and phi nodes.
3418 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3419 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3420 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3421 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3422 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3423 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3424
3425 // Result of top level CFG, Memory, IO and Value.
3426 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3427 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3428 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3429 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3430
3431 // Result control.
3432 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3433 result_rgn->init_req(_false_path, jobj_is_null);
3434
3435 // Result memory.
3436 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3437 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3438
3439 // Result IO.
3440 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3441 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3442
3443 // Result value.
3444 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3445 result_value->init_req(_false_path, null()); // return null
3446
3447 // Set output state.
3448 set_control(_gvn.transform(result_rgn));
3449 set_all_memory(_gvn.transform(result_mem));
3450 set_i_o(_gvn.transform(result_io));
3451 set_result(result_rgn, result_value);
3452 return true;
3453 }
3454
3455 /*
3456 * The intrinsic is a model of this pseudo-code:
3457 *
3458 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3459 * if (carrierThread != thread) { // is virtual thread
3460 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3461 * bool excluded = vthread_epoch_raw & excluded_mask;
3462 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3463 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3464 * if (!excluded) {
3465 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3466 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3467 * }
3468 * Atomic::release_store(&tl->_vthread, true);
3469 * return;
3470 * }
3471 * Atomic::release_store(&tl->_vthread, false);
3472 */
3473 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3474 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3475
3476 Node* input_memory_state = reset_memory();
3477 set_all_memory(input_memory_state);
3478
3479 Node* excluded_mask = _gvn.intcon(32768);
3480 Node* epoch_mask = _gvn.intcon(32767);
3481
3482 Node* const carrierThread = generate_current_thread(jt);
3483 // If thread != carrierThread, this is a virtual thread.
3484 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3485 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3486 IfNode* iff_thread_not_equal_carrierThread =
3487 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3488
3489 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3490
3491 // False branch, is carrierThread.
3492 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3493 // Store release
3494 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3495
3496 set_all_memory(input_memory_state);
3497
3498 // True branch, is virtual thread.
3499 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3500 set_control(thread_not_equal_carrierThread);
3501
3502 // Load the raw epoch value from the vthread.
3503 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3504 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3505 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3506
3507 // Mask off the excluded information from the epoch.
3508 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3509
3510 // Load the tid field from the thread.
3511 Node* tid = load_field_from_object(thread, "tid", "J");
3512
3513 // Store the vthread tid to the jfr thread local.
3514 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3515 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3516
3517 // Branch is_excluded to conditionalize updating the epoch .
3518 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3519 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3520 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3521
3522 // True branch, vthread is excluded, no need to write epoch info.
3523 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3524 set_control(excluded);
3525 Node* vthread_is_excluded = _gvn.intcon(1);
3526
3527 // False branch, vthread is included, update epoch info.
3528 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3529 set_control(included);
3530 Node* vthread_is_included = _gvn.intcon(0);
3531
3532 // Get epoch value.
3533 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3534
3535 // Store the vthread epoch to the jfr thread local.
3536 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3537 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3538
3539 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3540 record_for_igvn(excluded_rgn);
3541 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3542 record_for_igvn(excluded_mem);
3543 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3544 record_for_igvn(exclusion);
3545
3546 // Merge the excluded control and memory.
3547 excluded_rgn->init_req(_true_path, excluded);
3548 excluded_rgn->init_req(_false_path, included);
3549 excluded_mem->init_req(_true_path, tid_memory);
3550 excluded_mem->init_req(_false_path, included_memory);
3551 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3552 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3553
3554 // Set intermediate state.
3555 set_control(_gvn.transform(excluded_rgn));
3556 set_all_memory(excluded_mem);
3557
3558 // Store the vthread exclusion state to the jfr thread local.
3559 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3560 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3561
3562 // Store release
3563 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3564
3565 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3566 record_for_igvn(thread_compare_rgn);
3567 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3568 record_for_igvn(thread_compare_mem);
3569 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3570 record_for_igvn(vthread);
3571
3572 // Merge the thread_compare control and memory.
3573 thread_compare_rgn->init_req(_true_path, control());
3574 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3575 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3576 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3577
3578 // Set output state.
3579 set_control(_gvn.transform(thread_compare_rgn));
3580 set_all_memory(_gvn.transform(thread_compare_mem));
3581 }
3582
3583 #endif // JFR_HAVE_INTRINSICS
3584
3585 //------------------------inline_native_currentCarrierThread------------------
3586 bool LibraryCallKit::inline_native_currentCarrierThread() {
3587 Node* junk = nullptr;
3588 set_result(generate_current_thread(junk));
3589 return true;
3590 }
3591
3592 //------------------------inline_native_currentThread------------------
3593 bool LibraryCallKit::inline_native_currentThread() {
3594 Node* junk = nullptr;
3595 set_result(generate_virtual_thread(junk));
3596 return true;
3597 }
3598
3599 //------------------------inline_native_setVthread------------------
3600 bool LibraryCallKit::inline_native_setCurrentThread() {
3601 assert(C->method()->changes_current_thread(),
3602 "method changes current Thread but is not annotated ChangesCurrentThread");
3603 Node* arr = argument(1);
3604 Node* thread = _gvn.transform(new ThreadLocalNode());
3605 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3606 Node* thread_obj_handle
3607 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3608 thread_obj_handle = _gvn.transform(thread_obj_handle);
3609 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3610 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3611 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3612 return true;
3613 }
3614
3615 const Type* LibraryCallKit::scopedValueCache_type() {
3616 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3617 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3618 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3619
3620 // Because we create the scopedValue cache lazily we have to make the
3621 // type of the result BotPTR.
3622 bool xk = etype->klass_is_exact();
3623 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3624 return objects_type;
3625 }
3626
3627 Node* LibraryCallKit::scopedValueCache_helper() {
3628 Node* thread = _gvn.transform(new ThreadLocalNode());
3629 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3630 // We cannot use immutable_memory() because we might flip onto a
3631 // different carrier thread, at which point we'll need to use that
3632 // carrier thread's cache.
3633 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3634 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3635 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3636 }
3637
3638 //------------------------inline_native_scopedValueCache------------------
3639 bool LibraryCallKit::inline_native_scopedValueCache() {
3640 Node* cache_obj_handle = scopedValueCache_helper();
3641 const Type* objects_type = scopedValueCache_type();
3642 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3643
3644 return true;
3645 }
3646
3647 //------------------------inline_native_setScopedValueCache------------------
3648 bool LibraryCallKit::inline_native_setScopedValueCache() {
3649 Node* arr = argument(0);
3650 Node* cache_obj_handle = scopedValueCache_helper();
3651 const Type* objects_type = scopedValueCache_type();
3652
3653 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3654 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3655
3656 return true;
3657 }
3658
3659 //---------------------------load_mirror_from_klass----------------------------
3660 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3661 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3662 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3663 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3664 // mirror = ((OopHandle)mirror)->resolve();
3665 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3666 }
3667
3668 //-----------------------load_klass_from_mirror_common-------------------------
3669 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3670 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3671 // and branch to the given path on the region.
3672 // If never_see_null, take an uncommon trap on null, so we can optimistically
3673 // compile for the non-null case.
3674 // If the region is null, force never_see_null = true.
3675 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3676 bool never_see_null,
3677 RegionNode* region,
3678 int null_path,
3679 int offset) {
3680 if (region == nullptr) never_see_null = true;
3681 Node* p = basic_plus_adr(mirror, offset);
3682 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3683 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3684 Node* null_ctl = top();
3685 kls = null_check_oop(kls, &null_ctl, never_see_null);
3686 if (region != nullptr) {
3687 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3688 region->init_req(null_path, null_ctl);
3689 } else {
3690 assert(null_ctl == top(), "no loose ends");
3691 }
3692 return kls;
3693 }
3694
3695 //--------------------(inline_native_Class_query helpers)---------------------
3696 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3697 // Fall through if (mods & mask) == bits, take the guard otherwise.
3698 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3699 // Branch around if the given klass has the given modifier bit set.
3700 // Like generate_guard, adds a new path onto the region.
3701 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3702 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3703 Node* mask = intcon(modifier_mask);
3704 Node* bits = intcon(modifier_bits);
3705 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3706 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3707 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3708 return generate_fair_guard(bol, region);
3709 }
3710 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3711 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3712 }
3713 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3714 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3715 }
3716
3717 //-------------------------inline_native_Class_query-------------------
3718 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3719 const Type* return_type = TypeInt::BOOL;
3720 Node* prim_return_value = top(); // what happens if it's a primitive class?
3721 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3722 bool expect_prim = false; // most of these guys expect to work on refs
3723
3724 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3725
3726 Node* mirror = argument(0);
3727 Node* obj = top();
3728
3729 switch (id) {
3730 case vmIntrinsics::_isInstance:
3731 // nothing is an instance of a primitive type
3732 prim_return_value = intcon(0);
3733 obj = argument(1);
3734 break;
3735 case vmIntrinsics::_getModifiers:
3736 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3737 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3738 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3739 break;
3740 case vmIntrinsics::_isInterface:
3741 prim_return_value = intcon(0);
3742 break;
3743 case vmIntrinsics::_isArray:
3744 prim_return_value = intcon(0);
3745 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3746 break;
3747 case vmIntrinsics::_isPrimitive:
3748 prim_return_value = intcon(1);
3749 expect_prim = true; // obviously
3750 break;
3751 case vmIntrinsics::_isHidden:
3752 prim_return_value = intcon(0);
3753 break;
3754 case vmIntrinsics::_getSuperclass:
3755 prim_return_value = null();
3756 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3757 break;
3758 case vmIntrinsics::_getClassAccessFlags:
3759 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3760 return_type = TypeInt::INT; // not bool! 6297094
3761 break;
3762 default:
3763 fatal_unexpected_iid(id);
3764 break;
3765 }
3766
3767 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3768 if (mirror_con == nullptr) return false; // cannot happen?
3769
3770 #ifndef PRODUCT
3771 if (C->print_intrinsics() || C->print_inlining()) {
3772 ciType* k = mirror_con->java_mirror_type();
3773 if (k) {
3774 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3775 k->print_name();
3776 tty->cr();
3777 }
3778 }
3779 #endif
3780
3781 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3782 RegionNode* region = new RegionNode(PATH_LIMIT);
3783 record_for_igvn(region);
3784 PhiNode* phi = new PhiNode(region, return_type);
3785
3786 // The mirror will never be null of Reflection.getClassAccessFlags, however
3787 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3788 // if it is. See bug 4774291.
3789
3790 // For Reflection.getClassAccessFlags(), the null check occurs in
3791 // the wrong place; see inline_unsafe_access(), above, for a similar
3792 // situation.
3793 mirror = null_check(mirror);
3794 // If mirror or obj is dead, only null-path is taken.
3795 if (stopped()) return true;
3796
3797 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3798
3799 // Now load the mirror's klass metaobject, and null-check it.
3800 // Side-effects region with the control path if the klass is null.
3801 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3802 // If kls is null, we have a primitive mirror.
3803 phi->init_req(_prim_path, prim_return_value);
3804 if (stopped()) { set_result(region, phi); return true; }
3805 bool safe_for_replace = (region->in(_prim_path) == top());
3806
3807 Node* p; // handy temp
3808 Node* null_ctl;
3809
3810 // Now that we have the non-null klass, we can perform the real query.
3811 // For constant classes, the query will constant-fold in LoadNode::Value.
3812 Node* query_value = top();
3813 switch (id) {
3814 case vmIntrinsics::_isInstance:
3815 // nothing is an instance of a primitive type
3816 query_value = gen_instanceof(obj, kls, safe_for_replace);
3817 break;
3818
3819 case vmIntrinsics::_getModifiers:
3820 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3821 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3822 break;
3823
3824 case vmIntrinsics::_isInterface:
3825 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3826 if (generate_interface_guard(kls, region) != nullptr)
3827 // A guard was added. If the guard is taken, it was an interface.
3828 phi->add_req(intcon(1));
3829 // If we fall through, it's a plain class.
3830 query_value = intcon(0);
3831 break;
3832
3833 case vmIntrinsics::_isArray:
3834 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3835 if (generate_array_guard(kls, region) != nullptr)
3836 // A guard was added. If the guard is taken, it was an array.
3837 phi->add_req(intcon(1));
3838 // If we fall through, it's a plain class.
3839 query_value = intcon(0);
3840 break;
3841
3842 case vmIntrinsics::_isPrimitive:
3843 query_value = intcon(0); // "normal" path produces false
3844 break;
3845
3846 case vmIntrinsics::_isHidden:
3847 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3848 if (generate_hidden_class_guard(kls, region) != nullptr)
3849 // A guard was added. If the guard is taken, it was an hidden class.
3850 phi->add_req(intcon(1));
3851 // If we fall through, it's a plain class.
3852 query_value = intcon(0);
3853 break;
3854
3855
3856 case vmIntrinsics::_getSuperclass:
3857 // The rules here are somewhat unfortunate, but we can still do better
3858 // with random logic than with a JNI call.
3859 // Interfaces store null or Object as _super, but must report null.
3860 // Arrays store an intermediate super as _super, but must report Object.
3861 // Other types can report the actual _super.
3862 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3863 if (generate_interface_guard(kls, region) != nullptr)
3864 // A guard was added. If the guard is taken, it was an interface.
3865 phi->add_req(null());
3866 if (generate_array_guard(kls, region) != nullptr)
3867 // A guard was added. If the guard is taken, it was an array.
3868 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3869 // If we fall through, it's a plain class. Get its _super.
3870 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3871 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3872 null_ctl = top();
3873 kls = null_check_oop(kls, &null_ctl);
3874 if (null_ctl != top()) {
3875 // If the guard is taken, Object.superClass is null (both klass and mirror).
3876 region->add_req(null_ctl);
3877 phi ->add_req(null());
3878 }
3879 if (!stopped()) {
3880 query_value = load_mirror_from_klass(kls);
3881 }
3882 break;
3883
3884 case vmIntrinsics::_getClassAccessFlags:
3885 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3886 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3887 break;
3888
3889 default:
3890 fatal_unexpected_iid(id);
3891 break;
3892 }
3893
3894 // Fall-through is the normal case of a query to a real class.
3895 phi->init_req(1, query_value);
3896 region->init_req(1, control());
3897
3898 C->set_has_split_ifs(true); // Has chance for split-if optimization
3899 set_result(region, phi);
3900 return true;
3901 }
3902
3903 //-------------------------inline_Class_cast-------------------
3904 bool LibraryCallKit::inline_Class_cast() {
3905 Node* mirror = argument(0); // Class
3906 Node* obj = argument(1);
3907 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3908 if (mirror_con == nullptr) {
3909 return false; // dead path (mirror->is_top()).
3910 }
3911 if (obj == nullptr || obj->is_top()) {
3912 return false; // dead path
3913 }
3914 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3915
3916 // First, see if Class.cast() can be folded statically.
3917 // java_mirror_type() returns non-null for compile-time Class constants.
3918 ciType* tm = mirror_con->java_mirror_type();
3919 if (tm != nullptr && tm->is_klass() &&
3920 tp != nullptr) {
3921 if (!tp->is_loaded()) {
3922 // Don't use intrinsic when class is not loaded.
3923 return false;
3924 } else {
3925 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
3926 if (static_res == Compile::SSC_always_true) {
3927 // isInstance() is true - fold the code.
3928 set_result(obj);
3929 return true;
3930 } else if (static_res == Compile::SSC_always_false) {
3931 // Don't use intrinsic, have to throw ClassCastException.
3932 // If the reference is null, the non-intrinsic bytecode will
3933 // be optimized appropriately.
3934 return false;
3935 }
3936 }
3937 }
3938
3939 // Bailout intrinsic and do normal inlining if exception path is frequent.
3940 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3941 return false;
3942 }
3943
3944 // Generate dynamic checks.
3945 // Class.cast() is java implementation of _checkcast bytecode.
3946 // Do checkcast (Parse::do_checkcast()) optimizations here.
3947
3948 mirror = null_check(mirror);
3949 // If mirror is dead, only null-path is taken.
3950 if (stopped()) {
3951 return true;
3952 }
3953
3954 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
3955 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3956 RegionNode* region = new RegionNode(PATH_LIMIT);
3957 record_for_igvn(region);
3958
3959 // Now load the mirror's klass metaobject, and null-check it.
3960 // If kls is null, we have a primitive mirror and
3961 // nothing is an instance of a primitive type.
3962 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3963
3964 Node* res = top();
3965 if (!stopped()) {
3966 Node* bad_type_ctrl = top();
3967 // Do checkcast optimizations.
3968 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3969 region->init_req(_bad_type_path, bad_type_ctrl);
3970 }
3971 if (region->in(_prim_path) != top() ||
3972 region->in(_bad_type_path) != top()) {
3973 // Let Interpreter throw ClassCastException.
3974 PreserveJVMState pjvms(this);
3975 set_control(_gvn.transform(region));
3976 uncommon_trap(Deoptimization::Reason_intrinsic,
3977 Deoptimization::Action_maybe_recompile);
3978 }
3979 if (!stopped()) {
3980 set_result(res);
3981 }
3982 return true;
3983 }
3984
3985
3986 //--------------------------inline_native_subtype_check------------------------
3987 // This intrinsic takes the JNI calls out of the heart of
3988 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3989 bool LibraryCallKit::inline_native_subtype_check() {
3990 // Pull both arguments off the stack.
3991 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3992 args[0] = argument(0);
3993 args[1] = argument(1);
3994 Node* klasses[2]; // corresponding Klasses: superk, subk
3995 klasses[0] = klasses[1] = top();
3996
3997 enum {
3998 // A full decision tree on {superc is prim, subc is prim}:
3999 _prim_0_path = 1, // {P,N} => false
4000 // {P,P} & superc!=subc => false
4001 _prim_same_path, // {P,P} & superc==subc => true
4002 _prim_1_path, // {N,P} => false
4003 _ref_subtype_path, // {N,N} & subtype check wins => true
4004 _both_ref_path, // {N,N} & subtype check loses => false
4005 PATH_LIMIT
4006 };
4007
4008 RegionNode* region = new RegionNode(PATH_LIMIT);
4009 Node* phi = new PhiNode(region, TypeInt::BOOL);
4010 record_for_igvn(region);
4011
4012 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4013 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4014 int class_klass_offset = java_lang_Class::klass_offset();
4015
4016 // First null-check both mirrors and load each mirror's klass metaobject.
4017 int which_arg;
4018 for (which_arg = 0; which_arg <= 1; which_arg++) {
4019 Node* arg = args[which_arg];
4020 arg = null_check(arg);
4021 if (stopped()) break;
4022 args[which_arg] = arg;
4023
4024 Node* p = basic_plus_adr(arg, class_klass_offset);
4025 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4026 klasses[which_arg] = _gvn.transform(kls);
4027 }
4028
4029 // Having loaded both klasses, test each for null.
4030 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4031 for (which_arg = 0; which_arg <= 1; which_arg++) {
4032 Node* kls = klasses[which_arg];
4033 Node* null_ctl = top();
4034 kls = null_check_oop(kls, &null_ctl, never_see_null);
4035 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4036 region->init_req(prim_path, null_ctl);
4037 if (stopped()) break;
4038 klasses[which_arg] = kls;
4039 }
4040
4041 if (!stopped()) {
4042 // now we have two reference types, in klasses[0..1]
4043 Node* subk = klasses[1]; // the argument to isAssignableFrom
4044 Node* superk = klasses[0]; // the receiver
4045 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4046 // now we have a successful reference subtype check
4047 region->set_req(_ref_subtype_path, control());
4048 }
4049
4050 // If both operands are primitive (both klasses null), then
4051 // we must return true when they are identical primitives.
4052 // It is convenient to test this after the first null klass check.
4053 set_control(region->in(_prim_0_path)); // go back to first null check
4054 if (!stopped()) {
4055 // Since superc is primitive, make a guard for the superc==subc case.
4056 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4057 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4058 generate_guard(bol_eq, region, PROB_FAIR);
4059 if (region->req() == PATH_LIMIT+1) {
4060 // A guard was added. If the added guard is taken, superc==subc.
4061 region->swap_edges(PATH_LIMIT, _prim_same_path);
4062 region->del_req(PATH_LIMIT);
4063 }
4064 region->set_req(_prim_0_path, control()); // Not equal after all.
4065 }
4066
4067 // these are the only paths that produce 'true':
4068 phi->set_req(_prim_same_path, intcon(1));
4069 phi->set_req(_ref_subtype_path, intcon(1));
4070
4071 // pull together the cases:
4072 assert(region->req() == PATH_LIMIT, "sane region");
4073 for (uint i = 1; i < region->req(); i++) {
4074 Node* ctl = region->in(i);
4075 if (ctl == nullptr || ctl == top()) {
4076 region->set_req(i, top());
4077 phi ->set_req(i, top());
4078 } else if (phi->in(i) == nullptr) {
4079 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4080 }
4081 }
4082
4083 set_control(_gvn.transform(region));
4084 set_result(_gvn.transform(phi));
4085 return true;
4086 }
4087
4088 //---------------------generate_array_guard_common------------------------
4089 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4090 bool obj_array, bool not_array) {
4091
4092 if (stopped()) {
4093 return nullptr;
4094 }
4095
4096 // If obj_array/non_array==false/false:
4097 // Branch around if the given klass is in fact an array (either obj or prim).
4098 // If obj_array/non_array==false/true:
4099 // Branch around if the given klass is not an array klass of any kind.
4100 // If obj_array/non_array==true/true:
4101 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4102 // If obj_array/non_array==true/false:
4103 // Branch around if the kls is an oop array (Object[] or subtype)
4104 //
4105 // Like generate_guard, adds a new path onto the region.
4106 jint layout_con = 0;
4107 Node* layout_val = get_layout_helper(kls, layout_con);
4108 if (layout_val == nullptr) {
4109 bool query = (obj_array
4110 ? Klass::layout_helper_is_objArray(layout_con)
4111 : Klass::layout_helper_is_array(layout_con));
4112 if (query == not_array) {
4113 return nullptr; // never a branch
4114 } else { // always a branch
4115 Node* always_branch = control();
4116 if (region != nullptr)
4117 region->add_req(always_branch);
4118 set_control(top());
4119 return always_branch;
4120 }
4121 }
4122 // Now test the correct condition.
4123 jint nval = (obj_array
4124 ? (jint)(Klass::_lh_array_tag_type_value
4125 << Klass::_lh_array_tag_shift)
4126 : Klass::_lh_neutral_value);
4127 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4128 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4129 // invert the test if we are looking for a non-array
4130 if (not_array) btest = BoolTest(btest).negate();
4131 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4132 return generate_fair_guard(bol, region);
4133 }
4134
4135
4136 //-----------------------inline_native_newArray--------------------------
4137 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4138 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4139 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4140 Node* mirror;
4141 Node* count_val;
4142 if (uninitialized) {
4143 null_check_receiver();
4144 mirror = argument(1);
4145 count_val = argument(2);
4146 } else {
4147 mirror = argument(0);
4148 count_val = argument(1);
4149 }
4150
4151 mirror = null_check(mirror);
4152 // If mirror or obj is dead, only null-path is taken.
4153 if (stopped()) return true;
4154
4155 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4156 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4157 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4158 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4159 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4160
4161 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4162 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4163 result_reg, _slow_path);
4164 Node* normal_ctl = control();
4165 Node* no_array_ctl = result_reg->in(_slow_path);
4166
4167 // Generate code for the slow case. We make a call to newArray().
4168 set_control(no_array_ctl);
4169 if (!stopped()) {
4170 // Either the input type is void.class, or else the
4171 // array klass has not yet been cached. Either the
4172 // ensuing call will throw an exception, or else it
4173 // will cache the array klass for next time.
4174 PreserveJVMState pjvms(this);
4175 CallJavaNode* slow_call = nullptr;
4176 if (uninitialized) {
4177 // Generate optimized virtual call (holder class 'Unsafe' is final)
4178 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4179 } else {
4180 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4181 }
4182 Node* slow_result = set_results_for_java_call(slow_call);
4183 // this->control() comes from set_results_for_java_call
4184 result_reg->set_req(_slow_path, control());
4185 result_val->set_req(_slow_path, slow_result);
4186 result_io ->set_req(_slow_path, i_o());
4187 result_mem->set_req(_slow_path, reset_memory());
4188 }
4189
4190 set_control(normal_ctl);
4191 if (!stopped()) {
4192 // Normal case: The array type has been cached in the java.lang.Class.
4193 // The following call works fine even if the array type is polymorphic.
4194 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4195 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4196 result_reg->init_req(_normal_path, control());
4197 result_val->init_req(_normal_path, obj);
4198 result_io ->init_req(_normal_path, i_o());
4199 result_mem->init_req(_normal_path, reset_memory());
4200
4201 if (uninitialized) {
4202 // Mark the allocation so that zeroing is skipped
4203 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4204 alloc->maybe_set_complete(&_gvn);
4205 }
4206 }
4207
4208 // Return the combined state.
4209 set_i_o( _gvn.transform(result_io) );
4210 set_all_memory( _gvn.transform(result_mem));
4211
4212 C->set_has_split_ifs(true); // Has chance for split-if optimization
4213 set_result(result_reg, result_val);
4214 return true;
4215 }
4216
4217 //----------------------inline_native_getLength--------------------------
4218 // public static native int java.lang.reflect.Array.getLength(Object array);
4219 bool LibraryCallKit::inline_native_getLength() {
4220 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4221
4222 Node* array = null_check(argument(0));
4223 // If array is dead, only null-path is taken.
4224 if (stopped()) return true;
4225
4226 // Deoptimize if it is a non-array.
4227 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4228
4229 if (non_array != nullptr) {
4230 PreserveJVMState pjvms(this);
4231 set_control(non_array);
4232 uncommon_trap(Deoptimization::Reason_intrinsic,
4233 Deoptimization::Action_maybe_recompile);
4234 }
4235
4236 // If control is dead, only non-array-path is taken.
4237 if (stopped()) return true;
4238
4239 // The works fine even if the array type is polymorphic.
4240 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4241 Node* result = load_array_length(array);
4242
4243 C->set_has_split_ifs(true); // Has chance for split-if optimization
4244 set_result(result);
4245 return true;
4246 }
4247
4248 //------------------------inline_array_copyOf----------------------------
4249 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4250 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4251 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4252 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4253
4254 // Get the arguments.
4255 Node* original = argument(0);
4256 Node* start = is_copyOfRange? argument(1): intcon(0);
4257 Node* end = is_copyOfRange? argument(2): argument(1);
4258 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4259
4260 Node* newcopy = nullptr;
4261
4262 // Set the original stack and the reexecute bit for the interpreter to reexecute
4263 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4264 { PreserveReexecuteState preexecs(this);
4265 jvms()->set_should_reexecute(true);
4266
4267 array_type_mirror = null_check(array_type_mirror);
4268 original = null_check(original);
4269
4270 // Check if a null path was taken unconditionally.
4271 if (stopped()) return true;
4272
4273 Node* orig_length = load_array_length(original);
4274
4275 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4276 klass_node = null_check(klass_node);
4277
4278 RegionNode* bailout = new RegionNode(1);
4279 record_for_igvn(bailout);
4280
4281 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4282 // Bail out if that is so.
4283 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4284 if (not_objArray != nullptr) {
4285 // Improve the klass node's type from the new optimistic assumption:
4286 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4287 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4288 Node* cast = new CastPPNode(klass_node, akls);
4289 cast->init_req(0, control());
4290 klass_node = _gvn.transform(cast);
4291 }
4292
4293 // Bail out if either start or end is negative.
4294 generate_negative_guard(start, bailout, &start);
4295 generate_negative_guard(end, bailout, &end);
4296
4297 Node* length = end;
4298 if (_gvn.type(start) != TypeInt::ZERO) {
4299 length = _gvn.transform(new SubINode(end, start));
4300 }
4301
4302 // Bail out if length is negative.
4303 // Without this the new_array would throw
4304 // NegativeArraySizeException but IllegalArgumentException is what
4305 // should be thrown
4306 generate_negative_guard(length, bailout, &length);
4307
4308 if (bailout->req() > 1) {
4309 PreserveJVMState pjvms(this);
4310 set_control(_gvn.transform(bailout));
4311 uncommon_trap(Deoptimization::Reason_intrinsic,
4312 Deoptimization::Action_maybe_recompile);
4313 }
4314
4315 if (!stopped()) {
4316 // How many elements will we copy from the original?
4317 // The answer is MinI(orig_length - start, length).
4318 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4319 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4320
4321 // Generate a direct call to the right arraycopy function(s).
4322 // We know the copy is disjoint but we might not know if the
4323 // oop stores need checking.
4324 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4325 // This will fail a store-check if x contains any non-nulls.
4326
4327 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4328 // loads/stores but it is legal only if we're sure the
4329 // Arrays.copyOf would succeed. So we need all input arguments
4330 // to the copyOf to be validated, including that the copy to the
4331 // new array won't trigger an ArrayStoreException. That subtype
4332 // check can be optimized if we know something on the type of
4333 // the input array from type speculation.
4334 if (_gvn.type(klass_node)->singleton()) {
4335 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4336 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4337
4338 int test = C->static_subtype_check(superk, subk);
4339 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4340 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4341 if (t_original->speculative_type() != nullptr) {
4342 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4343 }
4344 }
4345 }
4346
4347 bool validated = false;
4348 // Reason_class_check rather than Reason_intrinsic because we
4349 // want to intrinsify even if this traps.
4350 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4351 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4352
4353 if (not_subtype_ctrl != top()) {
4354 PreserveJVMState pjvms(this);
4355 set_control(not_subtype_ctrl);
4356 uncommon_trap(Deoptimization::Reason_class_check,
4357 Deoptimization::Action_make_not_entrant);
4358 assert(stopped(), "Should be stopped");
4359 }
4360 validated = true;
4361 }
4362
4363 if (!stopped()) {
4364 newcopy = new_array(klass_node, length, 0); // no arguments to push
4365
4366 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4367 load_object_klass(original), klass_node);
4368 if (!is_copyOfRange) {
4369 ac->set_copyof(validated);
4370 } else {
4371 ac->set_copyofrange(validated);
4372 }
4373 Node* n = _gvn.transform(ac);
4374 if (n == ac) {
4375 ac->connect_outputs(this);
4376 } else {
4377 assert(validated, "shouldn't transform if all arguments not validated");
4378 set_all_memory(n);
4379 }
4380 }
4381 }
4382 } // original reexecute is set back here
4383
4384 C->set_has_split_ifs(true); // Has chance for split-if optimization
4385 if (!stopped()) {
4386 set_result(newcopy);
4387 }
4388 return true;
4389 }
4390
4391
4392 //----------------------generate_virtual_guard---------------------------
4393 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4394 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4395 RegionNode* slow_region) {
4396 ciMethod* method = callee();
4397 int vtable_index = method->vtable_index();
4398 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4399 "bad index %d", vtable_index);
4400 // Get the Method* out of the appropriate vtable entry.
4401 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4402 vtable_index*vtableEntry::size_in_bytes() +
4403 in_bytes(vtableEntry::method_offset());
4404 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4405 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4406
4407 // Compare the target method with the expected method (e.g., Object.hashCode).
4408 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4409
4410 Node* native_call = makecon(native_call_addr);
4411 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4412 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4413
4414 return generate_slow_guard(test_native, slow_region);
4415 }
4416
4417 //-----------------------generate_method_call----------------------------
4418 // Use generate_method_call to make a slow-call to the real
4419 // method if the fast path fails. An alternative would be to
4420 // use a stub like OptoRuntime::slow_arraycopy_Java.
4421 // This only works for expanding the current library call,
4422 // not another intrinsic. (E.g., don't use this for making an
4423 // arraycopy call inside of the copyOf intrinsic.)
4424 CallJavaNode*
4425 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4426 // When compiling the intrinsic method itself, do not use this technique.
4427 guarantee(callee() != C->method(), "cannot make slow-call to self");
4428
4429 ciMethod* method = callee();
4430 // ensure the JVMS we have will be correct for this call
4431 guarantee(method_id == method->intrinsic_id(), "must match");
4432
4433 const TypeFunc* tf = TypeFunc::make(method);
4434 if (res_not_null) {
4435 assert(tf->return_type() == T_OBJECT, "");
4436 const TypeTuple* range = tf->range();
4437 const Type** fields = TypeTuple::fields(range->cnt());
4438 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4439 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4440 tf = TypeFunc::make(tf->domain(), new_range);
4441 }
4442 CallJavaNode* slow_call;
4443 if (is_static) {
4444 assert(!is_virtual, "");
4445 slow_call = new CallStaticJavaNode(C, tf,
4446 SharedRuntime::get_resolve_static_call_stub(), method);
4447 } else if (is_virtual) {
4448 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4449 int vtable_index = Method::invalid_vtable_index;
4450 if (UseInlineCaches) {
4451 // Suppress the vtable call
4452 } else {
4453 // hashCode and clone are not a miranda methods,
4454 // so the vtable index is fixed.
4455 // No need to use the linkResolver to get it.
4456 vtable_index = method->vtable_index();
4457 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4458 "bad index %d", vtable_index);
4459 }
4460 slow_call = new CallDynamicJavaNode(tf,
4461 SharedRuntime::get_resolve_virtual_call_stub(),
4462 method, vtable_index);
4463 } else { // neither virtual nor static: opt_virtual
4464 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4465 slow_call = new CallStaticJavaNode(C, tf,
4466 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4467 slow_call->set_optimized_virtual(true);
4468 }
4469 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4470 // To be able to issue a direct call (optimized virtual or virtual)
4471 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4472 // about the method being invoked should be attached to the call site to
4473 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4474 slow_call->set_override_symbolic_info(true);
4475 }
4476 set_arguments_for_java_call(slow_call);
4477 set_edges_for_java_call(slow_call);
4478 return slow_call;
4479 }
4480
4481
4482 /**
4483 * Build special case code for calls to hashCode on an object. This call may
4484 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4485 * slightly different code.
4486 */
4487 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4488 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4489 assert(!(is_virtual && is_static), "either virtual, special, or static");
4490
4491 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4492
4493 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4494 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4495 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4496 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4497 Node* obj = nullptr;
4498 if (!is_static) {
4499 // Check for hashing null object
4500 obj = null_check_receiver();
4501 if (stopped()) return true; // unconditionally null
4502 result_reg->init_req(_null_path, top());
4503 result_val->init_req(_null_path, top());
4504 } else {
4505 // Do a null check, and return zero if null.
4506 // System.identityHashCode(null) == 0
4507 obj = argument(0);
4508 Node* null_ctl = top();
4509 obj = null_check_oop(obj, &null_ctl);
4510 result_reg->init_req(_null_path, null_ctl);
4511 result_val->init_req(_null_path, _gvn.intcon(0));
4512 }
4513
4514 // Unconditionally null? Then return right away.
4515 if (stopped()) {
4516 set_control( result_reg->in(_null_path));
4517 if (!stopped())
4518 set_result(result_val->in(_null_path));
4519 return true;
4520 }
4521
4522 // We only go to the fast case code if we pass a number of guards. The
4523 // paths which do not pass are accumulated in the slow_region.
4524 RegionNode* slow_region = new RegionNode(1);
4525 record_for_igvn(slow_region);
4526
4527 // If this is a virtual call, we generate a funny guard. We pull out
4528 // the vtable entry corresponding to hashCode() from the target object.
4529 // If the target method which we are calling happens to be the native
4530 // Object hashCode() method, we pass the guard. We do not need this
4531 // guard for non-virtual calls -- the caller is known to be the native
4532 // Object hashCode().
4533 if (is_virtual) {
4534 // After null check, get the object's klass.
4535 Node* obj_klass = load_object_klass(obj);
4536 generate_virtual_guard(obj_klass, slow_region);
4537 }
4538
4539 // Get the header out of the object, use LoadMarkNode when available
4540 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4541 // The control of the load must be null. Otherwise, the load can move before
4542 // the null check after castPP removal.
4543 Node* no_ctrl = nullptr;
4544 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4545
4546 // Test the header to see if it is unlocked.
4547 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4548 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4549 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4550 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4551 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4552
4553 generate_slow_guard(test_unlocked, slow_region);
4554
4555 // Get the hash value and check to see that it has been properly assigned.
4556 // We depend on hash_mask being at most 32 bits and avoid the use of
4557 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4558 // vm: see markWord.hpp.
4559 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4560 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4561 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4562 // This hack lets the hash bits live anywhere in the mark object now, as long
4563 // as the shift drops the relevant bits into the low 32 bits. Note that
4564 // Java spec says that HashCode is an int so there's no point in capturing
4565 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4566 hshifted_header = ConvX2I(hshifted_header);
4567 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4568
4569 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4570 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4571 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4572
4573 generate_slow_guard(test_assigned, slow_region);
4574
4575 Node* init_mem = reset_memory();
4576 // fill in the rest of the null path:
4577 result_io ->init_req(_null_path, i_o());
4578 result_mem->init_req(_null_path, init_mem);
4579
4580 result_val->init_req(_fast_path, hash_val);
4581 result_reg->init_req(_fast_path, control());
4582 result_io ->init_req(_fast_path, i_o());
4583 result_mem->init_req(_fast_path, init_mem);
4584
4585 // Generate code for the slow case. We make a call to hashCode().
4586 set_control(_gvn.transform(slow_region));
4587 if (!stopped()) {
4588 // No need for PreserveJVMState, because we're using up the present state.
4589 set_all_memory(init_mem);
4590 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4591 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4592 Node* slow_result = set_results_for_java_call(slow_call);
4593 // this->control() comes from set_results_for_java_call
4594 result_reg->init_req(_slow_path, control());
4595 result_val->init_req(_slow_path, slow_result);
4596 result_io ->set_req(_slow_path, i_o());
4597 result_mem ->set_req(_slow_path, reset_memory());
4598 }
4599
4600 // Return the combined state.
4601 set_i_o( _gvn.transform(result_io) );
4602 set_all_memory( _gvn.transform(result_mem));
4603
4604 set_result(result_reg, result_val);
4605 return true;
4606 }
4607
4608 //---------------------------inline_native_getClass----------------------------
4609 // public final native Class<?> java.lang.Object.getClass();
4610 //
4611 // Build special case code for calls to getClass on an object.
4612 bool LibraryCallKit::inline_native_getClass() {
4613 Node* obj = null_check_receiver();
4614 if (stopped()) return true;
4615 set_result(load_mirror_from_klass(load_object_klass(obj)));
4616 return true;
4617 }
4618
4619 //-----------------inline_native_Reflection_getCallerClass---------------------
4620 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4621 //
4622 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4623 //
4624 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4625 // in that it must skip particular security frames and checks for
4626 // caller sensitive methods.
4627 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4628 #ifndef PRODUCT
4629 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4630 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4631 }
4632 #endif
4633
4634 if (!jvms()->has_method()) {
4635 #ifndef PRODUCT
4636 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4637 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4638 }
4639 #endif
4640 return false;
4641 }
4642
4643 // Walk back up the JVM state to find the caller at the required
4644 // depth.
4645 JVMState* caller_jvms = jvms();
4646
4647 // Cf. JVM_GetCallerClass
4648 // NOTE: Start the loop at depth 1 because the current JVM state does
4649 // not include the Reflection.getCallerClass() frame.
4650 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4651 ciMethod* m = caller_jvms->method();
4652 switch (n) {
4653 case 0:
4654 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4655 break;
4656 case 1:
4657 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4658 if (!m->caller_sensitive()) {
4659 #ifndef PRODUCT
4660 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4661 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4662 }
4663 #endif
4664 return false; // bail-out; let JVM_GetCallerClass do the work
4665 }
4666 break;
4667 default:
4668 if (!m->is_ignored_by_security_stack_walk()) {
4669 // We have reached the desired frame; return the holder class.
4670 // Acquire method holder as java.lang.Class and push as constant.
4671 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4672 ciInstance* caller_mirror = caller_klass->java_mirror();
4673 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4674
4675 #ifndef PRODUCT
4676 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4677 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());
4678 tty->print_cr(" JVM state at this point:");
4679 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4680 ciMethod* m = jvms()->of_depth(i)->method();
4681 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4682 }
4683 }
4684 #endif
4685 return true;
4686 }
4687 break;
4688 }
4689 }
4690
4691 #ifndef PRODUCT
4692 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4693 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4694 tty->print_cr(" JVM state at this point:");
4695 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4696 ciMethod* m = jvms()->of_depth(i)->method();
4697 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4698 }
4699 }
4700 #endif
4701
4702 return false; // bail-out; let JVM_GetCallerClass do the work
4703 }
4704
4705 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4706 Node* arg = argument(0);
4707 Node* result = nullptr;
4708
4709 switch (id) {
4710 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4711 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4712 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4713 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4714 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4715 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4716
4717 case vmIntrinsics::_doubleToLongBits: {
4718 // two paths (plus control) merge in a wood
4719 RegionNode *r = new RegionNode(3);
4720 Node *phi = new PhiNode(r, TypeLong::LONG);
4721
4722 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4723 // Build the boolean node
4724 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4725
4726 // Branch either way.
4727 // NaN case is less traveled, which makes all the difference.
4728 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4729 Node *opt_isnan = _gvn.transform(ifisnan);
4730 assert( opt_isnan->is_If(), "Expect an IfNode");
4731 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4732 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4733
4734 set_control(iftrue);
4735
4736 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4737 Node *slow_result = longcon(nan_bits); // return NaN
4738 phi->init_req(1, _gvn.transform( slow_result ));
4739 r->init_req(1, iftrue);
4740
4741 // Else fall through
4742 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4743 set_control(iffalse);
4744
4745 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4746 r->init_req(2, iffalse);
4747
4748 // Post merge
4749 set_control(_gvn.transform(r));
4750 record_for_igvn(r);
4751
4752 C->set_has_split_ifs(true); // Has chance for split-if optimization
4753 result = phi;
4754 assert(result->bottom_type()->isa_long(), "must be");
4755 break;
4756 }
4757
4758 case vmIntrinsics::_floatToIntBits: {
4759 // two paths (plus control) merge in a wood
4760 RegionNode *r = new RegionNode(3);
4761 Node *phi = new PhiNode(r, TypeInt::INT);
4762
4763 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4764 // Build the boolean node
4765 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4766
4767 // Branch either way.
4768 // NaN case is less traveled, which makes all the difference.
4769 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4770 Node *opt_isnan = _gvn.transform(ifisnan);
4771 assert( opt_isnan->is_If(), "Expect an IfNode");
4772 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4773 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4774
4775 set_control(iftrue);
4776
4777 static const jint nan_bits = 0x7fc00000;
4778 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4779 phi->init_req(1, _gvn.transform( slow_result ));
4780 r->init_req(1, iftrue);
4781
4782 // Else fall through
4783 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4784 set_control(iffalse);
4785
4786 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4787 r->init_req(2, iffalse);
4788
4789 // Post merge
4790 set_control(_gvn.transform(r));
4791 record_for_igvn(r);
4792
4793 C->set_has_split_ifs(true); // Has chance for split-if optimization
4794 result = phi;
4795 assert(result->bottom_type()->isa_int(), "must be");
4796 break;
4797 }
4798
4799 default:
4800 fatal_unexpected_iid(id);
4801 break;
4802 }
4803 set_result(_gvn.transform(result));
4804 return true;
4805 }
4806
4807 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4808 Node* arg = argument(0);
4809 Node* result = nullptr;
4810
4811 switch (id) {
4812 case vmIntrinsics::_floatIsInfinite:
4813 result = new IsInfiniteFNode(arg);
4814 break;
4815 case vmIntrinsics::_floatIsFinite:
4816 result = new IsFiniteFNode(arg);
4817 break;
4818 case vmIntrinsics::_doubleIsInfinite:
4819 result = new IsInfiniteDNode(arg);
4820 break;
4821 case vmIntrinsics::_doubleIsFinite:
4822 result = new IsFiniteDNode(arg);
4823 break;
4824 default:
4825 fatal_unexpected_iid(id);
4826 break;
4827 }
4828 set_result(_gvn.transform(result));
4829 return true;
4830 }
4831
4832 //----------------------inline_unsafe_copyMemory-------------------------
4833 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4834
4835 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4836 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4837 const Type* base_t = gvn.type(base);
4838
4839 bool in_native = (base_t == TypePtr::NULL_PTR);
4840 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4841 bool is_mixed = !in_heap && !in_native;
4842
4843 if (is_mixed) {
4844 return true; // mixed accesses can touch both on-heap and off-heap memory
4845 }
4846 if (in_heap) {
4847 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
4848 if (!is_prim_array) {
4849 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4850 // there's not enough type information available to determine proper memory slice for it.
4851 return true;
4852 }
4853 }
4854 return false;
4855 }
4856
4857 bool LibraryCallKit::inline_unsafe_copyMemory() {
4858 if (callee()->is_static()) return false; // caller must have the capability!
4859 null_check_receiver(); // null-check receiver
4860 if (stopped()) return true;
4861
4862 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4863
4864 Node* src_base = argument(1); // type: oop
4865 Node* src_off = ConvL2X(argument(2)); // type: long
4866 Node* dst_base = argument(4); // type: oop
4867 Node* dst_off = ConvL2X(argument(5)); // type: long
4868 Node* size = ConvL2X(argument(7)); // type: long
4869
4870 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4871 "fieldOffset must be byte-scaled");
4872
4873 Node* src_addr = make_unsafe_address(src_base, src_off);
4874 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4875
4876 Node* thread = _gvn.transform(new ThreadLocalNode());
4877 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4878 BasicType doing_unsafe_access_bt = T_BYTE;
4879 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4880
4881 // update volatile field
4882 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4883
4884 int flags = RC_LEAF | RC_NO_FP;
4885
4886 const TypePtr* dst_type = TypePtr::BOTTOM;
4887
4888 // Adjust memory effects of the runtime call based on input values.
4889 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4890 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4891 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4892
4893 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4894 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4895 flags |= RC_NARROW_MEM; // narrow in memory
4896 }
4897 }
4898
4899 // Call it. Note that the length argument is not scaled.
4900 make_runtime_call(flags,
4901 OptoRuntime::fast_arraycopy_Type(),
4902 StubRoutines::unsafe_arraycopy(),
4903 "unsafe_arraycopy",
4904 dst_type,
4905 src_addr, dst_addr, size XTOP);
4906
4907 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4908
4909 return true;
4910 }
4911
4912 #undef XTOP
4913
4914 //------------------------clone_coping-----------------------------------
4915 // Helper function for inline_native_clone.
4916 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4917 assert(obj_size != nullptr, "");
4918 Node* raw_obj = alloc_obj->in(1);
4919 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4920
4921 AllocateNode* alloc = nullptr;
4922 if (ReduceBulkZeroing) {
4923 // We will be completely responsible for initializing this object -
4924 // mark Initialize node as complete.
4925 alloc = AllocateNode::Ideal_allocation(alloc_obj);
4926 // The object was just allocated - there should be no any stores!
4927 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
4928 // Mark as complete_with_arraycopy so that on AllocateNode
4929 // expansion, we know this AllocateNode is initialized by an array
4930 // copy and a StoreStore barrier exists after the array copy.
4931 alloc->initialization()->set_complete_with_arraycopy();
4932 }
4933
4934 Node* size = _gvn.transform(obj_size);
4935 access_clone(obj, alloc_obj, size, is_array);
4936
4937 // Do not let reads from the cloned object float above the arraycopy.
4938 if (alloc != nullptr) {
4939 // Do not let stores that initialize this object be reordered with
4940 // a subsequent store that would make this object accessible by
4941 // other threads.
4942 // Record what AllocateNode this StoreStore protects so that
4943 // escape analysis can go from the MemBarStoreStoreNode to the
4944 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4945 // based on the escape status of the AllocateNode.
4946 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4947 } else {
4948 insert_mem_bar(Op_MemBarCPUOrder);
4949 }
4950 }
4951
4952 //------------------------inline_native_clone----------------------------
4953 // protected native Object java.lang.Object.clone();
4954 //
4955 // Here are the simple edge cases:
4956 // null receiver => normal trap
4957 // virtual and clone was overridden => slow path to out-of-line clone
4958 // not cloneable or finalizer => slow path to out-of-line Object.clone
4959 //
4960 // The general case has two steps, allocation and copying.
4961 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4962 //
4963 // Copying also has two cases, oop arrays and everything else.
4964 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4965 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4966 //
4967 // These steps fold up nicely if and when the cloned object's klass
4968 // can be sharply typed as an object array, a type array, or an instance.
4969 //
4970 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4971 PhiNode* result_val;
4972
4973 // Set the reexecute bit for the interpreter to reexecute
4974 // the bytecode that invokes Object.clone if deoptimization happens.
4975 { PreserveReexecuteState preexecs(this);
4976 jvms()->set_should_reexecute(true);
4977
4978 Node* obj = null_check_receiver();
4979 if (stopped()) return true;
4980
4981 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4982
4983 // If we are going to clone an instance, we need its exact type to
4984 // know the number and types of fields to convert the clone to
4985 // loads/stores. Maybe a speculative type can help us.
4986 if (!obj_type->klass_is_exact() &&
4987 obj_type->speculative_type() != nullptr &&
4988 obj_type->speculative_type()->is_instance_klass()) {
4989 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4990 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4991 !spec_ik->has_injected_fields()) {
4992 if (!obj_type->isa_instptr() ||
4993 obj_type->is_instptr()->instance_klass()->has_subklass()) {
4994 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4995 }
4996 }
4997 }
4998
4999 // Conservatively insert a memory barrier on all memory slices.
5000 // Do not let writes into the original float below the clone.
5001 insert_mem_bar(Op_MemBarCPUOrder);
5002
5003 // paths into result_reg:
5004 enum {
5005 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5006 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5007 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5008 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5009 PATH_LIMIT
5010 };
5011 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5012 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5013 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5014 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5015 record_for_igvn(result_reg);
5016
5017 Node* obj_klass = load_object_klass(obj);
5018 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5019 if (array_ctl != nullptr) {
5020 // It's an array.
5021 PreserveJVMState pjvms(this);
5022 set_control(array_ctl);
5023 Node* obj_length = load_array_length(obj);
5024 Node* obj_size = nullptr;
5025 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
5026
5027 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5028 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5029 // If it is an oop array, it requires very special treatment,
5030 // because gc barriers are required when accessing the array.
5031 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5032 if (is_obja != nullptr) {
5033 PreserveJVMState pjvms2(this);
5034 set_control(is_obja);
5035 // Generate a direct call to the right arraycopy function(s).
5036 // Clones are always tightly coupled.
5037 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5038 ac->set_clone_oop_array();
5039 Node* n = _gvn.transform(ac);
5040 assert(n == ac, "cannot disappear");
5041 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5042
5043 result_reg->init_req(_objArray_path, control());
5044 result_val->init_req(_objArray_path, alloc_obj);
5045 result_i_o ->set_req(_objArray_path, i_o());
5046 result_mem ->set_req(_objArray_path, reset_memory());
5047 }
5048 }
5049 // Otherwise, there are no barriers to worry about.
5050 // (We can dispense with card marks if we know the allocation
5051 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5052 // causes the non-eden paths to take compensating steps to
5053 // simulate a fresh allocation, so that no further
5054 // card marks are required in compiled code to initialize
5055 // the object.)
5056
5057 if (!stopped()) {
5058 copy_to_clone(obj, alloc_obj, obj_size, true);
5059
5060 // Present the results of the copy.
5061 result_reg->init_req(_array_path, control());
5062 result_val->init_req(_array_path, alloc_obj);
5063 result_i_o ->set_req(_array_path, i_o());
5064 result_mem ->set_req(_array_path, reset_memory());
5065 }
5066 }
5067
5068 // We only go to the instance fast case code if we pass a number of guards.
5069 // The paths which do not pass are accumulated in the slow_region.
5070 RegionNode* slow_region = new RegionNode(1);
5071 record_for_igvn(slow_region);
5072 if (!stopped()) {
5073 // It's an instance (we did array above). Make the slow-path tests.
5074 // If this is a virtual call, we generate a funny guard. We grab
5075 // the vtable entry corresponding to clone() from the target object.
5076 // If the target method which we are calling happens to be the
5077 // Object clone() method, we pass the guard. We do not need this
5078 // guard for non-virtual calls; the caller is known to be the native
5079 // Object clone().
5080 if (is_virtual) {
5081 generate_virtual_guard(obj_klass, slow_region);
5082 }
5083
5084 // The object must be easily cloneable and must not have a finalizer.
5085 // Both of these conditions may be checked in a single test.
5086 // We could optimize the test further, but we don't care.
5087 generate_access_flags_guard(obj_klass,
5088 // Test both conditions:
5089 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5090 // Must be cloneable but not finalizer:
5091 JVM_ACC_IS_CLONEABLE_FAST,
5092 slow_region);
5093 }
5094
5095 if (!stopped()) {
5096 // It's an instance, and it passed the slow-path tests.
5097 PreserveJVMState pjvms(this);
5098 Node* obj_size = nullptr;
5099 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5100 // is reexecuted if deoptimization occurs and there could be problems when merging
5101 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5102 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5103
5104 copy_to_clone(obj, alloc_obj, obj_size, false);
5105
5106 // Present the results of the slow call.
5107 result_reg->init_req(_instance_path, control());
5108 result_val->init_req(_instance_path, alloc_obj);
5109 result_i_o ->set_req(_instance_path, i_o());
5110 result_mem ->set_req(_instance_path, reset_memory());
5111 }
5112
5113 // Generate code for the slow case. We make a call to clone().
5114 set_control(_gvn.transform(slow_region));
5115 if (!stopped()) {
5116 PreserveJVMState pjvms(this);
5117 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5118 // We need to deoptimize on exception (see comment above)
5119 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5120 // this->control() comes from set_results_for_java_call
5121 result_reg->init_req(_slow_path, control());
5122 result_val->init_req(_slow_path, slow_result);
5123 result_i_o ->set_req(_slow_path, i_o());
5124 result_mem ->set_req(_slow_path, reset_memory());
5125 }
5126
5127 // Return the combined state.
5128 set_control( _gvn.transform(result_reg));
5129 set_i_o( _gvn.transform(result_i_o));
5130 set_all_memory( _gvn.transform(result_mem));
5131 } // original reexecute is set back here
5132
5133 set_result(_gvn.transform(result_val));
5134 return true;
5135 }
5136
5137 // If we have a tightly coupled allocation, the arraycopy may take care
5138 // of the array initialization. If one of the guards we insert between
5139 // the allocation and the arraycopy causes a deoptimization, an
5140 // uninitialized array will escape the compiled method. To prevent that
5141 // we set the JVM state for uncommon traps between the allocation and
5142 // the arraycopy to the state before the allocation so, in case of
5143 // deoptimization, we'll reexecute the allocation and the
5144 // initialization.
5145 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5146 if (alloc != nullptr) {
5147 ciMethod* trap_method = alloc->jvms()->method();
5148 int trap_bci = alloc->jvms()->bci();
5149
5150 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5151 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5152 // Make sure there's no store between the allocation and the
5153 // arraycopy otherwise visible side effects could be rexecuted
5154 // in case of deoptimization and cause incorrect execution.
5155 bool no_interfering_store = true;
5156 Node* mem = alloc->in(TypeFunc::Memory);
5157 if (mem->is_MergeMem()) {
5158 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5159 Node* n = mms.memory();
5160 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5161 assert(n->is_Store(), "what else?");
5162 no_interfering_store = false;
5163 break;
5164 }
5165 }
5166 } else {
5167 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5168 Node* n = mms.memory();
5169 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5170 assert(n->is_Store(), "what else?");
5171 no_interfering_store = false;
5172 break;
5173 }
5174 }
5175 }
5176
5177 if (no_interfering_store) {
5178 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5179
5180 JVMState* saved_jvms = jvms();
5181 saved_reexecute_sp = _reexecute_sp;
5182
5183 set_jvms(sfpt->jvms());
5184 _reexecute_sp = jvms()->sp();
5185
5186 return saved_jvms;
5187 }
5188 }
5189 }
5190 return nullptr;
5191 }
5192
5193 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5194 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5195 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5196 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5197 uint size = alloc->req();
5198 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5199 old_jvms->set_map(sfpt);
5200 for (uint i = 0; i < size; i++) {
5201 sfpt->init_req(i, alloc->in(i));
5202 }
5203 // re-push array length for deoptimization
5204 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5205 old_jvms->set_sp(old_jvms->sp()+1);
5206 old_jvms->set_monoff(old_jvms->monoff()+1);
5207 old_jvms->set_scloff(old_jvms->scloff()+1);
5208 old_jvms->set_endoff(old_jvms->endoff()+1);
5209 old_jvms->set_should_reexecute(true);
5210
5211 sfpt->set_i_o(map()->i_o());
5212 sfpt->set_memory(map()->memory());
5213 sfpt->set_control(map()->control());
5214 return sfpt;
5215 }
5216
5217 // In case of a deoptimization, we restart execution at the
5218 // allocation, allocating a new array. We would leave an uninitialized
5219 // array in the heap that GCs wouldn't expect. Move the allocation
5220 // after the traps so we don't allocate the array if we
5221 // deoptimize. This is possible because tightly_coupled_allocation()
5222 // guarantees there's no observer of the allocated array at this point
5223 // and the control flow is simple enough.
5224 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5225 int saved_reexecute_sp, uint new_idx) {
5226 if (saved_jvms_before_guards != nullptr && !stopped()) {
5227 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5228
5229 assert(alloc != nullptr, "only with a tightly coupled allocation");
5230 // restore JVM state to the state at the arraycopy
5231 saved_jvms_before_guards->map()->set_control(map()->control());
5232 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5233 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5234 // If we've improved the types of some nodes (null check) while
5235 // emitting the guards, propagate them to the current state
5236 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5237 set_jvms(saved_jvms_before_guards);
5238 _reexecute_sp = saved_reexecute_sp;
5239
5240 // Remove the allocation from above the guards
5241 CallProjections callprojs;
5242 alloc->extract_projections(&callprojs, true);
5243 InitializeNode* init = alloc->initialization();
5244 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5245 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5246 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5247
5248 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5249 // the allocation (i.e. is only valid if the allocation succeeds):
5250 // 1) replace CastIINode with AllocateArrayNode's length here
5251 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5252 //
5253 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5254 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5255 Node* init_control = init->proj_out(TypeFunc::Control);
5256 Node* alloc_length = alloc->Ideal_length();
5257 #ifdef ASSERT
5258 Node* prev_cast = nullptr;
5259 #endif
5260 for (uint i = 0; i < init_control->outcnt(); i++) {
5261 Node* init_out = init_control->raw_out(i);
5262 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5263 #ifdef ASSERT
5264 if (prev_cast == nullptr) {
5265 prev_cast = init_out;
5266 } else {
5267 if (prev_cast->cmp(*init_out) == false) {
5268 prev_cast->dump();
5269 init_out->dump();
5270 assert(false, "not equal CastIINode");
5271 }
5272 }
5273 #endif
5274 C->gvn_replace_by(init_out, alloc_length);
5275 }
5276 }
5277 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5278
5279 // move the allocation here (after the guards)
5280 _gvn.hash_delete(alloc);
5281 alloc->set_req(TypeFunc::Control, control());
5282 alloc->set_req(TypeFunc::I_O, i_o());
5283 Node *mem = reset_memory();
5284 set_all_memory(mem);
5285 alloc->set_req(TypeFunc::Memory, mem);
5286 set_control(init->proj_out_or_null(TypeFunc::Control));
5287 set_i_o(callprojs.fallthrough_ioproj);
5288
5289 // Update memory as done in GraphKit::set_output_for_allocation()
5290 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5291 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5292 if (ary_type->isa_aryptr() && length_type != nullptr) {
5293 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5294 }
5295 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5296 int elemidx = C->get_alias_index(telemref);
5297 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5298 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5299
5300 Node* allocx = _gvn.transform(alloc);
5301 assert(allocx == alloc, "where has the allocation gone?");
5302 assert(dest->is_CheckCastPP(), "not an allocation result?");
5303
5304 _gvn.hash_delete(dest);
5305 dest->set_req(0, control());
5306 Node* destx = _gvn.transform(dest);
5307 assert(destx == dest, "where has the allocation result gone?");
5308
5309 array_ideal_length(alloc, ary_type, true);
5310 }
5311 }
5312
5313 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5314 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5315 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5316 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5317 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5318 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5319 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5320 JVMState* saved_jvms_before_guards) {
5321 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5322 // There is at least one unrelated uncommon trap which needs to be replaced.
5323 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5324
5325 JVMState* saved_jvms = jvms();
5326 const int saved_reexecute_sp = _reexecute_sp;
5327 set_jvms(sfpt->jvms());
5328 _reexecute_sp = jvms()->sp();
5329
5330 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5331
5332 // Restore state
5333 set_jvms(saved_jvms);
5334 _reexecute_sp = saved_reexecute_sp;
5335 }
5336 }
5337
5338 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5339 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5340 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5341 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5342 while (if_proj->is_IfProj()) {
5343 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5344 if (uncommon_trap != nullptr) {
5345 create_new_uncommon_trap(uncommon_trap);
5346 }
5347 assert(if_proj->in(0)->is_If(), "must be If");
5348 if_proj = if_proj->in(0)->in(0);
5349 }
5350 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5351 "must have reached control projection of init node");
5352 }
5353
5354 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5355 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5356 assert(trap_request != 0, "no valid UCT trap request");
5357 PreserveJVMState pjvms(this);
5358 set_control(uncommon_trap_call->in(0));
5359 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5360 Deoptimization::trap_request_action(trap_request));
5361 assert(stopped(), "Should be stopped");
5362 _gvn.hash_delete(uncommon_trap_call);
5363 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5364 }
5365
5366 //------------------------------inline_arraycopy-----------------------
5367 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5368 // Object dest, int destPos,
5369 // int length);
5370 bool LibraryCallKit::inline_arraycopy() {
5371 // Get the arguments.
5372 Node* src = argument(0); // type: oop
5373 Node* src_offset = argument(1); // type: int
5374 Node* dest = argument(2); // type: oop
5375 Node* dest_offset = argument(3); // type: int
5376 Node* length = argument(4); // type: int
5377
5378 uint new_idx = C->unique();
5379
5380 // Check for allocation before we add nodes that would confuse
5381 // tightly_coupled_allocation()
5382 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5383
5384 int saved_reexecute_sp = -1;
5385 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5386 // See arraycopy_restore_alloc_state() comment
5387 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5388 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5389 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5390 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5391
5392 // The following tests must be performed
5393 // (1) src and dest are arrays.
5394 // (2) src and dest arrays must have elements of the same BasicType
5395 // (3) src and dest must not be null.
5396 // (4) src_offset must not be negative.
5397 // (5) dest_offset must not be negative.
5398 // (6) length must not be negative.
5399 // (7) src_offset + length must not exceed length of src.
5400 // (8) dest_offset + length must not exceed length of dest.
5401 // (9) each element of an oop array must be assignable
5402
5403 // (3) src and dest must not be null.
5404 // always do this here because we need the JVM state for uncommon traps
5405 Node* null_ctl = top();
5406 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5407 assert(null_ctl->is_top(), "no null control here");
5408 dest = null_check(dest, T_ARRAY);
5409
5410 if (!can_emit_guards) {
5411 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5412 // guards but the arraycopy node could still take advantage of a
5413 // tightly allocated allocation. tightly_coupled_allocation() is
5414 // called again to make sure it takes the null check above into
5415 // account: the null check is mandatory and if it caused an
5416 // uncommon trap to be emitted then the allocation can't be
5417 // considered tightly coupled in this context.
5418 alloc = tightly_coupled_allocation(dest);
5419 }
5420
5421 bool validated = false;
5422
5423 const Type* src_type = _gvn.type(src);
5424 const Type* dest_type = _gvn.type(dest);
5425 const TypeAryPtr* top_src = src_type->isa_aryptr();
5426 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5427
5428 // Do we have the type of src?
5429 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5430 // Do we have the type of dest?
5431 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5432 // Is the type for src from speculation?
5433 bool src_spec = false;
5434 // Is the type for dest from speculation?
5435 bool dest_spec = false;
5436
5437 if ((!has_src || !has_dest) && can_emit_guards) {
5438 // We don't have sufficient type information, let's see if
5439 // speculative types can help. We need to have types for both src
5440 // and dest so that it pays off.
5441
5442 // Do we already have or could we have type information for src
5443 bool could_have_src = has_src;
5444 // Do we already have or could we have type information for dest
5445 bool could_have_dest = has_dest;
5446
5447 ciKlass* src_k = nullptr;
5448 if (!has_src) {
5449 src_k = src_type->speculative_type_not_null();
5450 if (src_k != nullptr && src_k->is_array_klass()) {
5451 could_have_src = true;
5452 }
5453 }
5454
5455 ciKlass* dest_k = nullptr;
5456 if (!has_dest) {
5457 dest_k = dest_type->speculative_type_not_null();
5458 if (dest_k != nullptr && dest_k->is_array_klass()) {
5459 could_have_dest = true;
5460 }
5461 }
5462
5463 if (could_have_src && could_have_dest) {
5464 // This is going to pay off so emit the required guards
5465 if (!has_src) {
5466 src = maybe_cast_profiled_obj(src, src_k, true);
5467 src_type = _gvn.type(src);
5468 top_src = src_type->isa_aryptr();
5469 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5470 src_spec = true;
5471 }
5472 if (!has_dest) {
5473 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5474 dest_type = _gvn.type(dest);
5475 top_dest = dest_type->isa_aryptr();
5476 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5477 dest_spec = true;
5478 }
5479 }
5480 }
5481
5482 if (has_src && has_dest && can_emit_guards) {
5483 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5484 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5485 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5486 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5487
5488 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5489 // If both arrays are object arrays then having the exact types
5490 // for both will remove the need for a subtype check at runtime
5491 // before the call and may make it possible to pick a faster copy
5492 // routine (without a subtype check on every element)
5493 // Do we have the exact type of src?
5494 bool could_have_src = src_spec;
5495 // Do we have the exact type of dest?
5496 bool could_have_dest = dest_spec;
5497 ciKlass* src_k = nullptr;
5498 ciKlass* dest_k = nullptr;
5499 if (!src_spec) {
5500 src_k = src_type->speculative_type_not_null();
5501 if (src_k != nullptr && src_k->is_array_klass()) {
5502 could_have_src = true;
5503 }
5504 }
5505 if (!dest_spec) {
5506 dest_k = dest_type->speculative_type_not_null();
5507 if (dest_k != nullptr && dest_k->is_array_klass()) {
5508 could_have_dest = true;
5509 }
5510 }
5511 if (could_have_src && could_have_dest) {
5512 // If we can have both exact types, emit the missing guards
5513 if (could_have_src && !src_spec) {
5514 src = maybe_cast_profiled_obj(src, src_k, true);
5515 }
5516 if (could_have_dest && !dest_spec) {
5517 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5518 }
5519 }
5520 }
5521 }
5522
5523 ciMethod* trap_method = method();
5524 int trap_bci = bci();
5525 if (saved_jvms_before_guards != nullptr) {
5526 trap_method = alloc->jvms()->method();
5527 trap_bci = alloc->jvms()->bci();
5528 }
5529
5530 bool negative_length_guard_generated = false;
5531
5532 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5533 can_emit_guards &&
5534 !src->is_top() && !dest->is_top()) {
5535 // validate arguments: enables transformation the ArrayCopyNode
5536 validated = true;
5537
5538 RegionNode* slow_region = new RegionNode(1);
5539 record_for_igvn(slow_region);
5540
5541 // (1) src and dest are arrays.
5542 generate_non_array_guard(load_object_klass(src), slow_region);
5543 generate_non_array_guard(load_object_klass(dest), slow_region);
5544
5545 // (2) src and dest arrays must have elements of the same BasicType
5546 // done at macro expansion or at Ideal transformation time
5547
5548 // (4) src_offset must not be negative.
5549 generate_negative_guard(src_offset, slow_region);
5550
5551 // (5) dest_offset must not be negative.
5552 generate_negative_guard(dest_offset, slow_region);
5553
5554 // (7) src_offset + length must not exceed length of src.
5555 generate_limit_guard(src_offset, length,
5556 load_array_length(src),
5557 slow_region);
5558
5559 // (8) dest_offset + length must not exceed length of dest.
5560 generate_limit_guard(dest_offset, length,
5561 load_array_length(dest),
5562 slow_region);
5563
5564 // (6) length must not be negative.
5565 // This is also checked in generate_arraycopy() during macro expansion, but
5566 // we also have to check it here for the case where the ArrayCopyNode will
5567 // be eliminated by Escape Analysis.
5568 if (EliminateAllocations) {
5569 generate_negative_guard(length, slow_region);
5570 negative_length_guard_generated = true;
5571 }
5572
5573 // (9) each element of an oop array must be assignable
5574 Node* dest_klass = load_object_klass(dest);
5575 if (src != dest) {
5576 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5577
5578 if (not_subtype_ctrl != top()) {
5579 PreserveJVMState pjvms(this);
5580 set_control(not_subtype_ctrl);
5581 uncommon_trap(Deoptimization::Reason_intrinsic,
5582 Deoptimization::Action_make_not_entrant);
5583 assert(stopped(), "Should be stopped");
5584 }
5585 }
5586 {
5587 PreserveJVMState pjvms(this);
5588 set_control(_gvn.transform(slow_region));
5589 uncommon_trap(Deoptimization::Reason_intrinsic,
5590 Deoptimization::Action_make_not_entrant);
5591 assert(stopped(), "Should be stopped");
5592 }
5593
5594 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5595 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5596 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5597 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5598 }
5599
5600 if (stopped()) {
5601 return true;
5602 }
5603
5604 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5605 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5606 // so the compiler has a chance to eliminate them: during macro expansion,
5607 // we have to set their control (CastPP nodes are eliminated).
5608 load_object_klass(src), load_object_klass(dest),
5609 load_array_length(src), load_array_length(dest));
5610
5611 ac->set_arraycopy(validated);
5612
5613 Node* n = _gvn.transform(ac);
5614 if (n == ac) {
5615 ac->connect_outputs(this);
5616 } else {
5617 assert(validated, "shouldn't transform if all arguments not validated");
5618 set_all_memory(n);
5619 }
5620 clear_upper_avx();
5621
5622
5623 return true;
5624 }
5625
5626
5627 // Helper function which determines if an arraycopy immediately follows
5628 // an allocation, with no intervening tests or other escapes for the object.
5629 AllocateArrayNode*
5630 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5631 if (stopped()) return nullptr; // no fast path
5632 if (!C->do_aliasing()) return nullptr; // no MergeMems around
5633
5634 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
5635 if (alloc == nullptr) return nullptr;
5636
5637 Node* rawmem = memory(Compile::AliasIdxRaw);
5638 // Is the allocation's memory state untouched?
5639 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5640 // Bail out if there have been raw-memory effects since the allocation.
5641 // (Example: There might have been a call or safepoint.)
5642 return nullptr;
5643 }
5644 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5645 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5646 return nullptr;
5647 }
5648
5649 // There must be no unexpected observers of this allocation.
5650 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5651 Node* obs = ptr->fast_out(i);
5652 if (obs != this->map()) {
5653 return nullptr;
5654 }
5655 }
5656
5657 // This arraycopy must unconditionally follow the allocation of the ptr.
5658 Node* alloc_ctl = ptr->in(0);
5659 Node* ctl = control();
5660 while (ctl != alloc_ctl) {
5661 // There may be guards which feed into the slow_region.
5662 // Any other control flow means that we might not get a chance
5663 // to finish initializing the allocated object.
5664 // Various low-level checks bottom out in uncommon traps. These
5665 // are considered safe since we've already checked above that
5666 // there is no unexpected observer of this allocation.
5667 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
5668 assert(ctl->in(0)->is_If(), "must be If");
5669 ctl = ctl->in(0)->in(0);
5670 } else {
5671 return nullptr;
5672 }
5673 }
5674
5675 // If we get this far, we have an allocation which immediately
5676 // precedes the arraycopy, and we can take over zeroing the new object.
5677 // The arraycopy will finish the initialization, and provide
5678 // a new control state to which we will anchor the destination pointer.
5679
5680 return alloc;
5681 }
5682
5683 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
5684 if (node->is_IfProj()) {
5685 Node* other_proj = node->as_IfProj()->other_if_proj();
5686 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
5687 Node* obs = other_proj->fast_out(j);
5688 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
5689 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5690 return obs->as_CallStaticJava();
5691 }
5692 }
5693 }
5694 return nullptr;
5695 }
5696
5697 //-------------inline_encodeISOArray-----------------------------------
5698 // encode char[] to byte[] in ISO_8859_1 or ASCII
5699 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5700 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5701 // no receiver since it is static method
5702 Node *src = argument(0);
5703 Node *src_offset = argument(1);
5704 Node *dst = argument(2);
5705 Node *dst_offset = argument(3);
5706 Node *length = argument(4);
5707
5708 src = must_be_not_null(src, true);
5709 dst = must_be_not_null(dst, true);
5710
5711 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
5712 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
5713 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
5714 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
5715 // failed array check
5716 return false;
5717 }
5718
5719 // Figure out the size and type of the elements we will be copying.
5720 BasicType src_elem = src_type->elem()->array_element_basic_type();
5721 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
5722 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5723 return false;
5724 }
5725
5726 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5727 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5728 // 'src_start' points to src array + scaled offset
5729 // 'dst_start' points to dst array + scaled offset
5730
5731 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5732 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5733 enc = _gvn.transform(enc);
5734 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5735 set_memory(res_mem, mtype);
5736 set_result(enc);
5737 clear_upper_avx();
5738
5739 return true;
5740 }
5741
5742 //-------------inline_multiplyToLen-----------------------------------
5743 bool LibraryCallKit::inline_multiplyToLen() {
5744 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5745
5746 address stubAddr = StubRoutines::multiplyToLen();
5747 if (stubAddr == nullptr) {
5748 return false; // Intrinsic's stub is not implemented on this platform
5749 }
5750 const char* stubName = "multiplyToLen";
5751
5752 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5753
5754 // no receiver because it is a static method
5755 Node* x = argument(0);
5756 Node* xlen = argument(1);
5757 Node* y = argument(2);
5758 Node* ylen = argument(3);
5759 Node* z = argument(4);
5760
5761 x = must_be_not_null(x, true);
5762 y = must_be_not_null(y, true);
5763
5764 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5765 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
5766 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5767 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
5768 // failed array check
5769 return false;
5770 }
5771
5772 BasicType x_elem = x_type->elem()->array_element_basic_type();
5773 BasicType y_elem = y_type->elem()->array_element_basic_type();
5774 if (x_elem != T_INT || y_elem != T_INT) {
5775 return false;
5776 }
5777
5778 // Set the original stack and the reexecute bit for the interpreter to reexecute
5779 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5780 // on the return from z array allocation in runtime.
5781 { PreserveReexecuteState preexecs(this);
5782 jvms()->set_should_reexecute(true);
5783
5784 Node* x_start = array_element_address(x, intcon(0), x_elem);
5785 Node* y_start = array_element_address(y, intcon(0), y_elem);
5786 // 'x_start' points to x array + scaled xlen
5787 // 'y_start' points to y array + scaled ylen
5788
5789 // Allocate the result array
5790 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5791 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5792 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5793
5794 IdealKit ideal(this);
5795
5796 #define __ ideal.
5797 Node* one = __ ConI(1);
5798 Node* zero = __ ConI(0);
5799 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5800 __ set(need_alloc, zero);
5801 __ set(z_alloc, z);
5802 __ if_then(z, BoolTest::eq, null()); {
5803 __ increment (need_alloc, one);
5804 } __ else_(); {
5805 // Update graphKit memory and control from IdealKit.
5806 sync_kit(ideal);
5807 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5808 cast->init_req(0, control());
5809 _gvn.set_type(cast, cast->bottom_type());
5810 C->record_for_igvn(cast);
5811
5812 Node* zlen_arg = load_array_length(cast);
5813 // Update IdealKit memory and control from graphKit.
5814 __ sync_kit(this);
5815 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5816 __ increment (need_alloc, one);
5817 } __ end_if();
5818 } __ end_if();
5819
5820 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5821 // Update graphKit memory and control from IdealKit.
5822 sync_kit(ideal);
5823 Node * narr = new_array(klass_node, zlen, 1);
5824 // Update IdealKit memory and control from graphKit.
5825 __ sync_kit(this);
5826 __ set(z_alloc, narr);
5827 } __ end_if();
5828
5829 sync_kit(ideal);
5830 z = __ value(z_alloc);
5831 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5832 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5833 // Final sync IdealKit and GraphKit.
5834 final_sync(ideal);
5835 #undef __
5836
5837 Node* z_start = array_element_address(z, intcon(0), T_INT);
5838
5839 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5840 OptoRuntime::multiplyToLen_Type(),
5841 stubAddr, stubName, TypePtr::BOTTOM,
5842 x_start, xlen, y_start, ylen, z_start, zlen);
5843 } // original reexecute is set back here
5844
5845 C->set_has_split_ifs(true); // Has chance for split-if optimization
5846 set_result(z);
5847 return true;
5848 }
5849
5850 //-------------inline_squareToLen------------------------------------
5851 bool LibraryCallKit::inline_squareToLen() {
5852 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5853
5854 address stubAddr = StubRoutines::squareToLen();
5855 if (stubAddr == nullptr) {
5856 return false; // Intrinsic's stub is not implemented on this platform
5857 }
5858 const char* stubName = "squareToLen";
5859
5860 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5861
5862 Node* x = argument(0);
5863 Node* len = argument(1);
5864 Node* z = argument(2);
5865 Node* zlen = argument(3);
5866
5867 x = must_be_not_null(x, true);
5868 z = must_be_not_null(z, true);
5869
5870 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5871 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
5872 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5873 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
5874 // failed array check
5875 return false;
5876 }
5877
5878 BasicType x_elem = x_type->elem()->array_element_basic_type();
5879 BasicType z_elem = z_type->elem()->array_element_basic_type();
5880 if (x_elem != T_INT || z_elem != T_INT) {
5881 return false;
5882 }
5883
5884
5885 Node* x_start = array_element_address(x, intcon(0), x_elem);
5886 Node* z_start = array_element_address(z, intcon(0), z_elem);
5887
5888 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5889 OptoRuntime::squareToLen_Type(),
5890 stubAddr, stubName, TypePtr::BOTTOM,
5891 x_start, len, z_start, zlen);
5892
5893 set_result(z);
5894 return true;
5895 }
5896
5897 //-------------inline_mulAdd------------------------------------------
5898 bool LibraryCallKit::inline_mulAdd() {
5899 assert(UseMulAddIntrinsic, "not implemented on this platform");
5900
5901 address stubAddr = StubRoutines::mulAdd();
5902 if (stubAddr == nullptr) {
5903 return false; // Intrinsic's stub is not implemented on this platform
5904 }
5905 const char* stubName = "mulAdd";
5906
5907 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5908
5909 Node* out = argument(0);
5910 Node* in = argument(1);
5911 Node* offset = argument(2);
5912 Node* len = argument(3);
5913 Node* k = argument(4);
5914
5915 in = must_be_not_null(in, true);
5916 out = must_be_not_null(out, true);
5917
5918 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
5919 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
5920 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
5921 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
5922 // failed array check
5923 return false;
5924 }
5925
5926 BasicType out_elem = out_type->elem()->array_element_basic_type();
5927 BasicType in_elem = in_type->elem()->array_element_basic_type();
5928 if (out_elem != T_INT || in_elem != T_INT) {
5929 return false;
5930 }
5931
5932 Node* outlen = load_array_length(out);
5933 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5934 Node* out_start = array_element_address(out, intcon(0), out_elem);
5935 Node* in_start = array_element_address(in, intcon(0), in_elem);
5936
5937 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5938 OptoRuntime::mulAdd_Type(),
5939 stubAddr, stubName, TypePtr::BOTTOM,
5940 out_start,in_start, new_offset, len, k);
5941 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5942 set_result(result);
5943 return true;
5944 }
5945
5946 //-------------inline_montgomeryMultiply-----------------------------------
5947 bool LibraryCallKit::inline_montgomeryMultiply() {
5948 address stubAddr = StubRoutines::montgomeryMultiply();
5949 if (stubAddr == nullptr) {
5950 return false; // Intrinsic's stub is not implemented on this platform
5951 }
5952
5953 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5954 const char* stubName = "montgomery_multiply";
5955
5956 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5957
5958 Node* a = argument(0);
5959 Node* b = argument(1);
5960 Node* n = argument(2);
5961 Node* len = argument(3);
5962 Node* inv = argument(4);
5963 Node* m = argument(6);
5964
5965 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
5966 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
5967 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
5968 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
5969 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
5970 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
5971 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
5972 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
5973 // failed array check
5974 return false;
5975 }
5976
5977 BasicType a_elem = a_type->elem()->array_element_basic_type();
5978 BasicType b_elem = b_type->elem()->array_element_basic_type();
5979 BasicType n_elem = n_type->elem()->array_element_basic_type();
5980 BasicType m_elem = m_type->elem()->array_element_basic_type();
5981 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5982 return false;
5983 }
5984
5985 // Make the call
5986 {
5987 Node* a_start = array_element_address(a, intcon(0), a_elem);
5988 Node* b_start = array_element_address(b, intcon(0), b_elem);
5989 Node* n_start = array_element_address(n, intcon(0), n_elem);
5990 Node* m_start = array_element_address(m, intcon(0), m_elem);
5991
5992 Node* call = make_runtime_call(RC_LEAF,
5993 OptoRuntime::montgomeryMultiply_Type(),
5994 stubAddr, stubName, TypePtr::BOTTOM,
5995 a_start, b_start, n_start, len, inv, top(),
5996 m_start);
5997 set_result(m);
5998 }
5999
6000 return true;
6001 }
6002
6003 bool LibraryCallKit::inline_montgomerySquare() {
6004 address stubAddr = StubRoutines::montgomerySquare();
6005 if (stubAddr == nullptr) {
6006 return false; // Intrinsic's stub is not implemented on this platform
6007 }
6008
6009 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6010 const char* stubName = "montgomery_square";
6011
6012 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6013
6014 Node* a = argument(0);
6015 Node* n = argument(1);
6016 Node* len = argument(2);
6017 Node* inv = argument(3);
6018 Node* m = argument(5);
6019
6020 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6021 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6022 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6023 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6024 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6025 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6026 // failed array check
6027 return false;
6028 }
6029
6030 BasicType a_elem = a_type->elem()->array_element_basic_type();
6031 BasicType n_elem = n_type->elem()->array_element_basic_type();
6032 BasicType m_elem = m_type->elem()->array_element_basic_type();
6033 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6034 return false;
6035 }
6036
6037 // Make the call
6038 {
6039 Node* a_start = array_element_address(a, intcon(0), a_elem);
6040 Node* n_start = array_element_address(n, intcon(0), n_elem);
6041 Node* m_start = array_element_address(m, intcon(0), m_elem);
6042
6043 Node* call = make_runtime_call(RC_LEAF,
6044 OptoRuntime::montgomerySquare_Type(),
6045 stubAddr, stubName, TypePtr::BOTTOM,
6046 a_start, n_start, len, inv, top(),
6047 m_start);
6048 set_result(m);
6049 }
6050
6051 return true;
6052 }
6053
6054 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6055 address stubAddr = nullptr;
6056 const char* stubName = nullptr;
6057
6058 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6059 if (stubAddr == nullptr) {
6060 return false; // Intrinsic's stub is not implemented on this platform
6061 }
6062
6063 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6064
6065 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6066
6067 Node* newArr = argument(0);
6068 Node* oldArr = argument(1);
6069 Node* newIdx = argument(2);
6070 Node* shiftCount = argument(3);
6071 Node* numIter = argument(4);
6072
6073 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6074 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6075 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6076 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6077 return false;
6078 }
6079
6080 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6081 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6082 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6083 return false;
6084 }
6085
6086 // Make the call
6087 {
6088 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6089 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6090
6091 Node* call = make_runtime_call(RC_LEAF,
6092 OptoRuntime::bigIntegerShift_Type(),
6093 stubAddr,
6094 stubName,
6095 TypePtr::BOTTOM,
6096 newArr_start,
6097 oldArr_start,
6098 newIdx,
6099 shiftCount,
6100 numIter);
6101 }
6102
6103 return true;
6104 }
6105
6106 //-------------inline_vectorizedMismatch------------------------------
6107 bool LibraryCallKit::inline_vectorizedMismatch() {
6108 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6109
6110 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6111 Node* obja = argument(0); // Object
6112 Node* aoffset = argument(1); // long
6113 Node* objb = argument(3); // Object
6114 Node* boffset = argument(4); // long
6115 Node* length = argument(6); // int
6116 Node* scale = argument(7); // int
6117
6118 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6119 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6120 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6121 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6122 scale == top()) {
6123 return false; // failed input validation
6124 }
6125
6126 Node* obja_adr = make_unsafe_address(obja, aoffset);
6127 Node* objb_adr = make_unsafe_address(objb, boffset);
6128
6129 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6130 //
6131 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6132 // if (length <= inline_limit) {
6133 // inline_path:
6134 // vmask = VectorMaskGen length
6135 // vload1 = LoadVectorMasked obja, vmask
6136 // vload2 = LoadVectorMasked objb, vmask
6137 // result1 = VectorCmpMasked vload1, vload2, vmask
6138 // } else {
6139 // call_stub_path:
6140 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6141 // }
6142 // exit_block:
6143 // return Phi(result1, result2);
6144 //
6145 enum { inline_path = 1, // input is small enough to process it all at once
6146 stub_path = 2, // input is too large; call into the VM
6147 PATH_LIMIT = 3
6148 };
6149
6150 Node* exit_block = new RegionNode(PATH_LIMIT);
6151 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6152 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6153
6154 Node* call_stub_path = control();
6155
6156 BasicType elem_bt = T_ILLEGAL;
6157
6158 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6159 if (scale_t->is_con()) {
6160 switch (scale_t->get_con()) {
6161 case 0: elem_bt = T_BYTE; break;
6162 case 1: elem_bt = T_SHORT; break;
6163 case 2: elem_bt = T_INT; break;
6164 case 3: elem_bt = T_LONG; break;
6165
6166 default: elem_bt = T_ILLEGAL; break; // not supported
6167 }
6168 }
6169
6170 int inline_limit = 0;
6171 bool do_partial_inline = false;
6172
6173 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6174 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6175 do_partial_inline = inline_limit >= 16;
6176 }
6177
6178 if (do_partial_inline) {
6179 assert(elem_bt != T_ILLEGAL, "sanity");
6180
6181 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6182 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6183 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6184
6185 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6186 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6187 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6188
6189 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6190
6191 if (!stopped()) {
6192 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6193
6194 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6195 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6196 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6197 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6198
6199 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6200 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6201 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6202 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6203
6204 exit_block->init_req(inline_path, control());
6205 memory_phi->init_req(inline_path, map()->memory());
6206 result_phi->init_req(inline_path, result);
6207
6208 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6209 clear_upper_avx();
6210 }
6211 }
6212 }
6213
6214 if (call_stub_path != nullptr) {
6215 set_control(call_stub_path);
6216
6217 Node* call = make_runtime_call(RC_LEAF,
6218 OptoRuntime::vectorizedMismatch_Type(),
6219 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6220 obja_adr, objb_adr, length, scale);
6221
6222 exit_block->init_req(stub_path, control());
6223 memory_phi->init_req(stub_path, map()->memory());
6224 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6225 }
6226
6227 exit_block = _gvn.transform(exit_block);
6228 memory_phi = _gvn.transform(memory_phi);
6229 result_phi = _gvn.transform(result_phi);
6230
6231 set_control(exit_block);
6232 set_all_memory(memory_phi);
6233 set_result(result_phi);
6234
6235 return true;
6236 }
6237
6238 //------------------------------inline_vectorizedHashcode----------------------------
6239 bool LibraryCallKit::inline_vectorizedHashCode() {
6240 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6241
6242 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6243 Node* array = argument(0);
6244 Node* offset = argument(1);
6245 Node* length = argument(2);
6246 Node* initialValue = argument(3);
6247 Node* basic_type = argument(4);
6248
6249 if (basic_type == top()) {
6250 return false; // failed input validation
6251 }
6252
6253 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6254 if (!basic_type_t->is_con()) {
6255 return false; // Only intrinsify if mode argument is constant
6256 }
6257
6258 array = must_be_not_null(array, true);
6259
6260 BasicType bt = (BasicType)basic_type_t->get_con();
6261
6262 // Resolve address of first element
6263 Node* array_start = array_element_address(array, offset, bt);
6264
6265 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6266 array_start, length, initialValue, basic_type)));
6267 clear_upper_avx();
6268
6269 return true;
6270 }
6271
6272 /**
6273 * Calculate CRC32 for byte.
6274 * int java.util.zip.CRC32.update(int crc, int b)
6275 */
6276 bool LibraryCallKit::inline_updateCRC32() {
6277 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6278 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6279 // no receiver since it is static method
6280 Node* crc = argument(0); // type: int
6281 Node* b = argument(1); // type: int
6282
6283 /*
6284 * int c = ~ crc;
6285 * b = timesXtoThe32[(b ^ c) & 0xFF];
6286 * b = b ^ (c >>> 8);
6287 * crc = ~b;
6288 */
6289
6290 Node* M1 = intcon(-1);
6291 crc = _gvn.transform(new XorINode(crc, M1));
6292 Node* result = _gvn.transform(new XorINode(crc, b));
6293 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6294
6295 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6296 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6297 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6298 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6299
6300 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6301 result = _gvn.transform(new XorINode(crc, result));
6302 result = _gvn.transform(new XorINode(result, M1));
6303 set_result(result);
6304 return true;
6305 }
6306
6307 /**
6308 * Calculate CRC32 for byte[] array.
6309 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6310 */
6311 bool LibraryCallKit::inline_updateBytesCRC32() {
6312 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6313 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6314 // no receiver since it is static method
6315 Node* crc = argument(0); // type: int
6316 Node* src = argument(1); // type: oop
6317 Node* offset = argument(2); // type: int
6318 Node* length = argument(3); // type: int
6319
6320 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6321 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6322 // failed array check
6323 return false;
6324 }
6325
6326 // Figure out the size and type of the elements we will be copying.
6327 BasicType src_elem = src_type->elem()->array_element_basic_type();
6328 if (src_elem != T_BYTE) {
6329 return false;
6330 }
6331
6332 // 'src_start' points to src array + scaled offset
6333 src = must_be_not_null(src, true);
6334 Node* src_start = array_element_address(src, offset, src_elem);
6335
6336 // We assume that range check is done by caller.
6337 // TODO: generate range check (offset+length < src.length) in debug VM.
6338
6339 // Call the stub.
6340 address stubAddr = StubRoutines::updateBytesCRC32();
6341 const char *stubName = "updateBytesCRC32";
6342
6343 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6344 stubAddr, stubName, TypePtr::BOTTOM,
6345 crc, src_start, length);
6346 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6347 set_result(result);
6348 return true;
6349 }
6350
6351 /**
6352 * Calculate CRC32 for ByteBuffer.
6353 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6354 */
6355 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6356 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6357 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6358 // no receiver since it is static method
6359 Node* crc = argument(0); // type: int
6360 Node* src = argument(1); // type: long
6361 Node* offset = argument(3); // type: int
6362 Node* length = argument(4); // type: int
6363
6364 src = ConvL2X(src); // adjust Java long to machine word
6365 Node* base = _gvn.transform(new CastX2PNode(src));
6366 offset = ConvI2X(offset);
6367
6368 // 'src_start' points to src array + scaled offset
6369 Node* src_start = basic_plus_adr(top(), base, offset);
6370
6371 // Call the stub.
6372 address stubAddr = StubRoutines::updateBytesCRC32();
6373 const char *stubName = "updateBytesCRC32";
6374
6375 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6376 stubAddr, stubName, TypePtr::BOTTOM,
6377 crc, src_start, length);
6378 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6379 set_result(result);
6380 return true;
6381 }
6382
6383 //------------------------------get_table_from_crc32c_class-----------------------
6384 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6385 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6386 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6387
6388 return table;
6389 }
6390
6391 //------------------------------inline_updateBytesCRC32C-----------------------
6392 //
6393 // Calculate CRC32C for byte[] array.
6394 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6395 //
6396 bool LibraryCallKit::inline_updateBytesCRC32C() {
6397 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6398 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6399 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6400 // no receiver since it is a static method
6401 Node* crc = argument(0); // type: int
6402 Node* src = argument(1); // type: oop
6403 Node* offset = argument(2); // type: int
6404 Node* end = argument(3); // type: int
6405
6406 Node* length = _gvn.transform(new SubINode(end, offset));
6407
6408 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6409 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6410 // failed array check
6411 return false;
6412 }
6413
6414 // Figure out the size and type of the elements we will be copying.
6415 BasicType src_elem = src_type->elem()->array_element_basic_type();
6416 if (src_elem != T_BYTE) {
6417 return false;
6418 }
6419
6420 // 'src_start' points to src array + scaled offset
6421 src = must_be_not_null(src, true);
6422 Node* src_start = array_element_address(src, offset, src_elem);
6423
6424 // static final int[] byteTable in class CRC32C
6425 Node* table = get_table_from_crc32c_class(callee()->holder());
6426 table = must_be_not_null(table, true);
6427 Node* table_start = array_element_address(table, intcon(0), T_INT);
6428
6429 // We assume that range check is done by caller.
6430 // TODO: generate range check (offset+length < src.length) in debug VM.
6431
6432 // Call the stub.
6433 address stubAddr = StubRoutines::updateBytesCRC32C();
6434 const char *stubName = "updateBytesCRC32C";
6435
6436 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6437 stubAddr, stubName, TypePtr::BOTTOM,
6438 crc, src_start, length, table_start);
6439 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6440 set_result(result);
6441 return true;
6442 }
6443
6444 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6445 //
6446 // Calculate CRC32C for DirectByteBuffer.
6447 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6448 //
6449 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6450 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6451 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6452 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6453 // no receiver since it is a static method
6454 Node* crc = argument(0); // type: int
6455 Node* src = argument(1); // type: long
6456 Node* offset = argument(3); // type: int
6457 Node* end = argument(4); // type: int
6458
6459 Node* length = _gvn.transform(new SubINode(end, offset));
6460
6461 src = ConvL2X(src); // adjust Java long to machine word
6462 Node* base = _gvn.transform(new CastX2PNode(src));
6463 offset = ConvI2X(offset);
6464
6465 // 'src_start' points to src array + scaled offset
6466 Node* src_start = basic_plus_adr(top(), base, offset);
6467
6468 // static final int[] byteTable in class CRC32C
6469 Node* table = get_table_from_crc32c_class(callee()->holder());
6470 table = must_be_not_null(table, true);
6471 Node* table_start = array_element_address(table, intcon(0), T_INT);
6472
6473 // Call the stub.
6474 address stubAddr = StubRoutines::updateBytesCRC32C();
6475 const char *stubName = "updateBytesCRC32C";
6476
6477 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6478 stubAddr, stubName, TypePtr::BOTTOM,
6479 crc, src_start, length, table_start);
6480 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6481 set_result(result);
6482 return true;
6483 }
6484
6485 //------------------------------inline_updateBytesAdler32----------------------
6486 //
6487 // Calculate Adler32 checksum for byte[] array.
6488 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6489 //
6490 bool LibraryCallKit::inline_updateBytesAdler32() {
6491 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6492 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6493 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6494 // no receiver since it is static method
6495 Node* crc = argument(0); // type: int
6496 Node* src = argument(1); // type: oop
6497 Node* offset = argument(2); // type: int
6498 Node* length = argument(3); // type: int
6499
6500 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6501 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6502 // failed array check
6503 return false;
6504 }
6505
6506 // Figure out the size and type of the elements we will be copying.
6507 BasicType src_elem = src_type->elem()->array_element_basic_type();
6508 if (src_elem != T_BYTE) {
6509 return false;
6510 }
6511
6512 // 'src_start' points to src array + scaled offset
6513 Node* src_start = array_element_address(src, offset, src_elem);
6514
6515 // We assume that range check is done by caller.
6516 // TODO: generate range check (offset+length < src.length) in debug VM.
6517
6518 // Call the stub.
6519 address stubAddr = StubRoutines::updateBytesAdler32();
6520 const char *stubName = "updateBytesAdler32";
6521
6522 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6523 stubAddr, stubName, TypePtr::BOTTOM,
6524 crc, src_start, length);
6525 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6526 set_result(result);
6527 return true;
6528 }
6529
6530 //------------------------------inline_updateByteBufferAdler32---------------
6531 //
6532 // Calculate Adler32 checksum for DirectByteBuffer.
6533 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6534 //
6535 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6536 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6537 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6538 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6539 // no receiver since it is static method
6540 Node* crc = argument(0); // type: int
6541 Node* src = argument(1); // type: long
6542 Node* offset = argument(3); // type: int
6543 Node* length = argument(4); // type: int
6544
6545 src = ConvL2X(src); // adjust Java long to machine word
6546 Node* base = _gvn.transform(new CastX2PNode(src));
6547 offset = ConvI2X(offset);
6548
6549 // 'src_start' points to src array + scaled offset
6550 Node* src_start = basic_plus_adr(top(), base, offset);
6551
6552 // Call the stub.
6553 address stubAddr = StubRoutines::updateBytesAdler32();
6554 const char *stubName = "updateBytesAdler32";
6555
6556 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6557 stubAddr, stubName, TypePtr::BOTTOM,
6558 crc, src_start, length);
6559
6560 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6561 set_result(result);
6562 return true;
6563 }
6564
6565 //----------------------------inline_reference_get----------------------------
6566 // public T java.lang.ref.Reference.get();
6567 bool LibraryCallKit::inline_reference_get() {
6568 const int referent_offset = java_lang_ref_Reference::referent_offset();
6569
6570 // Get the argument:
6571 Node* reference_obj = null_check_receiver();
6572 if (stopped()) return true;
6573
6574 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6575 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6576 decorators, /*is_static*/ false, nullptr);
6577 if (result == nullptr) return false;
6578
6579 // Add memory barrier to prevent commoning reads from this field
6580 // across safepoint since GC can change its value.
6581 insert_mem_bar(Op_MemBarCPUOrder);
6582
6583 set_result(result);
6584 return true;
6585 }
6586
6587 //----------------------------inline_reference_refersTo0----------------------------
6588 // bool java.lang.ref.Reference.refersTo0();
6589 // bool java.lang.ref.PhantomReference.refersTo0();
6590 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6591 // Get arguments:
6592 Node* reference_obj = null_check_receiver();
6593 Node* other_obj = argument(1);
6594 if (stopped()) return true;
6595
6596 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6597 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6598 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6599 decorators, /*is_static*/ false, nullptr);
6600 if (referent == nullptr) return false;
6601
6602 // Add memory barrier to prevent commoning reads from this field
6603 // across safepoint since GC can change its value.
6604 insert_mem_bar(Op_MemBarCPUOrder);
6605
6606 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6607 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6608 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6609
6610 RegionNode* region = new RegionNode(3);
6611 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6612
6613 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6614 region->init_req(1, if_true);
6615 phi->init_req(1, intcon(1));
6616
6617 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6618 region->init_req(2, if_false);
6619 phi->init_req(2, intcon(0));
6620
6621 set_control(_gvn.transform(region));
6622 record_for_igvn(region);
6623 set_result(_gvn.transform(phi));
6624 return true;
6625 }
6626
6627
6628 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6629 DecoratorSet decorators, bool is_static,
6630 ciInstanceKlass* fromKls) {
6631 if (fromKls == nullptr) {
6632 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6633 assert(tinst != nullptr, "obj is null");
6634 assert(tinst->is_loaded(), "obj is not loaded");
6635 fromKls = tinst->instance_klass();
6636 } else {
6637 assert(is_static, "only for static field access");
6638 }
6639 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6640 ciSymbol::make(fieldTypeString),
6641 is_static);
6642
6643 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6644 if (field == nullptr) return (Node *) nullptr;
6645
6646 if (is_static) {
6647 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6648 fromObj = makecon(tip);
6649 }
6650
6651 // Next code copied from Parse::do_get_xxx():
6652
6653 // Compute address and memory type.
6654 int offset = field->offset_in_bytes();
6655 bool is_vol = field->is_volatile();
6656 ciType* field_klass = field->type();
6657 assert(field_klass->is_loaded(), "should be loaded");
6658 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6659 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6660 BasicType bt = field->layout_type();
6661
6662 // Build the resultant type of the load
6663 const Type *type;
6664 if (bt == T_OBJECT) {
6665 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6666 } else {
6667 type = Type::get_const_basic_type(bt);
6668 }
6669
6670 if (is_vol) {
6671 decorators |= MO_SEQ_CST;
6672 }
6673
6674 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6675 }
6676
6677 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6678 bool is_exact /* true */, bool is_static /* false */,
6679 ciInstanceKlass * fromKls /* nullptr */) {
6680 if (fromKls == nullptr) {
6681 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6682 assert(tinst != nullptr, "obj is null");
6683 assert(tinst->is_loaded(), "obj is not loaded");
6684 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6685 fromKls = tinst->instance_klass();
6686 }
6687 else {
6688 assert(is_static, "only for static field access");
6689 }
6690 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6691 ciSymbol::make(fieldTypeString),
6692 is_static);
6693
6694 assert(field != nullptr, "undefined field");
6695 assert(!field->is_volatile(), "not defined for volatile fields");
6696
6697 if (is_static) {
6698 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6699 fromObj = makecon(tip);
6700 }
6701
6702 // Next code copied from Parse::do_get_xxx():
6703
6704 // Compute address and memory type.
6705 int offset = field->offset_in_bytes();
6706 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6707
6708 return adr;
6709 }
6710
6711 //------------------------------inline_aescrypt_Block-----------------------
6712 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6713 address stubAddr = nullptr;
6714 const char *stubName;
6715 assert(UseAES, "need AES instruction support");
6716
6717 switch(id) {
6718 case vmIntrinsics::_aescrypt_encryptBlock:
6719 stubAddr = StubRoutines::aescrypt_encryptBlock();
6720 stubName = "aescrypt_encryptBlock";
6721 break;
6722 case vmIntrinsics::_aescrypt_decryptBlock:
6723 stubAddr = StubRoutines::aescrypt_decryptBlock();
6724 stubName = "aescrypt_decryptBlock";
6725 break;
6726 default:
6727 break;
6728 }
6729 if (stubAddr == nullptr) return false;
6730
6731 Node* aescrypt_object = argument(0);
6732 Node* src = argument(1);
6733 Node* src_offset = argument(2);
6734 Node* dest = argument(3);
6735 Node* dest_offset = argument(4);
6736
6737 src = must_be_not_null(src, true);
6738 dest = must_be_not_null(dest, true);
6739
6740 // (1) src and dest are arrays.
6741 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6742 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6743 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6744 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6745
6746 // for the quick and dirty code we will skip all the checks.
6747 // we are just trying to get the call to be generated.
6748 Node* src_start = src;
6749 Node* dest_start = dest;
6750 if (src_offset != nullptr || dest_offset != nullptr) {
6751 assert(src_offset != nullptr && dest_offset != nullptr, "");
6752 src_start = array_element_address(src, src_offset, T_BYTE);
6753 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6754 }
6755
6756 // now need to get the start of its expanded key array
6757 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6758 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6759 if (k_start == nullptr) return false;
6760
6761 // Call the stub.
6762 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6763 stubAddr, stubName, TypePtr::BOTTOM,
6764 src_start, dest_start, k_start);
6765
6766 return true;
6767 }
6768
6769 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6770 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6771 address stubAddr = nullptr;
6772 const char *stubName = nullptr;
6773
6774 assert(UseAES, "need AES instruction support");
6775
6776 switch(id) {
6777 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6778 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6779 stubName = "cipherBlockChaining_encryptAESCrypt";
6780 break;
6781 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6782 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6783 stubName = "cipherBlockChaining_decryptAESCrypt";
6784 break;
6785 default:
6786 break;
6787 }
6788 if (stubAddr == nullptr) return false;
6789
6790 Node* cipherBlockChaining_object = argument(0);
6791 Node* src = argument(1);
6792 Node* src_offset = argument(2);
6793 Node* len = argument(3);
6794 Node* dest = argument(4);
6795 Node* dest_offset = argument(5);
6796
6797 src = must_be_not_null(src, false);
6798 dest = must_be_not_null(dest, false);
6799
6800 // (1) src and dest are arrays.
6801 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6802 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6803 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6804 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6805
6806 // checks are the responsibility of the caller
6807 Node* src_start = src;
6808 Node* dest_start = dest;
6809 if (src_offset != nullptr || dest_offset != nullptr) {
6810 assert(src_offset != nullptr && dest_offset != nullptr, "");
6811 src_start = array_element_address(src, src_offset, T_BYTE);
6812 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6813 }
6814
6815 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6816 // (because of the predicated logic executed earlier).
6817 // so we cast it here safely.
6818 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6819
6820 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6821 if (embeddedCipherObj == nullptr) return false;
6822
6823 // cast it to what we know it will be at runtime
6824 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6825 assert(tinst != nullptr, "CBC obj is null");
6826 assert(tinst->is_loaded(), "CBC obj is not loaded");
6827 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6828 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6829
6830 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6831 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6832 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6833 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6834 aescrypt_object = _gvn.transform(aescrypt_object);
6835
6836 // we need to get the start of the aescrypt_object's expanded key array
6837 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6838 if (k_start == nullptr) return false;
6839
6840 // similarly, get the start address of the r vector
6841 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6842 if (objRvec == nullptr) return false;
6843 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6844
6845 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6846 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6847 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6848 stubAddr, stubName, TypePtr::BOTTOM,
6849 src_start, dest_start, k_start, r_start, len);
6850
6851 // return cipher length (int)
6852 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6853 set_result(retvalue);
6854 return true;
6855 }
6856
6857 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6858 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6859 address stubAddr = nullptr;
6860 const char *stubName = nullptr;
6861
6862 assert(UseAES, "need AES instruction support");
6863
6864 switch (id) {
6865 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6866 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6867 stubName = "electronicCodeBook_encryptAESCrypt";
6868 break;
6869 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6870 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6871 stubName = "electronicCodeBook_decryptAESCrypt";
6872 break;
6873 default:
6874 break;
6875 }
6876
6877 if (stubAddr == nullptr) return false;
6878
6879 Node* electronicCodeBook_object = argument(0);
6880 Node* src = argument(1);
6881 Node* src_offset = argument(2);
6882 Node* len = argument(3);
6883 Node* dest = argument(4);
6884 Node* dest_offset = argument(5);
6885
6886 // (1) src and dest are arrays.
6887 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6888 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6889 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6890 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6891
6892 // checks are the responsibility of the caller
6893 Node* src_start = src;
6894 Node* dest_start = dest;
6895 if (src_offset != nullptr || dest_offset != nullptr) {
6896 assert(src_offset != nullptr && dest_offset != nullptr, "");
6897 src_start = array_element_address(src, src_offset, T_BYTE);
6898 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6899 }
6900
6901 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6902 // (because of the predicated logic executed earlier).
6903 // so we cast it here safely.
6904 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6905
6906 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6907 if (embeddedCipherObj == nullptr) return false;
6908
6909 // cast it to what we know it will be at runtime
6910 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6911 assert(tinst != nullptr, "ECB obj is null");
6912 assert(tinst->is_loaded(), "ECB obj is not loaded");
6913 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6914 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6915
6916 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6917 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6918 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6919 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6920 aescrypt_object = _gvn.transform(aescrypt_object);
6921
6922 // we need to get the start of the aescrypt_object's expanded key array
6923 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6924 if (k_start == nullptr) return false;
6925
6926 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6927 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6928 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6929 stubAddr, stubName, TypePtr::BOTTOM,
6930 src_start, dest_start, k_start, len);
6931
6932 // return cipher length (int)
6933 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6934 set_result(retvalue);
6935 return true;
6936 }
6937
6938 //------------------------------inline_counterMode_AESCrypt-----------------------
6939 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6940 assert(UseAES, "need AES instruction support");
6941 if (!UseAESCTRIntrinsics) return false;
6942
6943 address stubAddr = nullptr;
6944 const char *stubName = nullptr;
6945 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6946 stubAddr = StubRoutines::counterMode_AESCrypt();
6947 stubName = "counterMode_AESCrypt";
6948 }
6949 if (stubAddr == nullptr) return false;
6950
6951 Node* counterMode_object = argument(0);
6952 Node* src = argument(1);
6953 Node* src_offset = argument(2);
6954 Node* len = argument(3);
6955 Node* dest = argument(4);
6956 Node* dest_offset = argument(5);
6957
6958 // (1) src and dest are arrays.
6959 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6960 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6961 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6962 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6963
6964 // checks are the responsibility of the caller
6965 Node* src_start = src;
6966 Node* dest_start = dest;
6967 if (src_offset != nullptr || dest_offset != nullptr) {
6968 assert(src_offset != nullptr && dest_offset != nullptr, "");
6969 src_start = array_element_address(src, src_offset, T_BYTE);
6970 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6971 }
6972
6973 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6974 // (because of the predicated logic executed earlier).
6975 // so we cast it here safely.
6976 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6977 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6978 if (embeddedCipherObj == nullptr) return false;
6979 // cast it to what we know it will be at runtime
6980 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6981 assert(tinst != nullptr, "CTR obj is null");
6982 assert(tinst->is_loaded(), "CTR obj is not loaded");
6983 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6984 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6985 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6986 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6987 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6988 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6989 aescrypt_object = _gvn.transform(aescrypt_object);
6990 // we need to get the start of the aescrypt_object's expanded key array
6991 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6992 if (k_start == nullptr) return false;
6993 // similarly, get the start address of the r vector
6994 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6995 if (obj_counter == nullptr) return false;
6996 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6997
6998 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6999 if (saved_encCounter == nullptr) return false;
7000 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7001 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7002
7003 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7004 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7005 OptoRuntime::counterMode_aescrypt_Type(),
7006 stubAddr, stubName, TypePtr::BOTTOM,
7007 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7008
7009 // return cipher length (int)
7010 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7011 set_result(retvalue);
7012 return true;
7013 }
7014
7015 //------------------------------get_key_start_from_aescrypt_object-----------------------
7016 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7017 #if defined(PPC64) || defined(S390)
7018 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7019 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7020 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7021 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7022 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7023 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7024 if (objSessionK == nullptr) {
7025 return (Node *) nullptr;
7026 }
7027 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7028 #else
7029 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7030 #endif // PPC64
7031 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7032 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7033
7034 // now have the array, need to get the start address of the K array
7035 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7036 return k_start;
7037 }
7038
7039 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7040 // Return node representing slow path of predicate check.
7041 // the pseudo code we want to emulate with this predicate is:
7042 // for encryption:
7043 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7044 // for decryption:
7045 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7046 // note cipher==plain is more conservative than the original java code but that's OK
7047 //
7048 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7049 // The receiver was checked for null already.
7050 Node* objCBC = argument(0);
7051
7052 Node* src = argument(1);
7053 Node* dest = argument(4);
7054
7055 // Load embeddedCipher field of CipherBlockChaining object.
7056 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7057
7058 // get AESCrypt klass for instanceOf check
7059 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7060 // will have same classloader as CipherBlockChaining object
7061 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7062 assert(tinst != nullptr, "CBCobj is null");
7063 assert(tinst->is_loaded(), "CBCobj is not loaded");
7064
7065 // we want to do an instanceof comparison against the AESCrypt class
7066 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7067 if (!klass_AESCrypt->is_loaded()) {
7068 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7069 Node* ctrl = control();
7070 set_control(top()); // no regular fast path
7071 return ctrl;
7072 }
7073
7074 src = must_be_not_null(src, true);
7075 dest = must_be_not_null(dest, true);
7076
7077 // Resolve oops to stable for CmpP below.
7078 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7079
7080 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7081 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7082 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7083
7084 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7085
7086 // for encryption, we are done
7087 if (!decrypting)
7088 return instof_false; // even if it is null
7089
7090 // for decryption, we need to add a further check to avoid
7091 // taking the intrinsic path when cipher and plain are the same
7092 // see the original java code for why.
7093 RegionNode* region = new RegionNode(3);
7094 region->init_req(1, instof_false);
7095
7096 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7097 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7098 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7099 region->init_req(2, src_dest_conjoint);
7100
7101 record_for_igvn(region);
7102 return _gvn.transform(region);
7103 }
7104
7105 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7106 // Return node representing slow path of predicate check.
7107 // the pseudo code we want to emulate with this predicate is:
7108 // for encryption:
7109 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7110 // for decryption:
7111 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7112 // note cipher==plain is more conservative than the original java code but that's OK
7113 //
7114 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7115 // The receiver was checked for null already.
7116 Node* objECB = argument(0);
7117
7118 // Load embeddedCipher field of ElectronicCodeBook object.
7119 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7120
7121 // get AESCrypt klass for instanceOf check
7122 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7123 // will have same classloader as ElectronicCodeBook object
7124 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7125 assert(tinst != nullptr, "ECBobj is null");
7126 assert(tinst->is_loaded(), "ECBobj is not loaded");
7127
7128 // we want to do an instanceof comparison against the AESCrypt class
7129 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7130 if (!klass_AESCrypt->is_loaded()) {
7131 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7132 Node* ctrl = control();
7133 set_control(top()); // no regular fast path
7134 return ctrl;
7135 }
7136 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7137
7138 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7139 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7140 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7141
7142 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7143
7144 // for encryption, we are done
7145 if (!decrypting)
7146 return instof_false; // even if it is null
7147
7148 // for decryption, we need to add a further check to avoid
7149 // taking the intrinsic path when cipher and plain are the same
7150 // see the original java code for why.
7151 RegionNode* region = new RegionNode(3);
7152 region->init_req(1, instof_false);
7153 Node* src = argument(1);
7154 Node* dest = argument(4);
7155 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7156 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7157 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7158 region->init_req(2, src_dest_conjoint);
7159
7160 record_for_igvn(region);
7161 return _gvn.transform(region);
7162 }
7163
7164 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7165 // Return node representing slow path of predicate check.
7166 // the pseudo code we want to emulate with this predicate is:
7167 // for encryption:
7168 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7169 // for decryption:
7170 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7171 // note cipher==plain is more conservative than the original java code but that's OK
7172 //
7173
7174 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7175 // The receiver was checked for null already.
7176 Node* objCTR = argument(0);
7177
7178 // Load embeddedCipher field of CipherBlockChaining object.
7179 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7180
7181 // get AESCrypt klass for instanceOf check
7182 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7183 // will have same classloader as CipherBlockChaining object
7184 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7185 assert(tinst != nullptr, "CTRobj is null");
7186 assert(tinst->is_loaded(), "CTRobj is not loaded");
7187
7188 // we want to do an instanceof comparison against the AESCrypt class
7189 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7190 if (!klass_AESCrypt->is_loaded()) {
7191 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7192 Node* ctrl = control();
7193 set_control(top()); // no regular fast path
7194 return ctrl;
7195 }
7196
7197 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7198 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7199 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7200 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7201 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7202
7203 return instof_false; // even if it is null
7204 }
7205
7206 //------------------------------inline_ghash_processBlocks
7207 bool LibraryCallKit::inline_ghash_processBlocks() {
7208 address stubAddr;
7209 const char *stubName;
7210 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7211
7212 stubAddr = StubRoutines::ghash_processBlocks();
7213 stubName = "ghash_processBlocks";
7214
7215 Node* data = argument(0);
7216 Node* offset = argument(1);
7217 Node* len = argument(2);
7218 Node* state = argument(3);
7219 Node* subkeyH = argument(4);
7220
7221 state = must_be_not_null(state, true);
7222 subkeyH = must_be_not_null(subkeyH, true);
7223 data = must_be_not_null(data, true);
7224
7225 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7226 assert(state_start, "state is null");
7227 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7228 assert(subkeyH_start, "subkeyH is null");
7229 Node* data_start = array_element_address(data, offset, T_BYTE);
7230 assert(data_start, "data is null");
7231
7232 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7233 OptoRuntime::ghash_processBlocks_Type(),
7234 stubAddr, stubName, TypePtr::BOTTOM,
7235 state_start, subkeyH_start, data_start, len);
7236 return true;
7237 }
7238
7239 //------------------------------inline_chacha20Block
7240 bool LibraryCallKit::inline_chacha20Block() {
7241 address stubAddr;
7242 const char *stubName;
7243 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7244
7245 stubAddr = StubRoutines::chacha20Block();
7246 stubName = "chacha20Block";
7247
7248 Node* state = argument(0);
7249 Node* result = argument(1);
7250
7251 state = must_be_not_null(state, true);
7252 result = must_be_not_null(result, true);
7253
7254 Node* state_start = array_element_address(state, intcon(0), T_INT);
7255 assert(state_start, "state is null");
7256 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7257 assert(result_start, "result is null");
7258
7259 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7260 OptoRuntime::chacha20Block_Type(),
7261 stubAddr, stubName, TypePtr::BOTTOM,
7262 state_start, result_start);
7263 // return key stream length (int)
7264 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7265 set_result(retvalue);
7266 return true;
7267 }
7268
7269 bool LibraryCallKit::inline_base64_encodeBlock() {
7270 address stubAddr;
7271 const char *stubName;
7272 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7273 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7274 stubAddr = StubRoutines::base64_encodeBlock();
7275 stubName = "encodeBlock";
7276
7277 if (!stubAddr) return false;
7278 Node* base64obj = argument(0);
7279 Node* src = argument(1);
7280 Node* offset = argument(2);
7281 Node* len = argument(3);
7282 Node* dest = argument(4);
7283 Node* dp = argument(5);
7284 Node* isURL = argument(6);
7285
7286 src = must_be_not_null(src, true);
7287 dest = must_be_not_null(dest, true);
7288
7289 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7290 assert(src_start, "source array is null");
7291 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7292 assert(dest_start, "destination array is null");
7293
7294 Node* base64 = make_runtime_call(RC_LEAF,
7295 OptoRuntime::base64_encodeBlock_Type(),
7296 stubAddr, stubName, TypePtr::BOTTOM,
7297 src_start, offset, len, dest_start, dp, isURL);
7298 return true;
7299 }
7300
7301 bool LibraryCallKit::inline_base64_decodeBlock() {
7302 address stubAddr;
7303 const char *stubName;
7304 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7305 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7306 stubAddr = StubRoutines::base64_decodeBlock();
7307 stubName = "decodeBlock";
7308
7309 if (!stubAddr) return false;
7310 Node* base64obj = argument(0);
7311 Node* src = argument(1);
7312 Node* src_offset = argument(2);
7313 Node* len = argument(3);
7314 Node* dest = argument(4);
7315 Node* dest_offset = argument(5);
7316 Node* isURL = argument(6);
7317 Node* isMIME = argument(7);
7318
7319 src = must_be_not_null(src, true);
7320 dest = must_be_not_null(dest, true);
7321
7322 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7323 assert(src_start, "source array is null");
7324 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7325 assert(dest_start, "destination array is null");
7326
7327 Node* call = make_runtime_call(RC_LEAF,
7328 OptoRuntime::base64_decodeBlock_Type(),
7329 stubAddr, stubName, TypePtr::BOTTOM,
7330 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7331 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7332 set_result(result);
7333 return true;
7334 }
7335
7336 bool LibraryCallKit::inline_poly1305_processBlocks() {
7337 address stubAddr;
7338 const char *stubName;
7339 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7340 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7341 stubAddr = StubRoutines::poly1305_processBlocks();
7342 stubName = "poly1305_processBlocks";
7343
7344 if (!stubAddr) return false;
7345 null_check_receiver(); // null-check receiver
7346 if (stopped()) return true;
7347
7348 Node* input = argument(1);
7349 Node* input_offset = argument(2);
7350 Node* len = argument(3);
7351 Node* alimbs = argument(4);
7352 Node* rlimbs = argument(5);
7353
7354 input = must_be_not_null(input, true);
7355 alimbs = must_be_not_null(alimbs, true);
7356 rlimbs = must_be_not_null(rlimbs, true);
7357
7358 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7359 assert(input_start, "input array is null");
7360 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7361 assert(acc_start, "acc array is null");
7362 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7363 assert(r_start, "r array is null");
7364
7365 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7366 OptoRuntime::poly1305_processBlocks_Type(),
7367 stubAddr, stubName, TypePtr::BOTTOM,
7368 input_start, len, acc_start, r_start);
7369 return true;
7370 }
7371
7372 //------------------------------inline_digestBase_implCompress-----------------------
7373 //
7374 // Calculate MD5 for single-block byte[] array.
7375 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7376 //
7377 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7378 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7379 //
7380 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7381 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7382 //
7383 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7384 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7385 //
7386 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7387 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7388 //
7389 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7390 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7391
7392 Node* digestBase_obj = argument(0);
7393 Node* src = argument(1); // type oop
7394 Node* ofs = argument(2); // type int
7395
7396 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7397 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7398 // failed array check
7399 return false;
7400 }
7401 // Figure out the size and type of the elements we will be copying.
7402 BasicType src_elem = src_type->elem()->array_element_basic_type();
7403 if (src_elem != T_BYTE) {
7404 return false;
7405 }
7406 // 'src_start' points to src array + offset
7407 src = must_be_not_null(src, true);
7408 Node* src_start = array_element_address(src, ofs, src_elem);
7409 Node* state = nullptr;
7410 Node* block_size = nullptr;
7411 address stubAddr;
7412 const char *stubName;
7413
7414 switch(id) {
7415 case vmIntrinsics::_md5_implCompress:
7416 assert(UseMD5Intrinsics, "need MD5 instruction support");
7417 state = get_state_from_digest_object(digestBase_obj, T_INT);
7418 stubAddr = StubRoutines::md5_implCompress();
7419 stubName = "md5_implCompress";
7420 break;
7421 case vmIntrinsics::_sha_implCompress:
7422 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7423 state = get_state_from_digest_object(digestBase_obj, T_INT);
7424 stubAddr = StubRoutines::sha1_implCompress();
7425 stubName = "sha1_implCompress";
7426 break;
7427 case vmIntrinsics::_sha2_implCompress:
7428 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7429 state = get_state_from_digest_object(digestBase_obj, T_INT);
7430 stubAddr = StubRoutines::sha256_implCompress();
7431 stubName = "sha256_implCompress";
7432 break;
7433 case vmIntrinsics::_sha5_implCompress:
7434 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7435 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7436 stubAddr = StubRoutines::sha512_implCompress();
7437 stubName = "sha512_implCompress";
7438 break;
7439 case vmIntrinsics::_sha3_implCompress:
7440 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7441 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7442 stubAddr = StubRoutines::sha3_implCompress();
7443 stubName = "sha3_implCompress";
7444 block_size = get_block_size_from_digest_object(digestBase_obj);
7445 if (block_size == nullptr) return false;
7446 break;
7447 default:
7448 fatal_unexpected_iid(id);
7449 return false;
7450 }
7451 if (state == nullptr) return false;
7452
7453 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7454 if (stubAddr == nullptr) return false;
7455
7456 // Call the stub.
7457 Node* call;
7458 if (block_size == nullptr) {
7459 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7460 stubAddr, stubName, TypePtr::BOTTOM,
7461 src_start, state);
7462 } else {
7463 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7464 stubAddr, stubName, TypePtr::BOTTOM,
7465 src_start, state, block_size);
7466 }
7467
7468 return true;
7469 }
7470
7471 //------------------------------inline_digestBase_implCompressMB-----------------------
7472 //
7473 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7474 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7475 //
7476 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7477 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7478 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7479 assert((uint)predicate < 5, "sanity");
7480 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7481
7482 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7483 Node* src = argument(1); // byte[] array
7484 Node* ofs = argument(2); // type int
7485 Node* limit = argument(3); // type int
7486
7487 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7488 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7489 // failed array check
7490 return false;
7491 }
7492 // Figure out the size and type of the elements we will be copying.
7493 BasicType src_elem = src_type->elem()->array_element_basic_type();
7494 if (src_elem != T_BYTE) {
7495 return false;
7496 }
7497 // 'src_start' points to src array + offset
7498 src = must_be_not_null(src, false);
7499 Node* src_start = array_element_address(src, ofs, src_elem);
7500
7501 const char* klass_digestBase_name = nullptr;
7502 const char* stub_name = nullptr;
7503 address stub_addr = nullptr;
7504 BasicType elem_type = T_INT;
7505
7506 switch (predicate) {
7507 case 0:
7508 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7509 klass_digestBase_name = "sun/security/provider/MD5";
7510 stub_name = "md5_implCompressMB";
7511 stub_addr = StubRoutines::md5_implCompressMB();
7512 }
7513 break;
7514 case 1:
7515 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7516 klass_digestBase_name = "sun/security/provider/SHA";
7517 stub_name = "sha1_implCompressMB";
7518 stub_addr = StubRoutines::sha1_implCompressMB();
7519 }
7520 break;
7521 case 2:
7522 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7523 klass_digestBase_name = "sun/security/provider/SHA2";
7524 stub_name = "sha256_implCompressMB";
7525 stub_addr = StubRoutines::sha256_implCompressMB();
7526 }
7527 break;
7528 case 3:
7529 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7530 klass_digestBase_name = "sun/security/provider/SHA5";
7531 stub_name = "sha512_implCompressMB";
7532 stub_addr = StubRoutines::sha512_implCompressMB();
7533 elem_type = T_LONG;
7534 }
7535 break;
7536 case 4:
7537 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7538 klass_digestBase_name = "sun/security/provider/SHA3";
7539 stub_name = "sha3_implCompressMB";
7540 stub_addr = StubRoutines::sha3_implCompressMB();
7541 elem_type = T_BYTE;
7542 }
7543 break;
7544 default:
7545 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7546 }
7547 if (klass_digestBase_name != nullptr) {
7548 assert(stub_addr != nullptr, "Stub is generated");
7549 if (stub_addr == nullptr) return false;
7550
7551 // get DigestBase klass to lookup for SHA klass
7552 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7553 assert(tinst != nullptr, "digestBase_obj is not instance???");
7554 assert(tinst->is_loaded(), "DigestBase is not loaded");
7555
7556 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7557 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7558 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7559 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7560 }
7561 return false;
7562 }
7563
7564 //------------------------------inline_digestBase_implCompressMB-----------------------
7565 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7566 BasicType elem_type, address stubAddr, const char *stubName,
7567 Node* src_start, Node* ofs, Node* limit) {
7568 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7569 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7570 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7571 digest_obj = _gvn.transform(digest_obj);
7572
7573 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7574 if (state == nullptr) return false;
7575
7576 Node* block_size = nullptr;
7577 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7578 block_size = get_block_size_from_digest_object(digest_obj);
7579 if (block_size == nullptr) return false;
7580 }
7581
7582 // Call the stub.
7583 Node* call;
7584 if (block_size == nullptr) {
7585 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7586 OptoRuntime::digestBase_implCompressMB_Type(false),
7587 stubAddr, stubName, TypePtr::BOTTOM,
7588 src_start, state, ofs, limit);
7589 } else {
7590 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7591 OptoRuntime::digestBase_implCompressMB_Type(true),
7592 stubAddr, stubName, TypePtr::BOTTOM,
7593 src_start, state, block_size, ofs, limit);
7594 }
7595
7596 // return ofs (int)
7597 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7598 set_result(result);
7599
7600 return true;
7601 }
7602
7603 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7604 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7605 assert(UseAES, "need AES instruction support");
7606 address stubAddr = nullptr;
7607 const char *stubName = nullptr;
7608 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7609 stubName = "galoisCounterMode_AESCrypt";
7610
7611 if (stubAddr == nullptr) return false;
7612
7613 Node* in = argument(0);
7614 Node* inOfs = argument(1);
7615 Node* len = argument(2);
7616 Node* ct = argument(3);
7617 Node* ctOfs = argument(4);
7618 Node* out = argument(5);
7619 Node* outOfs = argument(6);
7620 Node* gctr_object = argument(7);
7621 Node* ghash_object = argument(8);
7622
7623 // (1) in, ct and out are arrays.
7624 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7625 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
7626 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7627 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
7628 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
7629 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
7630
7631 // checks are the responsibility of the caller
7632 Node* in_start = in;
7633 Node* ct_start = ct;
7634 Node* out_start = out;
7635 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
7636 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
7637 in_start = array_element_address(in, inOfs, T_BYTE);
7638 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7639 out_start = array_element_address(out, outOfs, T_BYTE);
7640 }
7641
7642 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7643 // (because of the predicated logic executed earlier).
7644 // so we cast it here safely.
7645 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7646 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7647 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7648 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7649 Node* state = load_field_from_object(ghash_object, "state", "[J");
7650
7651 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
7652 return false;
7653 }
7654 // cast it to what we know it will be at runtime
7655 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7656 assert(tinst != nullptr, "GCTR obj is null");
7657 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7658 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7659 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7660 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7661 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7662 const TypeOopPtr* xtype = aklass->as_instance_type();
7663 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7664 aescrypt_object = _gvn.transform(aescrypt_object);
7665 // we need to get the start of the aescrypt_object's expanded key array
7666 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7667 if (k_start == nullptr) return false;
7668 // similarly, get the start address of the r vector
7669 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7670 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7671 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7672
7673
7674 // Call the stub, passing params
7675 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7676 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7677 stubAddr, stubName, TypePtr::BOTTOM,
7678 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7679
7680 // return cipher length (int)
7681 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7682 set_result(retvalue);
7683
7684 return true;
7685 }
7686
7687 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7688 // Return node representing slow path of predicate check.
7689 // the pseudo code we want to emulate with this predicate is:
7690 // for encryption:
7691 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7692 // for decryption:
7693 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7694 // note cipher==plain is more conservative than the original java code but that's OK
7695 //
7696
7697 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7698 // The receiver was checked for null already.
7699 Node* objGCTR = argument(7);
7700 // Load embeddedCipher field of GCTR object.
7701 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7702 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
7703
7704 // get AESCrypt klass for instanceOf check
7705 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7706 // will have same classloader as CipherBlockChaining object
7707 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7708 assert(tinst != nullptr, "GCTR obj is null");
7709 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7710
7711 // we want to do an instanceof comparison against the AESCrypt class
7712 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7713 if (!klass_AESCrypt->is_loaded()) {
7714 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7715 Node* ctrl = control();
7716 set_control(top()); // no regular fast path
7717 return ctrl;
7718 }
7719
7720 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7721 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7722 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7723 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7724 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7725
7726 return instof_false; // even if it is null
7727 }
7728
7729 //------------------------------get_state_from_digest_object-----------------------
7730 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7731 const char* state_type;
7732 switch (elem_type) {
7733 case T_BYTE: state_type = "[B"; break;
7734 case T_INT: state_type = "[I"; break;
7735 case T_LONG: state_type = "[J"; break;
7736 default: ShouldNotReachHere();
7737 }
7738 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7739 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7740 if (digest_state == nullptr) return (Node *) nullptr;
7741
7742 // now have the array, need to get the start address of the state array
7743 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7744 return state;
7745 }
7746
7747 //------------------------------get_block_size_from_sha3_object----------------------------------
7748 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
7749 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
7750 assert (block_size != nullptr, "sanity");
7751 return block_size;
7752 }
7753
7754 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7755 // Return node representing slow path of predicate check.
7756 // the pseudo code we want to emulate with this predicate is:
7757 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7758 //
7759 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7760 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7761 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7762 assert((uint)predicate < 5, "sanity");
7763
7764 // The receiver was checked for null already.
7765 Node* digestBaseObj = argument(0);
7766
7767 // get DigestBase klass for instanceOf check
7768 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7769 assert(tinst != nullptr, "digestBaseObj is null");
7770 assert(tinst->is_loaded(), "DigestBase is not loaded");
7771
7772 const char* klass_name = nullptr;
7773 switch (predicate) {
7774 case 0:
7775 if (UseMD5Intrinsics) {
7776 // we want to do an instanceof comparison against the MD5 class
7777 klass_name = "sun/security/provider/MD5";
7778 }
7779 break;
7780 case 1:
7781 if (UseSHA1Intrinsics) {
7782 // we want to do an instanceof comparison against the SHA class
7783 klass_name = "sun/security/provider/SHA";
7784 }
7785 break;
7786 case 2:
7787 if (UseSHA256Intrinsics) {
7788 // we want to do an instanceof comparison against the SHA2 class
7789 klass_name = "sun/security/provider/SHA2";
7790 }
7791 break;
7792 case 3:
7793 if (UseSHA512Intrinsics) {
7794 // we want to do an instanceof comparison against the SHA5 class
7795 klass_name = "sun/security/provider/SHA5";
7796 }
7797 break;
7798 case 4:
7799 if (UseSHA3Intrinsics) {
7800 // we want to do an instanceof comparison against the SHA3 class
7801 klass_name = "sun/security/provider/SHA3";
7802 }
7803 break;
7804 default:
7805 fatal("unknown SHA intrinsic predicate: %d", predicate);
7806 }
7807
7808 ciKlass* klass = nullptr;
7809 if (klass_name != nullptr) {
7810 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7811 }
7812 if ((klass == nullptr) || !klass->is_loaded()) {
7813 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7814 Node* ctrl = control();
7815 set_control(top()); // no intrinsic path
7816 return ctrl;
7817 }
7818 ciInstanceKlass* instklass = klass->as_instance_klass();
7819
7820 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7821 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7822 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7823 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7824
7825 return instof_false; // even if it is null
7826 }
7827
7828 //-------------inline_fma-----------------------------------
7829 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7830 Node *a = nullptr;
7831 Node *b = nullptr;
7832 Node *c = nullptr;
7833 Node* result = nullptr;
7834 switch (id) {
7835 case vmIntrinsics::_fmaD:
7836 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7837 // no receiver since it is static method
7838 a = round_double_node(argument(0));
7839 b = round_double_node(argument(2));
7840 c = round_double_node(argument(4));
7841 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7842 break;
7843 case vmIntrinsics::_fmaF:
7844 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7845 a = argument(0);
7846 b = argument(1);
7847 c = argument(2);
7848 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7849 break;
7850 default:
7851 fatal_unexpected_iid(id); break;
7852 }
7853 set_result(result);
7854 return true;
7855 }
7856
7857 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7858 // argument(0) is receiver
7859 Node* codePoint = argument(1);
7860 Node* n = nullptr;
7861
7862 switch (id) {
7863 case vmIntrinsics::_isDigit :
7864 n = new DigitNode(control(), codePoint);
7865 break;
7866 case vmIntrinsics::_isLowerCase :
7867 n = new LowerCaseNode(control(), codePoint);
7868 break;
7869 case vmIntrinsics::_isUpperCase :
7870 n = new UpperCaseNode(control(), codePoint);
7871 break;
7872 case vmIntrinsics::_isWhitespace :
7873 n = new WhitespaceNode(control(), codePoint);
7874 break;
7875 default:
7876 fatal_unexpected_iid(id);
7877 }
7878
7879 set_result(_gvn.transform(n));
7880 return true;
7881 }
7882
7883 //------------------------------inline_fp_min_max------------------------------
7884 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7885 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7886
7887 // The intrinsic should be used only when the API branches aren't predictable,
7888 // the last one performing the most important comparison. The following heuristic
7889 // uses the branch statistics to eventually bail out if necessary.
7890
7891 ciMethodData *md = callee()->method_data();
7892
7893 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
7894 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7895
7896 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7897 // Bail out if the call-site didn't contribute enough to the statistics.
7898 return false;
7899 }
7900
7901 uint taken = 0, not_taken = 0;
7902
7903 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7904 if (p->is_BranchData()) {
7905 taken = ((ciBranchData*)p)->taken();
7906 not_taken = ((ciBranchData*)p)->not_taken();
7907 }
7908 }
7909
7910 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7911 balance = balance < 0 ? -balance : balance;
7912 if ( balance > 0.2 ) {
7913 // Bail out if the most important branch is predictable enough.
7914 return false;
7915 }
7916 }
7917 */
7918
7919 Node *a = nullptr;
7920 Node *b = nullptr;
7921 Node *n = nullptr;
7922 switch (id) {
7923 case vmIntrinsics::_maxF:
7924 case vmIntrinsics::_minF:
7925 case vmIntrinsics::_maxF_strict:
7926 case vmIntrinsics::_minF_strict:
7927 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7928 a = argument(0);
7929 b = argument(1);
7930 break;
7931 case vmIntrinsics::_maxD:
7932 case vmIntrinsics::_minD:
7933 case vmIntrinsics::_maxD_strict:
7934 case vmIntrinsics::_minD_strict:
7935 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7936 a = round_double_node(argument(0));
7937 b = round_double_node(argument(2));
7938 break;
7939 default:
7940 fatal_unexpected_iid(id);
7941 break;
7942 }
7943 switch (id) {
7944 case vmIntrinsics::_maxF:
7945 case vmIntrinsics::_maxF_strict:
7946 n = new MaxFNode(a, b);
7947 break;
7948 case vmIntrinsics::_minF:
7949 case vmIntrinsics::_minF_strict:
7950 n = new MinFNode(a, b);
7951 break;
7952 case vmIntrinsics::_maxD:
7953 case vmIntrinsics::_maxD_strict:
7954 n = new MaxDNode(a, b);
7955 break;
7956 case vmIntrinsics::_minD:
7957 case vmIntrinsics::_minD_strict:
7958 n = new MinDNode(a, b);
7959 break;
7960 default:
7961 fatal_unexpected_iid(id);
7962 break;
7963 }
7964 set_result(_gvn.transform(n));
7965 return true;
7966 }
7967
7968 bool LibraryCallKit::inline_profileBoolean() {
7969 Node* counts = argument(1);
7970 const TypeAryPtr* ary = nullptr;
7971 ciArray* aobj = nullptr;
7972 if (counts->is_Con()
7973 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
7974 && (aobj = ary->const_oop()->as_array()) != nullptr
7975 && (aobj->length() == 2)) {
7976 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7977 jint false_cnt = aobj->element_value(0).as_int();
7978 jint true_cnt = aobj->element_value(1).as_int();
7979
7980 if (C->log() != nullptr) {
7981 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7982 false_cnt, true_cnt);
7983 }
7984
7985 if (false_cnt + true_cnt == 0) {
7986 // According to profile, never executed.
7987 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7988 Deoptimization::Action_reinterpret);
7989 return true;
7990 }
7991
7992 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7993 // is a number of each value occurrences.
7994 Node* result = argument(0);
7995 if (false_cnt == 0 || true_cnt == 0) {
7996 // According to profile, one value has been never seen.
7997 int expected_val = (false_cnt == 0) ? 1 : 0;
7998
7999 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8000 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8001
8002 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8003 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8004 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8005
8006 { // Slow path: uncommon trap for never seen value and then reexecute
8007 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8008 // the value has been seen at least once.
8009 PreserveJVMState pjvms(this);
8010 PreserveReexecuteState preexecs(this);
8011 jvms()->set_should_reexecute(true);
8012
8013 set_control(slow_path);
8014 set_i_o(i_o());
8015
8016 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8017 Deoptimization::Action_reinterpret);
8018 }
8019 // The guard for never seen value enables sharpening of the result and
8020 // returning a constant. It allows to eliminate branches on the same value
8021 // later on.
8022 set_control(fast_path);
8023 result = intcon(expected_val);
8024 }
8025 // Stop profiling.
8026 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8027 // By replacing method body with profile data (represented as ProfileBooleanNode
8028 // on IR level) we effectively disable profiling.
8029 // It enables full speed execution once optimized code is generated.
8030 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8031 C->record_for_igvn(profile);
8032 set_result(profile);
8033 return true;
8034 } else {
8035 // Continue profiling.
8036 // Profile data isn't available at the moment. So, execute method's bytecode version.
8037 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8038 // is compiled and counters aren't available since corresponding MethodHandle
8039 // isn't a compile-time constant.
8040 return false;
8041 }
8042 }
8043
8044 bool LibraryCallKit::inline_isCompileConstant() {
8045 Node* n = argument(0);
8046 set_result(n->is_Con() ? intcon(1) : intcon(0));
8047 return true;
8048 }
8049
8050 //------------------------------- inline_getObjectSize --------------------------------------
8051 //
8052 // Calculate the runtime size of the object/array.
8053 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8054 //
8055 bool LibraryCallKit::inline_getObjectSize() {
8056 Node* obj = argument(3);
8057 Node* klass_node = load_object_klass(obj);
8058
8059 jint layout_con = Klass::_lh_neutral_value;
8060 Node* layout_val = get_layout_helper(klass_node, layout_con);
8061 int layout_is_con = (layout_val == nullptr);
8062
8063 if (layout_is_con) {
8064 // Layout helper is constant, can figure out things at compile time.
8065
8066 if (Klass::layout_helper_is_instance(layout_con)) {
8067 // Instance case: layout_con contains the size itself.
8068 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8069 set_result(size);
8070 } else {
8071 // Array case: size is round(header + element_size*arraylength).
8072 // Since arraylength is different for every array instance, we have to
8073 // compute the whole thing at runtime.
8074
8075 Node* arr_length = load_array_length(obj);
8076
8077 int round_mask = MinObjAlignmentInBytes - 1;
8078 int hsize = Klass::layout_helper_header_size(layout_con);
8079 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8080
8081 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8082 round_mask = 0; // strength-reduce it if it goes away completely
8083 }
8084 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8085 Node* header_size = intcon(hsize + round_mask);
8086
8087 Node* lengthx = ConvI2X(arr_length);
8088 Node* headerx = ConvI2X(header_size);
8089
8090 Node* abody = lengthx;
8091 if (eshift != 0) {
8092 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8093 }
8094 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8095 if (round_mask != 0) {
8096 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8097 }
8098 size = ConvX2L(size);
8099 set_result(size);
8100 }
8101 } else {
8102 // Layout helper is not constant, need to test for array-ness at runtime.
8103
8104 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8105 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8106 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8107 record_for_igvn(result_reg);
8108
8109 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8110 if (array_ctl != nullptr) {
8111 // Array case: size is round(header + element_size*arraylength).
8112 // Since arraylength is different for every array instance, we have to
8113 // compute the whole thing at runtime.
8114
8115 PreserveJVMState pjvms(this);
8116 set_control(array_ctl);
8117 Node* arr_length = load_array_length(obj);
8118
8119 int round_mask = MinObjAlignmentInBytes - 1;
8120 Node* mask = intcon(round_mask);
8121
8122 Node* hss = intcon(Klass::_lh_header_size_shift);
8123 Node* hsm = intcon(Klass::_lh_header_size_mask);
8124 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8125 header_size = _gvn.transform(new AndINode(header_size, hsm));
8126 header_size = _gvn.transform(new AddINode(header_size, mask));
8127
8128 // There is no need to mask or shift this value.
8129 // The semantics of LShiftINode include an implicit mask to 0x1F.
8130 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8131 Node* elem_shift = layout_val;
8132
8133 Node* lengthx = ConvI2X(arr_length);
8134 Node* headerx = ConvI2X(header_size);
8135
8136 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8137 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8138 if (round_mask != 0) {
8139 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8140 }
8141 size = ConvX2L(size);
8142
8143 result_reg->init_req(_array_path, control());
8144 result_val->init_req(_array_path, size);
8145 }
8146
8147 if (!stopped()) {
8148 // Instance case: the layout helper gives us instance size almost directly,
8149 // but we need to mask out the _lh_instance_slow_path_bit.
8150 Node* size = ConvI2X(layout_val);
8151 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8152 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8153 size = _gvn.transform(new AndXNode(size, mask));
8154 size = ConvX2L(size);
8155
8156 result_reg->init_req(_instance_path, control());
8157 result_val->init_req(_instance_path, size);
8158 }
8159
8160 set_result(result_reg, result_val);
8161 }
8162
8163 return true;
8164 }
8165
8166 //------------------------------- inline_blackhole --------------------------------------
8167 //
8168 // Make sure all arguments to this node are alive.
8169 // This matches methods that were requested to be blackholed through compile commands.
8170 //
8171 bool LibraryCallKit::inline_blackhole() {
8172 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8173 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8174 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8175
8176 // Blackhole node pinches only the control, not memory. This allows
8177 // the blackhole to be pinned in the loop that computes blackholed
8178 // values, but have no other side effects, like breaking the optimizations
8179 // across the blackhole.
8180
8181 Node* bh = _gvn.transform(new BlackholeNode(control()));
8182 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8183
8184 // Bind call arguments as blackhole arguments to keep them alive
8185 uint nargs = callee()->arg_size();
8186 for (uint i = 0; i < nargs; i++) {
8187 bh->add_req(argument(i));
8188 }
8189
8190 return true;
8191 }