1 /* 2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/macroAssembler.hpp" 27 #include "ci/ciFlatArrayKlass.hpp" 28 #include "ci/ciUtilities.inline.hpp" 29 #include "classfile/vmIntrinsics.hpp" 30 #include "compiler/compileBroker.hpp" 31 #include "compiler/compileLog.hpp" 32 #include "gc/shared/barrierSet.hpp" 33 #include "jfr/support/jfrIntrinsics.hpp" 34 #include "memory/resourceArea.hpp" 35 #include "oops/klass.inline.hpp" 36 #include "oops/objArrayKlass.hpp" 37 #include "opto/addnode.hpp" 38 #include "opto/arraycopynode.hpp" 39 #include "opto/c2compiler.hpp" 40 #include "opto/castnode.hpp" 41 #include "opto/cfgnode.hpp" 42 #include "opto/convertnode.hpp" 43 #include "opto/countbitsnode.hpp" 44 #include "opto/idealKit.hpp" 45 #include "opto/library_call.hpp" 46 #include "opto/mathexactnode.hpp" 47 #include "opto/mulnode.hpp" 48 #include "opto/narrowptrnode.hpp" 49 #include "opto/opaquenode.hpp" 50 #include "opto/parse.hpp" 51 #include "opto/runtime.hpp" 52 #include "opto/rootnode.hpp" 53 #include "opto/subnode.hpp" 54 #include "prims/jvmtiExport.hpp" 55 #include "prims/jvmtiThreadState.hpp" 56 #include "prims/unsafe.hpp" 57 #include "runtime/jniHandles.inline.hpp" 58 #include "runtime/objectMonitor.hpp" 59 #include "runtime/sharedRuntime.hpp" 60 #include "runtime/stubRoutines.hpp" 61 #include "utilities/macros.hpp" 62 #include "utilities/powerOfTwo.hpp" 63 64 //---------------------------make_vm_intrinsic---------------------------- 65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) { 66 vmIntrinsicID id = m->intrinsic_id(); 67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic"); 68 69 if (!m->is_loaded()) { 70 // Do not attempt to inline unloaded methods. 71 return nullptr; 72 } 73 74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization); 75 bool is_available = false; 76 77 { 78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag 79 // the compiler must transition to '_thread_in_vm' state because both 80 // methods access VM-internal data. 81 VM_ENTRY_MARK; 82 methodHandle mh(THREAD, m->get_Method()); 83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive()); 84 if (is_available && is_virtual) { 85 is_available = vmIntrinsics::does_virtual_dispatch(id); 86 } 87 } 88 89 if (is_available) { 90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility"); 91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?"); 92 return new LibraryIntrinsic(m, is_virtual, 93 vmIntrinsics::predicates_needed(id), 94 vmIntrinsics::does_virtual_dispatch(id), 95 id); 96 } else { 97 return nullptr; 98 } 99 } 100 101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) { 102 LibraryCallKit kit(jvms, this); 103 Compile* C = kit.C; 104 int nodes = C->unique(); 105 #ifndef PRODUCT 106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 107 char buf[1000]; 108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 109 tty->print_cr("Intrinsic %s", str); 110 } 111 #endif 112 ciMethod* callee = kit.callee(); 113 const int bci = kit.bci(); 114 #ifdef ASSERT 115 Node* ctrl = kit.control(); 116 #endif 117 // Try to inline the intrinsic. 118 if (callee->check_intrinsic_candidate() && 119 kit.try_to_inline(_last_predicate)) { 120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)" 121 : "(intrinsic)"; 122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg); 123 if (C->print_intrinsics() || C->print_inlining()) { 124 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg); 125 } 126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 127 if (C->log()) { 128 C->log()->elem("intrinsic id='%s'%s nodes='%d'", 129 vmIntrinsics::name_at(intrinsic_id()), 130 (is_virtual() ? " virtual='1'" : ""), 131 C->unique() - nodes); 132 } 133 // Push the result from the inlined method onto the stack. 134 kit.push_result(); 135 C->print_inlining_update(this); 136 return kit.transfer_exceptions_into_jvms(); 137 } 138 139 // The intrinsic bailed out 140 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out"); 141 if (jvms->has_method()) { 142 // Not a root compile. 143 const char* msg; 144 if (callee->intrinsic_candidate()) { 145 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)"; 146 } else { 147 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated" 148 : "failed to inline (intrinsic), method not annotated"; 149 } 150 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg); 151 if (C->print_intrinsics() || C->print_inlining()) { 152 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg); 153 } 154 } else { 155 // Root compile 156 ResourceMark rm; 157 stringStream msg_stream; 158 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 159 vmIntrinsics::name_at(intrinsic_id()), 160 is_virtual() ? " (virtual)" : "", bci); 161 const char *msg = msg_stream.freeze(); 162 log_debug(jit, inlining)("%s", msg); 163 if (C->print_intrinsics() || C->print_inlining()) { 164 tty->print("%s", msg); 165 } 166 } 167 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 168 C->print_inlining_update(this); 169 170 return nullptr; 171 } 172 173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) { 174 LibraryCallKit kit(jvms, this); 175 Compile* C = kit.C; 176 int nodes = C->unique(); 177 _last_predicate = predicate; 178 #ifndef PRODUCT 179 assert(is_predicated() && predicate < predicates_count(), "sanity"); 180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 181 char buf[1000]; 182 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 183 tty->print_cr("Predicate for intrinsic %s", str); 184 } 185 #endif 186 ciMethod* callee = kit.callee(); 187 const int bci = kit.bci(); 188 189 Node* slow_ctl = kit.try_to_predicate(predicate); 190 if (!kit.failing()) { 191 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)" 192 : "(intrinsic, predicate)"; 193 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg); 194 if (C->print_intrinsics() || C->print_inlining()) { 195 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg); 196 } 197 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 198 if (C->log()) { 199 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'", 200 vmIntrinsics::name_at(intrinsic_id()), 201 (is_virtual() ? " virtual='1'" : ""), 202 C->unique() - nodes); 203 } 204 return slow_ctl; // Could be null if the check folds. 205 } 206 207 // The intrinsic bailed out 208 if (jvms->has_method()) { 209 // Not a root compile. 210 const char* msg = "failed to generate predicate for intrinsic"; 211 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg); 212 if (C->print_intrinsics() || C->print_inlining()) { 213 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg); 214 } 215 } else { 216 // Root compile 217 ResourceMark rm; 218 stringStream msg_stream; 219 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 220 vmIntrinsics::name_at(intrinsic_id()), 221 is_virtual() ? " (virtual)" : "", bci); 222 const char *msg = msg_stream.freeze(); 223 log_debug(jit, inlining)("%s", msg); 224 if (C->print_intrinsics() || C->print_inlining()) { 225 C->print_inlining_stream()->print("%s", msg); 226 } 227 } 228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 229 return nullptr; 230 } 231 232 bool LibraryCallKit::try_to_inline(int predicate) { 233 // Handle symbolic names for otherwise undistinguished boolean switches: 234 const bool is_store = true; 235 const bool is_compress = true; 236 const bool is_static = true; 237 const bool is_volatile = true; 238 239 if (!jvms()->has_method()) { 240 // Root JVMState has a null method. 241 assert(map()->memory()->Opcode() == Op_Parm, ""); 242 // Insert the memory aliasing node 243 set_all_memory(reset_memory()); 244 } 245 assert(merged_memory(), ""); 246 247 switch (intrinsic_id()) { 248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static); 249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static); 250 case vmIntrinsics::_getClass: return inline_native_getClass(); 251 252 case vmIntrinsics::_ceil: 253 case vmIntrinsics::_floor: 254 case vmIntrinsics::_rint: 255 case vmIntrinsics::_dsin: 256 case vmIntrinsics::_dcos: 257 case vmIntrinsics::_dtan: 258 case vmIntrinsics::_dabs: 259 case vmIntrinsics::_fabs: 260 case vmIntrinsics::_iabs: 261 case vmIntrinsics::_labs: 262 case vmIntrinsics::_datan2: 263 case vmIntrinsics::_dsqrt: 264 case vmIntrinsics::_dsqrt_strict: 265 case vmIntrinsics::_dexp: 266 case vmIntrinsics::_dlog: 267 case vmIntrinsics::_dlog10: 268 case vmIntrinsics::_dpow: 269 case vmIntrinsics::_dcopySign: 270 case vmIntrinsics::_fcopySign: 271 case vmIntrinsics::_dsignum: 272 case vmIntrinsics::_roundF: 273 case vmIntrinsics::_roundD: 274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id()); 275 276 case vmIntrinsics::_notify: 277 case vmIntrinsics::_notifyAll: 278 return inline_notify(intrinsic_id()); 279 280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh(); 289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh(); 290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 294 295 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 296 297 case vmIntrinsics::_arraySort: return inline_array_sort(); 298 case vmIntrinsics::_arrayPartition: return inline_array_partition(); 299 300 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 301 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 302 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 303 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 304 305 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 306 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 307 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 308 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 309 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 310 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 311 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U); 312 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L); 313 314 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 315 316 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode(); 317 318 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 319 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 320 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 321 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 322 323 case vmIntrinsics::_compressStringC: 324 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 325 case vmIntrinsics::_inflateStringC: 326 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 327 328 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer(); 329 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer(); 330 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false); 331 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false); 332 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false); 333 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false); 334 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false); 335 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false); 336 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false); 337 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false); 338 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false); 339 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true); 340 341 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false); 342 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false); 343 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false); 344 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false); 345 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false); 346 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false); 347 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false); 348 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false); 349 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false); 350 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true); 351 352 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false); 353 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false); 354 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false); 355 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false); 356 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false); 357 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false); 358 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false); 359 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false); 360 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false); 361 362 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false); 363 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false); 364 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false); 365 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false); 366 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false); 367 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false); 368 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false); 369 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false); 370 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false); 371 372 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true); 373 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true); 374 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true); 375 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true); 376 377 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true); 378 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true); 379 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true); 380 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true); 381 382 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false); 383 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false); 384 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false); 385 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false); 386 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false); 387 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false); 388 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false); 389 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false); 390 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false); 391 392 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false); 393 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false); 394 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false); 395 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false); 396 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false); 397 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false); 398 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false); 399 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false); 400 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false); 401 402 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false); 403 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false); 404 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false); 405 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false); 406 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false); 407 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false); 408 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false); 409 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false); 410 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false); 411 412 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false); 413 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false); 414 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false); 415 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false); 416 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false); 417 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false); 418 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false); 419 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false); 420 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false); 421 422 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 423 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile); 424 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile); 425 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 426 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 427 428 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 429 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 430 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 431 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile); 432 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed); 433 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire); 434 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release); 435 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile); 436 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed); 437 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire); 438 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release); 439 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile); 440 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 441 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 442 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 443 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile); 444 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 445 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 446 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 447 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile); 448 449 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 450 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 451 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 452 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile); 453 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire); 454 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release); 455 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile); 456 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire); 457 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release); 458 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 459 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 460 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 461 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 462 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 463 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 464 465 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile); 466 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile); 467 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 468 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 469 470 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile); 471 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile); 472 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 473 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 474 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 475 476 case vmIntrinsics::_loadFence: 477 case vmIntrinsics::_storeFence: 478 case vmIntrinsics::_storeStoreFence: 479 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 480 481 case vmIntrinsics::_onSpinWait: return inline_onspinwait(); 482 483 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread(); 484 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 485 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread(); 486 487 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache(); 488 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache(); 489 490 #if INCLUDE_JVMTI 491 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()), 492 "notifyJvmtiStart", true, false); 493 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()), 494 "notifyJvmtiEnd", false, true); 495 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()), 496 "notifyJvmtiMount", false, false); 497 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()), 498 "notifyJvmtiUnmount", false, false); 499 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide(); 500 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync(); 501 #endif 502 503 #ifdef JFR_HAVE_INTRINSICS 504 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime"); 505 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter(); 506 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit(); 507 #endif 508 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 509 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 510 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0(); 511 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true); 512 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false); 513 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 514 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 515 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray(); 516 case vmIntrinsics::_getLength: return inline_native_getLength(); 517 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 518 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 519 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 520 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 521 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT); 522 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG); 523 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 524 525 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true); 526 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false); 527 case vmIntrinsics::_newNullRestrictedArray: return inline_newNullRestrictedArray(); 528 529 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 530 531 case vmIntrinsics::_isInstance: 532 case vmIntrinsics::_getModifiers: 533 case vmIntrinsics::_isInterface: 534 case vmIntrinsics::_isArray: 535 case vmIntrinsics::_isPrimitive: 536 case vmIntrinsics::_isHidden: 537 case vmIntrinsics::_getSuperclass: 538 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id()); 539 540 case vmIntrinsics::_floatToRawIntBits: 541 case vmIntrinsics::_floatToIntBits: 542 case vmIntrinsics::_intBitsToFloat: 543 case vmIntrinsics::_doubleToRawLongBits: 544 case vmIntrinsics::_doubleToLongBits: 545 case vmIntrinsics::_longBitsToDouble: 546 case vmIntrinsics::_floatToFloat16: 547 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id()); 548 549 case vmIntrinsics::_floatIsFinite: 550 case vmIntrinsics::_floatIsInfinite: 551 case vmIntrinsics::_doubleIsFinite: 552 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id()); 553 554 case vmIntrinsics::_numberOfLeadingZeros_i: 555 case vmIntrinsics::_numberOfLeadingZeros_l: 556 case vmIntrinsics::_numberOfTrailingZeros_i: 557 case vmIntrinsics::_numberOfTrailingZeros_l: 558 case vmIntrinsics::_bitCount_i: 559 case vmIntrinsics::_bitCount_l: 560 case vmIntrinsics::_reverse_i: 561 case vmIntrinsics::_reverse_l: 562 case vmIntrinsics::_reverseBytes_i: 563 case vmIntrinsics::_reverseBytes_l: 564 case vmIntrinsics::_reverseBytes_s: 565 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id()); 566 567 case vmIntrinsics::_compress_i: 568 case vmIntrinsics::_compress_l: 569 case vmIntrinsics::_expand_i: 570 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id()); 571 572 case vmIntrinsics::_compareUnsigned_i: 573 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id()); 574 575 case vmIntrinsics::_divideUnsigned_i: 576 case vmIntrinsics::_divideUnsigned_l: 577 case vmIntrinsics::_remainderUnsigned_i: 578 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id()); 579 580 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass(); 581 582 case vmIntrinsics::_Reference_get: return inline_reference_get(); 583 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false); 584 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true); 585 586 case vmIntrinsics::_Class_cast: return inline_Class_cast(); 587 588 case vmIntrinsics::_aescrypt_encryptBlock: 589 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id()); 590 591 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 592 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 593 return inline_cipherBlockChaining_AESCrypt(intrinsic_id()); 594 595 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 596 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 597 return inline_electronicCodeBook_AESCrypt(intrinsic_id()); 598 599 case vmIntrinsics::_counterMode_AESCrypt: 600 return inline_counterMode_AESCrypt(intrinsic_id()); 601 602 case vmIntrinsics::_galoisCounterMode_AESCrypt: 603 return inline_galoisCounterMode_AESCrypt(); 604 605 case vmIntrinsics::_md5_implCompress: 606 case vmIntrinsics::_sha_implCompress: 607 case vmIntrinsics::_sha2_implCompress: 608 case vmIntrinsics::_sha5_implCompress: 609 case vmIntrinsics::_sha3_implCompress: 610 return inline_digestBase_implCompress(intrinsic_id()); 611 612 case vmIntrinsics::_digestBase_implCompressMB: 613 return inline_digestBase_implCompressMB(predicate); 614 615 case vmIntrinsics::_multiplyToLen: 616 return inline_multiplyToLen(); 617 618 case vmIntrinsics::_squareToLen: 619 return inline_squareToLen(); 620 621 case vmIntrinsics::_mulAdd: 622 return inline_mulAdd(); 623 624 case vmIntrinsics::_montgomeryMultiply: 625 return inline_montgomeryMultiply(); 626 case vmIntrinsics::_montgomerySquare: 627 return inline_montgomerySquare(); 628 629 case vmIntrinsics::_bigIntegerRightShiftWorker: 630 return inline_bigIntegerShift(true); 631 case vmIntrinsics::_bigIntegerLeftShiftWorker: 632 return inline_bigIntegerShift(false); 633 634 case vmIntrinsics::_vectorizedMismatch: 635 return inline_vectorizedMismatch(); 636 637 case vmIntrinsics::_ghash_processBlocks: 638 return inline_ghash_processBlocks(); 639 case vmIntrinsics::_chacha20Block: 640 return inline_chacha20Block(); 641 case vmIntrinsics::_base64_encodeBlock: 642 return inline_base64_encodeBlock(); 643 case vmIntrinsics::_base64_decodeBlock: 644 return inline_base64_decodeBlock(); 645 case vmIntrinsics::_poly1305_processBlocks: 646 return inline_poly1305_processBlocks(); 647 648 case vmIntrinsics::_encodeISOArray: 649 case vmIntrinsics::_encodeByteISOArray: 650 return inline_encodeISOArray(false); 651 case vmIntrinsics::_encodeAsciiArray: 652 return inline_encodeISOArray(true); 653 654 case vmIntrinsics::_updateCRC32: 655 return inline_updateCRC32(); 656 case vmIntrinsics::_updateBytesCRC32: 657 return inline_updateBytesCRC32(); 658 case vmIntrinsics::_updateByteBufferCRC32: 659 return inline_updateByteBufferCRC32(); 660 661 case vmIntrinsics::_updateBytesCRC32C: 662 return inline_updateBytesCRC32C(); 663 case vmIntrinsics::_updateDirectByteBufferCRC32C: 664 return inline_updateDirectByteBufferCRC32C(); 665 666 case vmIntrinsics::_updateBytesAdler32: 667 return inline_updateBytesAdler32(); 668 case vmIntrinsics::_updateByteBufferAdler32: 669 return inline_updateByteBufferAdler32(); 670 671 case vmIntrinsics::_profileBoolean: 672 return inline_profileBoolean(); 673 case vmIntrinsics::_isCompileConstant: 674 return inline_isCompileConstant(); 675 676 case vmIntrinsics::_countPositives: 677 return inline_countPositives(); 678 679 case vmIntrinsics::_fmaD: 680 case vmIntrinsics::_fmaF: 681 return inline_fma(intrinsic_id()); 682 683 case vmIntrinsics::_isDigit: 684 case vmIntrinsics::_isLowerCase: 685 case vmIntrinsics::_isUpperCase: 686 case vmIntrinsics::_isWhitespace: 687 return inline_character_compare(intrinsic_id()); 688 689 case vmIntrinsics::_min: 690 case vmIntrinsics::_max: 691 case vmIntrinsics::_min_strict: 692 case vmIntrinsics::_max_strict: 693 return inline_min_max(intrinsic_id()); 694 695 case vmIntrinsics::_maxF: 696 case vmIntrinsics::_minF: 697 case vmIntrinsics::_maxD: 698 case vmIntrinsics::_minD: 699 case vmIntrinsics::_maxF_strict: 700 case vmIntrinsics::_minF_strict: 701 case vmIntrinsics::_maxD_strict: 702 case vmIntrinsics::_minD_strict: 703 return inline_fp_min_max(intrinsic_id()); 704 705 case vmIntrinsics::_VectorUnaryOp: 706 return inline_vector_nary_operation(1); 707 case vmIntrinsics::_VectorBinaryOp: 708 return inline_vector_nary_operation(2); 709 case vmIntrinsics::_VectorTernaryOp: 710 return inline_vector_nary_operation(3); 711 case vmIntrinsics::_VectorFromBitsCoerced: 712 return inline_vector_frombits_coerced(); 713 case vmIntrinsics::_VectorShuffleIota: 714 return inline_vector_shuffle_iota(); 715 case vmIntrinsics::_VectorMaskOp: 716 return inline_vector_mask_operation(); 717 case vmIntrinsics::_VectorShuffleToVector: 718 return inline_vector_shuffle_to_vector(); 719 case vmIntrinsics::_VectorLoadOp: 720 return inline_vector_mem_operation(/*is_store=*/false); 721 case vmIntrinsics::_VectorLoadMaskedOp: 722 return inline_vector_mem_masked_operation(/*is_store*/false); 723 case vmIntrinsics::_VectorStoreOp: 724 return inline_vector_mem_operation(/*is_store=*/true); 725 case vmIntrinsics::_VectorStoreMaskedOp: 726 return inline_vector_mem_masked_operation(/*is_store=*/true); 727 case vmIntrinsics::_VectorGatherOp: 728 return inline_vector_gather_scatter(/*is_scatter*/ false); 729 case vmIntrinsics::_VectorScatterOp: 730 return inline_vector_gather_scatter(/*is_scatter*/ true); 731 case vmIntrinsics::_VectorReductionCoerced: 732 return inline_vector_reduction(); 733 case vmIntrinsics::_VectorTest: 734 return inline_vector_test(); 735 case vmIntrinsics::_VectorBlend: 736 return inline_vector_blend(); 737 case vmIntrinsics::_VectorRearrange: 738 return inline_vector_rearrange(); 739 case vmIntrinsics::_VectorCompare: 740 return inline_vector_compare(); 741 case vmIntrinsics::_VectorBroadcastInt: 742 return inline_vector_broadcast_int(); 743 case vmIntrinsics::_VectorConvert: 744 return inline_vector_convert(); 745 case vmIntrinsics::_VectorInsert: 746 return inline_vector_insert(); 747 case vmIntrinsics::_VectorExtract: 748 return inline_vector_extract(); 749 case vmIntrinsics::_VectorCompressExpand: 750 return inline_vector_compress_expand(); 751 case vmIntrinsics::_IndexVector: 752 return inline_index_vector(); 753 case vmIntrinsics::_IndexPartiallyInUpperRange: 754 return inline_index_partially_in_upper_range(); 755 756 case vmIntrinsics::_getObjectSize: 757 return inline_getObjectSize(); 758 759 case vmIntrinsics::_blackhole: 760 return inline_blackhole(); 761 762 default: 763 // If you get here, it may be that someone has added a new intrinsic 764 // to the list in vmIntrinsics.hpp without implementing it here. 765 #ifndef PRODUCT 766 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 767 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 768 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id())); 769 } 770 #endif 771 return false; 772 } 773 } 774 775 Node* LibraryCallKit::try_to_predicate(int predicate) { 776 if (!jvms()->has_method()) { 777 // Root JVMState has a null method. 778 assert(map()->memory()->Opcode() == Op_Parm, ""); 779 // Insert the memory aliasing node 780 set_all_memory(reset_memory()); 781 } 782 assert(merged_memory(), ""); 783 784 switch (intrinsic_id()) { 785 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 786 return inline_cipherBlockChaining_AESCrypt_predicate(false); 787 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 788 return inline_cipherBlockChaining_AESCrypt_predicate(true); 789 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 790 return inline_electronicCodeBook_AESCrypt_predicate(false); 791 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 792 return inline_electronicCodeBook_AESCrypt_predicate(true); 793 case vmIntrinsics::_counterMode_AESCrypt: 794 return inline_counterMode_AESCrypt_predicate(); 795 case vmIntrinsics::_digestBase_implCompressMB: 796 return inline_digestBase_implCompressMB_predicate(predicate); 797 case vmIntrinsics::_galoisCounterMode_AESCrypt: 798 return inline_galoisCounterMode_AESCrypt_predicate(); 799 800 default: 801 // If you get here, it may be that someone has added a new intrinsic 802 // to the list in vmIntrinsics.hpp without implementing it here. 803 #ifndef PRODUCT 804 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 805 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 806 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id())); 807 } 808 #endif 809 Node* slow_ctl = control(); 810 set_control(top()); // No fast path intrinsic 811 return slow_ctl; 812 } 813 } 814 815 //------------------------------set_result------------------------------- 816 // Helper function for finishing intrinsics. 817 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 818 record_for_igvn(region); 819 set_control(_gvn.transform(region)); 820 set_result( _gvn.transform(value)); 821 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 822 } 823 824 //------------------------------generate_guard--------------------------- 825 // Helper function for generating guarded fast-slow graph structures. 826 // The given 'test', if true, guards a slow path. If the test fails 827 // then a fast path can be taken. (We generally hope it fails.) 828 // In all cases, GraphKit::control() is updated to the fast path. 829 // The returned value represents the control for the slow path. 830 // The return value is never 'top'; it is either a valid control 831 // or null if it is obvious that the slow path can never be taken. 832 // Also, if region and the slow control are not null, the slow edge 833 // is appended to the region. 834 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 835 if (stopped()) { 836 // Already short circuited. 837 return nullptr; 838 } 839 840 // Build an if node and its projections. 841 // If test is true we take the slow path, which we assume is uncommon. 842 if (_gvn.type(test) == TypeInt::ZERO) { 843 // The slow branch is never taken. No need to build this guard. 844 return nullptr; 845 } 846 847 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 848 849 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 850 if (if_slow == top()) { 851 // The slow branch is never taken. No need to build this guard. 852 return nullptr; 853 } 854 855 if (region != nullptr) 856 region->add_req(if_slow); 857 858 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 859 set_control(if_fast); 860 861 return if_slow; 862 } 863 864 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 865 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 866 } 867 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 868 return generate_guard(test, region, PROB_FAIR); 869 } 870 871 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 872 Node* *pos_index) { 873 if (stopped()) 874 return nullptr; // already stopped 875 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 876 return nullptr; // index is already adequately typed 877 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 878 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 879 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 880 if (is_neg != nullptr && pos_index != nullptr) { 881 // Emulate effect of Parse::adjust_map_after_if. 882 Node* ccast = new CastIINode(control(), index, TypeInt::POS); 883 (*pos_index) = _gvn.transform(ccast); 884 } 885 return is_neg; 886 } 887 888 // Make sure that 'position' is a valid limit index, in [0..length]. 889 // There are two equivalent plans for checking this: 890 // A. (offset + copyLength) unsigned<= arrayLength 891 // B. offset <= (arrayLength - copyLength) 892 // We require that all of the values above, except for the sum and 893 // difference, are already known to be non-negative. 894 // Plan A is robust in the face of overflow, if offset and copyLength 895 // are both hugely positive. 896 // 897 // Plan B is less direct and intuitive, but it does not overflow at 898 // all, since the difference of two non-negatives is always 899 // representable. Whenever Java methods must perform the equivalent 900 // check they generally use Plan B instead of Plan A. 901 // For the moment we use Plan A. 902 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 903 Node* subseq_length, 904 Node* array_length, 905 RegionNode* region) { 906 if (stopped()) 907 return nullptr; // already stopped 908 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 909 if (zero_offset && subseq_length->eqv_uncast(array_length)) 910 return nullptr; // common case of whole-array copy 911 Node* last = subseq_length; 912 if (!zero_offset) // last += offset 913 last = _gvn.transform(new AddINode(last, offset)); 914 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 915 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 916 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 917 return is_over; 918 } 919 920 // Emit range checks for the given String.value byte array 921 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) { 922 if (stopped()) { 923 return; // already stopped 924 } 925 RegionNode* bailout = new RegionNode(1); 926 record_for_igvn(bailout); 927 if (char_count) { 928 // Convert char count to byte count 929 count = _gvn.transform(new LShiftINode(count, intcon(1))); 930 } 931 932 // Offset and count must not be negative 933 generate_negative_guard(offset, bailout); 934 generate_negative_guard(count, bailout); 935 // Offset + count must not exceed length of array 936 generate_limit_guard(offset, count, load_array_length(array), bailout); 937 938 if (bailout->req() > 1) { 939 PreserveJVMState pjvms(this); 940 set_control(_gvn.transform(bailout)); 941 uncommon_trap(Deoptimization::Reason_intrinsic, 942 Deoptimization::Action_maybe_recompile); 943 } 944 } 945 946 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset, 947 bool is_immutable) { 948 ciKlass* thread_klass = env()->Thread_klass(); 949 const Type* thread_type 950 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 951 952 Node* thread = _gvn.transform(new ThreadLocalNode()); 953 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset)); 954 tls_output = thread; 955 956 Node* thread_obj_handle 957 = (is_immutable 958 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(), 959 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered) 960 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered)); 961 thread_obj_handle = _gvn.transform(thread_obj_handle); 962 963 DecoratorSet decorators = IN_NATIVE; 964 if (is_immutable) { 965 decorators |= C2_IMMUTABLE_MEMORY; 966 } 967 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators); 968 } 969 970 //--------------------------generate_current_thread-------------------- 971 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 972 return current_thread_helper(tls_output, JavaThread::threadObj_offset(), 973 /*is_immutable*/false); 974 } 975 976 //--------------------------generate_virtual_thread-------------------- 977 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) { 978 return current_thread_helper(tls_output, JavaThread::vthread_offset(), 979 !C->method()->changes_current_thread()); 980 } 981 982 //------------------------------make_string_method_node------------------------ 983 // Helper method for String intrinsic functions. This version is called with 984 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 985 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 986 // containing the lengths of str1 and str2. 987 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 988 Node* result = nullptr; 989 switch (opcode) { 990 case Op_StrIndexOf: 991 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 992 str1_start, cnt1, str2_start, cnt2, ae); 993 break; 994 case Op_StrComp: 995 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 996 str1_start, cnt1, str2_start, cnt2, ae); 997 break; 998 case Op_StrEquals: 999 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1000 // Use the constant length if there is one because optimized match rule may exist. 1001 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1002 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1003 break; 1004 default: 1005 ShouldNotReachHere(); 1006 return nullptr; 1007 } 1008 1009 // All these intrinsics have checks. 1010 C->set_has_split_ifs(true); // Has chance for split-if optimization 1011 clear_upper_avx(); 1012 1013 return _gvn.transform(result); 1014 } 1015 1016 //------------------------------inline_string_compareTo------------------------ 1017 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1018 Node* arg1 = argument(0); 1019 Node* arg2 = argument(1); 1020 1021 arg1 = must_be_not_null(arg1, true); 1022 arg2 = must_be_not_null(arg2, true); 1023 1024 // Get start addr and length of first argument 1025 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1026 Node* arg1_cnt = load_array_length(arg1); 1027 1028 // Get start addr and length of second argument 1029 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1030 Node* arg2_cnt = load_array_length(arg2); 1031 1032 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1033 set_result(result); 1034 return true; 1035 } 1036 1037 //------------------------------inline_string_equals------------------------ 1038 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1039 Node* arg1 = argument(0); 1040 Node* arg2 = argument(1); 1041 1042 // paths (plus control) merge 1043 RegionNode* region = new RegionNode(3); 1044 Node* phi = new PhiNode(region, TypeInt::BOOL); 1045 1046 if (!stopped()) { 1047 1048 arg1 = must_be_not_null(arg1, true); 1049 arg2 = must_be_not_null(arg2, true); 1050 1051 // Get start addr and length of first argument 1052 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1053 Node* arg1_cnt = load_array_length(arg1); 1054 1055 // Get start addr and length of second argument 1056 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1057 Node* arg2_cnt = load_array_length(arg2); 1058 1059 // Check for arg1_cnt != arg2_cnt 1060 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1061 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1062 Node* if_ne = generate_slow_guard(bol, nullptr); 1063 if (if_ne != nullptr) { 1064 phi->init_req(2, intcon(0)); 1065 region->init_req(2, if_ne); 1066 } 1067 1068 // Check for count == 0 is done by assembler code for StrEquals. 1069 1070 if (!stopped()) { 1071 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1072 phi->init_req(1, equals); 1073 region->init_req(1, control()); 1074 } 1075 } 1076 1077 // post merge 1078 set_control(_gvn.transform(region)); 1079 record_for_igvn(region); 1080 1081 set_result(_gvn.transform(phi)); 1082 return true; 1083 } 1084 1085 //------------------------------inline_array_equals---------------------------- 1086 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1087 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1088 Node* arg1 = argument(0); 1089 Node* arg2 = argument(1); 1090 1091 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1092 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1093 clear_upper_avx(); 1094 1095 return true; 1096 } 1097 1098 1099 //------------------------------inline_countPositives------------------------------ 1100 bool LibraryCallKit::inline_countPositives() { 1101 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1102 return false; 1103 } 1104 1105 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters"); 1106 // no receiver since it is static method 1107 Node* ba = argument(0); 1108 Node* offset = argument(1); 1109 Node* len = argument(2); 1110 1111 ba = must_be_not_null(ba, true); 1112 1113 // Range checks 1114 generate_string_range_check(ba, offset, len, false); 1115 if (stopped()) { 1116 return true; 1117 } 1118 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1119 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1120 set_result(_gvn.transform(result)); 1121 clear_upper_avx(); 1122 return true; 1123 } 1124 1125 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) { 1126 Node* index = argument(0); 1127 Node* length = bt == T_INT ? argument(1) : argument(2); 1128 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1129 return false; 1130 } 1131 1132 // check that length is positive 1133 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt)); 1134 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1135 1136 { 1137 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1138 uncommon_trap(Deoptimization::Reason_intrinsic, 1139 Deoptimization::Action_make_not_entrant); 1140 } 1141 1142 if (stopped()) { 1143 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success 1144 return true; 1145 } 1146 1147 // length is now known positive, add a cast node to make this explicit 1148 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long(); 1149 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type( 1150 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), 1151 ConstraintCastNode::RegularDependency, bt); 1152 casted_length = _gvn.transform(casted_length); 1153 replace_in_map(length, casted_length); 1154 length = casted_length; 1155 1156 // Use an unsigned comparison for the range check itself 1157 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true)); 1158 BoolTest::mask btest = BoolTest::lt; 1159 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1160 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1161 _gvn.set_type(rc, rc->Value(&_gvn)); 1162 if (!rc_bool->is_Con()) { 1163 record_for_igvn(rc); 1164 } 1165 set_control(_gvn.transform(new IfTrueNode(rc))); 1166 { 1167 PreserveJVMState pjvms(this); 1168 set_control(_gvn.transform(new IfFalseNode(rc))); 1169 uncommon_trap(Deoptimization::Reason_range_check, 1170 Deoptimization::Action_make_not_entrant); 1171 } 1172 1173 if (stopped()) { 1174 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success 1175 return true; 1176 } 1177 1178 // index is now known to be >= 0 and < length, cast it 1179 Node* result = ConstraintCastNode::make_cast_for_basic_type( 1180 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), 1181 ConstraintCastNode::RegularDependency, bt); 1182 result = _gvn.transform(result); 1183 set_result(result); 1184 replace_in_map(index, result); 1185 return true; 1186 } 1187 1188 //------------------------------inline_string_indexOf------------------------ 1189 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1190 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1191 return false; 1192 } 1193 Node* src = argument(0); 1194 Node* tgt = argument(1); 1195 1196 // Make the merge point 1197 RegionNode* result_rgn = new RegionNode(4); 1198 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 1199 1200 src = must_be_not_null(src, true); 1201 tgt = must_be_not_null(tgt, true); 1202 1203 // Get start addr and length of source string 1204 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1205 Node* src_count = load_array_length(src); 1206 1207 // Get start addr and length of substring 1208 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1209 Node* tgt_count = load_array_length(tgt); 1210 1211 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1212 // Divide src size by 2 if String is UTF16 encoded 1213 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1214 } 1215 if (ae == StrIntrinsicNode::UU) { 1216 // Divide substring size by 2 if String is UTF16 encoded 1217 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1218 } 1219 1220 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae); 1221 if (result != nullptr) { 1222 result_phi->init_req(3, result); 1223 result_rgn->init_req(3, control()); 1224 } 1225 set_control(_gvn.transform(result_rgn)); 1226 record_for_igvn(result_rgn); 1227 set_result(_gvn.transform(result_phi)); 1228 1229 return true; 1230 } 1231 1232 //-----------------------------inline_string_indexOf----------------------- 1233 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1234 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1235 return false; 1236 } 1237 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1238 return false; 1239 } 1240 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1241 Node* src = argument(0); // byte[] 1242 Node* src_count = argument(1); // char count 1243 Node* tgt = argument(2); // byte[] 1244 Node* tgt_count = argument(3); // char count 1245 Node* from_index = argument(4); // char index 1246 1247 src = must_be_not_null(src, true); 1248 tgt = must_be_not_null(tgt, true); 1249 1250 // Multiply byte array index by 2 if String is UTF16 encoded 1251 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1252 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1253 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1254 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1255 1256 // Range checks 1257 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1258 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1259 if (stopped()) { 1260 return true; 1261 } 1262 1263 RegionNode* region = new RegionNode(5); 1264 Node* phi = new PhiNode(region, TypeInt::INT); 1265 1266 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae); 1267 if (result != nullptr) { 1268 // The result is index relative to from_index if substring was found, -1 otherwise. 1269 // Generate code which will fold into cmove. 1270 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1271 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1272 1273 Node* if_lt = generate_slow_guard(bol, nullptr); 1274 if (if_lt != nullptr) { 1275 // result == -1 1276 phi->init_req(3, result); 1277 region->init_req(3, if_lt); 1278 } 1279 if (!stopped()) { 1280 result = _gvn.transform(new AddINode(result, from_index)); 1281 phi->init_req(4, result); 1282 region->init_req(4, control()); 1283 } 1284 } 1285 1286 set_control(_gvn.transform(region)); 1287 record_for_igvn(region); 1288 set_result(_gvn.transform(phi)); 1289 clear_upper_avx(); 1290 1291 return true; 1292 } 1293 1294 // Create StrIndexOfNode with fast path checks 1295 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1296 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1297 // Check for substr count > string count 1298 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1299 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1300 Node* if_gt = generate_slow_guard(bol, nullptr); 1301 if (if_gt != nullptr) { 1302 phi->init_req(1, intcon(-1)); 1303 region->init_req(1, if_gt); 1304 } 1305 if (!stopped()) { 1306 // Check for substr count == 0 1307 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1308 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1309 Node* if_zero = generate_slow_guard(bol, nullptr); 1310 if (if_zero != nullptr) { 1311 phi->init_req(2, intcon(0)); 1312 region->init_req(2, if_zero); 1313 } 1314 } 1315 if (!stopped()) { 1316 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1317 } 1318 return nullptr; 1319 } 1320 1321 //-----------------------------inline_string_indexOfChar----------------------- 1322 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) { 1323 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1324 return false; 1325 } 1326 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1327 return false; 1328 } 1329 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1330 Node* src = argument(0); // byte[] 1331 Node* int_ch = argument(1); 1332 Node* from_index = argument(2); 1333 Node* max = argument(3); 1334 1335 src = must_be_not_null(src, true); 1336 1337 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1338 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1339 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1340 1341 // Range checks 1342 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U); 1343 1344 // Check for int_ch >= 0 1345 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0))); 1346 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge)); 1347 { 1348 BuildCutout unless(this, int_ch_bol, PROB_MAX); 1349 uncommon_trap(Deoptimization::Reason_intrinsic, 1350 Deoptimization::Action_maybe_recompile); 1351 } 1352 if (stopped()) { 1353 return true; 1354 } 1355 1356 RegionNode* region = new RegionNode(3); 1357 Node* phi = new PhiNode(region, TypeInt::INT); 1358 1359 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae); 1360 C->set_has_split_ifs(true); // Has chance for split-if optimization 1361 _gvn.transform(result); 1362 1363 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1364 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1365 1366 Node* if_lt = generate_slow_guard(bol, nullptr); 1367 if (if_lt != nullptr) { 1368 // result == -1 1369 phi->init_req(2, result); 1370 region->init_req(2, if_lt); 1371 } 1372 if (!stopped()) { 1373 result = _gvn.transform(new AddINode(result, from_index)); 1374 phi->init_req(1, result); 1375 region->init_req(1, control()); 1376 } 1377 set_control(_gvn.transform(region)); 1378 record_for_igvn(region); 1379 set_result(_gvn.transform(phi)); 1380 clear_upper_avx(); 1381 1382 return true; 1383 } 1384 //---------------------------inline_string_copy--------------------- 1385 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1386 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1387 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1388 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1389 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1390 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1391 bool LibraryCallKit::inline_string_copy(bool compress) { 1392 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1393 return false; 1394 } 1395 int nargs = 5; // 2 oops, 3 ints 1396 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1397 1398 Node* src = argument(0); 1399 Node* src_offset = argument(1); 1400 Node* dst = argument(2); 1401 Node* dst_offset = argument(3); 1402 Node* length = argument(4); 1403 1404 // Check for allocation before we add nodes that would confuse 1405 // tightly_coupled_allocation() 1406 AllocateArrayNode* alloc = tightly_coupled_allocation(dst); 1407 1408 // Figure out the size and type of the elements we will be copying. 1409 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 1410 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr(); 1411 if (src_type == nullptr || dst_type == nullptr) { 1412 return false; 1413 } 1414 BasicType src_elem = src_type->elem()->array_element_basic_type(); 1415 BasicType dst_elem = dst_type->elem()->array_element_basic_type(); 1416 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1417 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1418 "Unsupported array types for inline_string_copy"); 1419 1420 src = must_be_not_null(src, true); 1421 dst = must_be_not_null(dst, true); 1422 1423 // Convert char[] offsets to byte[] offsets 1424 bool convert_src = (compress && src_elem == T_BYTE); 1425 bool convert_dst = (!compress && dst_elem == T_BYTE); 1426 if (convert_src) { 1427 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1428 } else if (convert_dst) { 1429 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1430 } 1431 1432 // Range checks 1433 generate_string_range_check(src, src_offset, length, convert_src); 1434 generate_string_range_check(dst, dst_offset, length, convert_dst); 1435 if (stopped()) { 1436 return true; 1437 } 1438 1439 Node* src_start = array_element_address(src, src_offset, src_elem); 1440 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1441 // 'src_start' points to src array + scaled offset 1442 // 'dst_start' points to dst array + scaled offset 1443 Node* count = nullptr; 1444 if (compress) { 1445 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1446 } else { 1447 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1448 } 1449 1450 if (alloc != nullptr) { 1451 if (alloc->maybe_set_complete(&_gvn)) { 1452 // "You break it, you buy it." 1453 InitializeNode* init = alloc->initialization(); 1454 assert(init->is_complete(), "we just did this"); 1455 init->set_complete_with_arraycopy(); 1456 assert(dst->is_CheckCastPP(), "sanity"); 1457 assert(dst->in(0)->in(0) == init, "dest pinned"); 1458 } 1459 // Do not let stores that initialize this object be reordered with 1460 // a subsequent store that would make this object accessible by 1461 // other threads. 1462 // Record what AllocateNode this StoreStore protects so that 1463 // escape analysis can go from the MemBarStoreStoreNode to the 1464 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1465 // based on the escape status of the AllocateNode. 1466 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1467 } 1468 if (compress) { 1469 set_result(_gvn.transform(count)); 1470 } 1471 clear_upper_avx(); 1472 1473 return true; 1474 } 1475 1476 #ifdef _LP64 1477 #define XTOP ,top() /*additional argument*/ 1478 #else //_LP64 1479 #define XTOP /*no additional argument*/ 1480 #endif //_LP64 1481 1482 //------------------------inline_string_toBytesU-------------------------- 1483 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1484 bool LibraryCallKit::inline_string_toBytesU() { 1485 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1486 return false; 1487 } 1488 // Get the arguments. 1489 Node* value = argument(0); 1490 Node* offset = argument(1); 1491 Node* length = argument(2); 1492 1493 Node* newcopy = nullptr; 1494 1495 // Set the original stack and the reexecute bit for the interpreter to reexecute 1496 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1497 { PreserveReexecuteState preexecs(this); 1498 jvms()->set_should_reexecute(true); 1499 1500 // Check if a null path was taken unconditionally. 1501 value = null_check(value); 1502 1503 RegionNode* bailout = new RegionNode(1); 1504 record_for_igvn(bailout); 1505 1506 // Range checks 1507 generate_negative_guard(offset, bailout); 1508 generate_negative_guard(length, bailout); 1509 generate_limit_guard(offset, length, load_array_length(value), bailout); 1510 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1511 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1512 1513 if (bailout->req() > 1) { 1514 PreserveJVMState pjvms(this); 1515 set_control(_gvn.transform(bailout)); 1516 uncommon_trap(Deoptimization::Reason_intrinsic, 1517 Deoptimization::Action_maybe_recompile); 1518 } 1519 if (stopped()) { 1520 return true; 1521 } 1522 1523 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1524 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1525 newcopy = new_array(klass_node, size, 0); // no arguments to push 1526 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy); 1527 guarantee(alloc != nullptr, "created above"); 1528 1529 // Calculate starting addresses. 1530 Node* src_start = array_element_address(value, offset, T_CHAR); 1531 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1532 1533 // Check if src array address is aligned to HeapWordSize (dst is always aligned) 1534 const TypeInt* toffset = gvn().type(offset)->is_int(); 1535 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1536 1537 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1538 const char* copyfunc_name = "arraycopy"; 1539 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1540 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1541 OptoRuntime::fast_arraycopy_Type(), 1542 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1543 src_start, dst_start, ConvI2X(length) XTOP); 1544 // Do not let reads from the cloned object float above the arraycopy. 1545 if (alloc->maybe_set_complete(&_gvn)) { 1546 // "You break it, you buy it." 1547 InitializeNode* init = alloc->initialization(); 1548 assert(init->is_complete(), "we just did this"); 1549 init->set_complete_with_arraycopy(); 1550 assert(newcopy->is_CheckCastPP(), "sanity"); 1551 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1552 } 1553 // Do not let stores that initialize this object be reordered with 1554 // a subsequent store that would make this object accessible by 1555 // other threads. 1556 // Record what AllocateNode this StoreStore protects so that 1557 // escape analysis can go from the MemBarStoreStoreNode to the 1558 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1559 // based on the escape status of the AllocateNode. 1560 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1561 } // original reexecute is set back here 1562 1563 C->set_has_split_ifs(true); // Has chance for split-if optimization 1564 if (!stopped()) { 1565 set_result(newcopy); 1566 } 1567 clear_upper_avx(); 1568 1569 return true; 1570 } 1571 1572 //------------------------inline_string_getCharsU-------------------------- 1573 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1574 bool LibraryCallKit::inline_string_getCharsU() { 1575 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1576 return false; 1577 } 1578 1579 // Get the arguments. 1580 Node* src = argument(0); 1581 Node* src_begin = argument(1); 1582 Node* src_end = argument(2); // exclusive offset (i < src_end) 1583 Node* dst = argument(3); 1584 Node* dst_begin = argument(4); 1585 1586 // Check for allocation before we add nodes that would confuse 1587 // tightly_coupled_allocation() 1588 AllocateArrayNode* alloc = tightly_coupled_allocation(dst); 1589 1590 // Check if a null path was taken unconditionally. 1591 src = null_check(src); 1592 dst = null_check(dst); 1593 if (stopped()) { 1594 return true; 1595 } 1596 1597 // Get length and convert char[] offset to byte[] offset 1598 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1599 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1600 1601 // Range checks 1602 generate_string_range_check(src, src_begin, length, true); 1603 generate_string_range_check(dst, dst_begin, length, false); 1604 if (stopped()) { 1605 return true; 1606 } 1607 1608 if (!stopped()) { 1609 // Calculate starting addresses. 1610 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1611 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1612 1613 // Check if array addresses are aligned to HeapWordSize 1614 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1615 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1616 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1617 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1618 1619 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1620 const char* copyfunc_name = "arraycopy"; 1621 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1622 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1623 OptoRuntime::fast_arraycopy_Type(), 1624 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1625 src_start, dst_start, ConvI2X(length) XTOP); 1626 // Do not let reads from the cloned object float above the arraycopy. 1627 if (alloc != nullptr) { 1628 if (alloc->maybe_set_complete(&_gvn)) { 1629 // "You break it, you buy it." 1630 InitializeNode* init = alloc->initialization(); 1631 assert(init->is_complete(), "we just did this"); 1632 init->set_complete_with_arraycopy(); 1633 assert(dst->is_CheckCastPP(), "sanity"); 1634 assert(dst->in(0)->in(0) == init, "dest pinned"); 1635 } 1636 // Do not let stores that initialize this object be reordered with 1637 // a subsequent store that would make this object accessible by 1638 // other threads. 1639 // Record what AllocateNode this StoreStore protects so that 1640 // escape analysis can go from the MemBarStoreStoreNode to the 1641 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1642 // based on the escape status of the AllocateNode. 1643 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1644 } else { 1645 insert_mem_bar(Op_MemBarCPUOrder); 1646 } 1647 } 1648 1649 C->set_has_split_ifs(true); // Has chance for split-if optimization 1650 return true; 1651 } 1652 1653 //----------------------inline_string_char_access---------------------------- 1654 // Store/Load char to/from byte[] array. 1655 // static void StringUTF16.putChar(byte[] val, int index, int c) 1656 // static char StringUTF16.getChar(byte[] val, int index) 1657 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1658 Node* value = argument(0); 1659 Node* index = argument(1); 1660 Node* ch = is_store ? argument(2) : nullptr; 1661 1662 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1663 // correctly requires matched array shapes. 1664 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1665 "sanity: byte[] and char[] bases agree"); 1666 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1667 "sanity: byte[] and char[] scales agree"); 1668 1669 // Bail when getChar over constants is requested: constant folding would 1670 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1671 // Java method would constant fold nicely instead. 1672 if (!is_store && value->is_Con() && index->is_Con()) { 1673 return false; 1674 } 1675 1676 // Save state and restore on bailout 1677 uint old_sp = sp(); 1678 SafePointNode* old_map = clone_map(); 1679 1680 value = must_be_not_null(value, true); 1681 1682 Node* adr = array_element_address(value, index, T_CHAR); 1683 if (adr->is_top()) { 1684 set_map(old_map); 1685 set_sp(old_sp); 1686 return false; 1687 } 1688 destruct_map_clone(old_map); 1689 if (is_store) { 1690 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED); 1691 } else { 1692 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); 1693 set_result(ch); 1694 } 1695 return true; 1696 } 1697 1698 //--------------------------round_double_node-------------------------------- 1699 // Round a double node if necessary. 1700 Node* LibraryCallKit::round_double_node(Node* n) { 1701 if (Matcher::strict_fp_requires_explicit_rounding) { 1702 #ifdef IA32 1703 if (UseSSE < 2) { 1704 n = _gvn.transform(new RoundDoubleNode(nullptr, n)); 1705 } 1706 #else 1707 Unimplemented(); 1708 #endif // IA32 1709 } 1710 return n; 1711 } 1712 1713 //------------------------------inline_math----------------------------------- 1714 // public static double Math.abs(double) 1715 // public static double Math.sqrt(double) 1716 // public static double Math.log(double) 1717 // public static double Math.log10(double) 1718 // public static double Math.round(double) 1719 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) { 1720 Node* arg = round_double_node(argument(0)); 1721 Node* n = nullptr; 1722 switch (id) { 1723 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1724 case vmIntrinsics::_dsqrt: 1725 case vmIntrinsics::_dsqrt_strict: 1726 n = new SqrtDNode(C, control(), arg); break; 1727 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break; 1728 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break; 1729 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break; 1730 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break; 1731 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break; 1732 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break; 1733 default: fatal_unexpected_iid(id); break; 1734 } 1735 set_result(_gvn.transform(n)); 1736 return true; 1737 } 1738 1739 //------------------------------inline_math----------------------------------- 1740 // public static float Math.abs(float) 1741 // public static int Math.abs(int) 1742 // public static long Math.abs(long) 1743 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1744 Node* arg = argument(0); 1745 Node* n = nullptr; 1746 switch (id) { 1747 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break; 1748 case vmIntrinsics::_iabs: n = new AbsINode( arg); break; 1749 case vmIntrinsics::_labs: n = new AbsLNode( arg); break; 1750 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break; 1751 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break; 1752 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break; 1753 default: fatal_unexpected_iid(id); break; 1754 } 1755 set_result(_gvn.transform(n)); 1756 return true; 1757 } 1758 1759 //------------------------------runtime_math----------------------------- 1760 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1761 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1762 "must be (DD)D or (D)D type"); 1763 1764 // Inputs 1765 Node* a = round_double_node(argument(0)); 1766 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr; 1767 1768 const TypePtr* no_memory_effects = nullptr; 1769 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName, 1770 no_memory_effects, 1771 a, top(), b, b ? top() : nullptr); 1772 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1773 #ifdef ASSERT 1774 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1775 assert(value_top == top(), "second value must be top"); 1776 #endif 1777 1778 set_result(value); 1779 return true; 1780 } 1781 1782 //------------------------------inline_math_pow----------------------------- 1783 bool LibraryCallKit::inline_math_pow() { 1784 Node* exp = round_double_node(argument(2)); 1785 const TypeD* d = _gvn.type(exp)->isa_double_constant(); 1786 if (d != nullptr) { 1787 if (d->getd() == 2.0) { 1788 // Special case: pow(x, 2.0) => x * x 1789 Node* base = round_double_node(argument(0)); 1790 set_result(_gvn.transform(new MulDNode(base, base))); 1791 return true; 1792 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) { 1793 // Special case: pow(x, 0.5) => sqrt(x) 1794 Node* base = round_double_node(argument(0)); 1795 Node* zero = _gvn.zerocon(T_DOUBLE); 1796 1797 RegionNode* region = new RegionNode(3); 1798 Node* phi = new PhiNode(region, Type::DOUBLE); 1799 1800 Node* cmp = _gvn.transform(new CmpDNode(base, zero)); 1801 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0. 1802 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0). 1803 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0. 1804 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le)); 1805 1806 Node* if_pow = generate_slow_guard(test, nullptr); 1807 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base)); 1808 phi->init_req(1, value_sqrt); 1809 region->init_req(1, control()); 1810 1811 if (if_pow != nullptr) { 1812 set_control(if_pow); 1813 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() : 1814 CAST_FROM_FN_PTR(address, SharedRuntime::dpow); 1815 const TypePtr* no_memory_effects = nullptr; 1816 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW", 1817 no_memory_effects, base, top(), exp, top()); 1818 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1819 #ifdef ASSERT 1820 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1821 assert(value_top == top(), "second value must be top"); 1822 #endif 1823 phi->init_req(2, value_pow); 1824 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control))); 1825 } 1826 1827 C->set_has_split_ifs(true); // Has chance for split-if optimization 1828 set_control(_gvn.transform(region)); 1829 record_for_igvn(region); 1830 set_result(_gvn.transform(phi)); 1831 1832 return true; 1833 } 1834 } 1835 1836 return StubRoutines::dpow() != nullptr ? 1837 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1838 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW"); 1839 } 1840 1841 //------------------------------inline_math_native----------------------------- 1842 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1843 switch (id) { 1844 case vmIntrinsics::_dsin: 1845 return StubRoutines::dsin() != nullptr ? 1846 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1847 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN"); 1848 case vmIntrinsics::_dcos: 1849 return StubRoutines::dcos() != nullptr ? 1850 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1851 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS"); 1852 case vmIntrinsics::_dtan: 1853 return StubRoutines::dtan() != nullptr ? 1854 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") : 1855 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN"); 1856 case vmIntrinsics::_dexp: 1857 return StubRoutines::dexp() != nullptr ? 1858 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1859 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP"); 1860 case vmIntrinsics::_dlog: 1861 return StubRoutines::dlog() != nullptr ? 1862 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1863 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG"); 1864 case vmIntrinsics::_dlog10: 1865 return StubRoutines::dlog10() != nullptr ? 1866 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") : 1867 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10"); 1868 1869 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false; 1870 case vmIntrinsics::_ceil: 1871 case vmIntrinsics::_floor: 1872 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false; 1873 1874 case vmIntrinsics::_dsqrt: 1875 case vmIntrinsics::_dsqrt_strict: 1876 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false; 1877 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false; 1878 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false; 1879 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false; 1880 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false; 1881 1882 case vmIntrinsics::_dpow: return inline_math_pow(); 1883 case vmIntrinsics::_dcopySign: return inline_double_math(id); 1884 case vmIntrinsics::_fcopySign: return inline_math(id); 1885 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false; 1886 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false; 1887 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false; 1888 1889 // These intrinsics are not yet correctly implemented 1890 case vmIntrinsics::_datan2: 1891 return false; 1892 1893 default: 1894 fatal_unexpected_iid(id); 1895 return false; 1896 } 1897 } 1898 1899 //----------------------------inline_notify-----------------------------------* 1900 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1901 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1902 address func; 1903 if (id == vmIntrinsics::_notify) { 1904 func = OptoRuntime::monitor_notify_Java(); 1905 } else { 1906 func = OptoRuntime::monitor_notifyAll_Java(); 1907 } 1908 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0)); 1909 make_slow_call_ex(call, env()->Throwable_klass(), false); 1910 return true; 1911 } 1912 1913 1914 //----------------------------inline_min_max----------------------------------- 1915 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1916 set_result(generate_min_max(id, argument(0), argument(1))); 1917 return true; 1918 } 1919 1920 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) { 1921 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 1922 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 1923 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 1924 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 1925 1926 { 1927 PreserveJVMState pjvms(this); 1928 PreserveReexecuteState preexecs(this); 1929 jvms()->set_should_reexecute(true); 1930 1931 set_control(slow_path); 1932 set_i_o(i_o()); 1933 1934 uncommon_trap(Deoptimization::Reason_intrinsic, 1935 Deoptimization::Action_none); 1936 } 1937 1938 set_control(fast_path); 1939 set_result(math); 1940 } 1941 1942 template <typename OverflowOp> 1943 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 1944 typedef typename OverflowOp::MathOp MathOp; 1945 1946 MathOp* mathOp = new MathOp(arg1, arg2); 1947 Node* operation = _gvn.transform( mathOp ); 1948 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 1949 inline_math_mathExact(operation, ofcheck); 1950 return true; 1951 } 1952 1953 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 1954 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 1955 } 1956 1957 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 1958 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 1959 } 1960 1961 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 1962 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 1963 } 1964 1965 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 1966 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 1967 } 1968 1969 bool LibraryCallKit::inline_math_negateExactI() { 1970 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 1971 } 1972 1973 bool LibraryCallKit::inline_math_negateExactL() { 1974 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 1975 } 1976 1977 bool LibraryCallKit::inline_math_multiplyExactI() { 1978 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 1979 } 1980 1981 bool LibraryCallKit::inline_math_multiplyExactL() { 1982 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 1983 } 1984 1985 bool LibraryCallKit::inline_math_multiplyHigh() { 1986 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2)))); 1987 return true; 1988 } 1989 1990 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() { 1991 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2)))); 1992 return true; 1993 } 1994 1995 Node* 1996 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) { 1997 Node* result_val = nullptr; 1998 switch (id) { 1999 case vmIntrinsics::_min: 2000 case vmIntrinsics::_min_strict: 2001 result_val = _gvn.transform(new MinINode(x0, y0)); 2002 break; 2003 case vmIntrinsics::_max: 2004 case vmIntrinsics::_max_strict: 2005 result_val = _gvn.transform(new MaxINode(x0, y0)); 2006 break; 2007 default: 2008 fatal_unexpected_iid(id); 2009 break; 2010 } 2011 return result_val; 2012 } 2013 2014 inline int 2015 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) { 2016 const TypePtr* base_type = TypePtr::NULL_PTR; 2017 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr(); 2018 if (base_type == nullptr) { 2019 // Unknown type. 2020 return Type::AnyPtr; 2021 } else if (base_type == TypePtr::NULL_PTR) { 2022 // Since this is a null+long form, we have to switch to a rawptr. 2023 base = _gvn.transform(new CastX2PNode(offset)); 2024 offset = MakeConX(0); 2025 return Type::RawPtr; 2026 } else if (base_type->base() == Type::RawPtr) { 2027 return Type::RawPtr; 2028 } else if (base_type->isa_oopptr()) { 2029 // Base is never null => always a heap address. 2030 if (!TypePtr::NULL_PTR->higher_equal(base_type)) { 2031 return Type::OopPtr; 2032 } 2033 // Offset is small => always a heap address. 2034 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2035 if (offset_type != nullptr && 2036 base_type->offset() == 0 && // (should always be?) 2037 offset_type->_lo >= 0 && 2038 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2039 return Type::OopPtr; 2040 } else if (type == T_OBJECT) { 2041 // off heap access to an oop doesn't make any sense. Has to be on 2042 // heap. 2043 return Type::OopPtr; 2044 } 2045 // Otherwise, it might either be oop+off or null+addr. 2046 return Type::AnyPtr; 2047 } else { 2048 // No information: 2049 return Type::AnyPtr; 2050 } 2051 } 2052 2053 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) { 2054 Node* uncasted_base = base; 2055 int kind = classify_unsafe_addr(uncasted_base, offset, type); 2056 if (kind == Type::RawPtr) { 2057 return basic_plus_adr(top(), uncasted_base, offset); 2058 } else if (kind == Type::AnyPtr) { 2059 assert(base == uncasted_base, "unexpected base change"); 2060 if (can_cast) { 2061 if (!_gvn.type(base)->speculative_maybe_null() && 2062 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 2063 // According to profiling, this access is always on 2064 // heap. Casting the base to not null and thus avoiding membars 2065 // around the access should allow better optimizations 2066 Node* null_ctl = top(); 2067 base = null_check_oop(base, &null_ctl, true, true, true); 2068 assert(null_ctl->is_top(), "no null control here"); 2069 return basic_plus_adr(base, offset); 2070 } else if (_gvn.type(base)->speculative_always_null() && 2071 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 2072 // According to profiling, this access is always off 2073 // heap. 2074 base = null_assert(base); 2075 Node* raw_base = _gvn.transform(new CastX2PNode(offset)); 2076 offset = MakeConX(0); 2077 return basic_plus_adr(top(), raw_base, offset); 2078 } 2079 } 2080 // We don't know if it's an on heap or off heap access. Fall back 2081 // to raw memory access. 2082 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM)); 2083 return basic_plus_adr(top(), raw, offset); 2084 } else { 2085 assert(base == uncasted_base, "unexpected base change"); 2086 // We know it's an on heap access so base can't be null 2087 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) { 2088 base = must_be_not_null(base, true); 2089 } 2090 return basic_plus_adr(base, offset); 2091 } 2092 } 2093 2094 //--------------------------inline_number_methods----------------------------- 2095 // inline int Integer.numberOfLeadingZeros(int) 2096 // inline int Long.numberOfLeadingZeros(long) 2097 // 2098 // inline int Integer.numberOfTrailingZeros(int) 2099 // inline int Long.numberOfTrailingZeros(long) 2100 // 2101 // inline int Integer.bitCount(int) 2102 // inline int Long.bitCount(long) 2103 // 2104 // inline char Character.reverseBytes(char) 2105 // inline short Short.reverseBytes(short) 2106 // inline int Integer.reverseBytes(int) 2107 // inline long Long.reverseBytes(long) 2108 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2109 Node* arg = argument(0); 2110 Node* n = nullptr; 2111 switch (id) { 2112 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2113 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2114 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2115 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2116 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2117 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2118 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break; 2119 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break; 2120 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break; 2121 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break; 2122 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break; 2123 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break; 2124 default: fatal_unexpected_iid(id); break; 2125 } 2126 set_result(_gvn.transform(n)); 2127 return true; 2128 } 2129 2130 //--------------------------inline_bitshuffle_methods----------------------------- 2131 // inline int Integer.compress(int, int) 2132 // inline int Integer.expand(int, int) 2133 // inline long Long.compress(long, long) 2134 // inline long Long.expand(long, long) 2135 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) { 2136 Node* n = nullptr; 2137 switch (id) { 2138 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break; 2139 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break; 2140 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break; 2141 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break; 2142 default: fatal_unexpected_iid(id); break; 2143 } 2144 set_result(_gvn.transform(n)); 2145 return true; 2146 } 2147 2148 //--------------------------inline_number_methods----------------------------- 2149 // inline int Integer.compareUnsigned(int, int) 2150 // inline int Long.compareUnsigned(long, long) 2151 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) { 2152 Node* arg1 = argument(0); 2153 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1); 2154 Node* n = nullptr; 2155 switch (id) { 2156 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break; 2157 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break; 2158 default: fatal_unexpected_iid(id); break; 2159 } 2160 set_result(_gvn.transform(n)); 2161 return true; 2162 } 2163 2164 //--------------------------inline_unsigned_divmod_methods----------------------------- 2165 // inline int Integer.divideUnsigned(int, int) 2166 // inline int Integer.remainderUnsigned(int, int) 2167 // inline long Long.divideUnsigned(long, long) 2168 // inline long Long.remainderUnsigned(long, long) 2169 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) { 2170 Node* n = nullptr; 2171 switch (id) { 2172 case vmIntrinsics::_divideUnsigned_i: { 2173 zero_check_int(argument(1)); 2174 // Compile-time detect of null-exception 2175 if (stopped()) { 2176 return true; // keep the graph constructed so far 2177 } 2178 n = new UDivINode(control(), argument(0), argument(1)); 2179 break; 2180 } 2181 case vmIntrinsics::_divideUnsigned_l: { 2182 zero_check_long(argument(2)); 2183 // Compile-time detect of null-exception 2184 if (stopped()) { 2185 return true; // keep the graph constructed so far 2186 } 2187 n = new UDivLNode(control(), argument(0), argument(2)); 2188 break; 2189 } 2190 case vmIntrinsics::_remainderUnsigned_i: { 2191 zero_check_int(argument(1)); 2192 // Compile-time detect of null-exception 2193 if (stopped()) { 2194 return true; // keep the graph constructed so far 2195 } 2196 n = new UModINode(control(), argument(0), argument(1)); 2197 break; 2198 } 2199 case vmIntrinsics::_remainderUnsigned_l: { 2200 zero_check_long(argument(2)); 2201 // Compile-time detect of null-exception 2202 if (stopped()) { 2203 return true; // keep the graph constructed so far 2204 } 2205 n = new UModLNode(control(), argument(0), argument(2)); 2206 break; 2207 } 2208 default: fatal_unexpected_iid(id); break; 2209 } 2210 set_result(_gvn.transform(n)); 2211 return true; 2212 } 2213 2214 //----------------------------inline_unsafe_access---------------------------- 2215 2216 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) { 2217 // Attempt to infer a sharper value type from the offset and base type. 2218 ciKlass* sharpened_klass = nullptr; 2219 bool null_free = false; 2220 2221 // See if it is an instance field, with an object type. 2222 if (alias_type->field() != nullptr) { 2223 if (alias_type->field()->type()->is_klass()) { 2224 sharpened_klass = alias_type->field()->type()->as_klass(); 2225 null_free = alias_type->field()->is_null_free(); 2226 } 2227 } 2228 2229 const TypeOopPtr* result = nullptr; 2230 // See if it is a narrow oop array. 2231 if (adr_type->isa_aryptr()) { 2232 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2233 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr(); 2234 null_free = adr_type->is_aryptr()->is_null_free(); 2235 if (elem_type != nullptr && elem_type->is_loaded()) { 2236 // Sharpen the value type. 2237 result = elem_type; 2238 } 2239 } 2240 } 2241 2242 // The sharpened class might be unloaded if there is no class loader 2243 // contraint in place. 2244 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) { 2245 // Sharpen the value type. 2246 result = TypeOopPtr::make_from_klass(sharpened_klass); 2247 if (null_free) { 2248 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr(); 2249 } 2250 } 2251 if (result != nullptr) { 2252 #ifndef PRODUCT 2253 if (C->print_intrinsics() || C->print_inlining()) { 2254 tty->print(" from base type: "); adr_type->dump(); tty->cr(); 2255 tty->print(" sharpened value: "); result->dump(); tty->cr(); 2256 } 2257 #endif 2258 } 2259 return result; 2260 } 2261 2262 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) { 2263 switch (kind) { 2264 case Relaxed: 2265 return MO_UNORDERED; 2266 case Opaque: 2267 return MO_RELAXED; 2268 case Acquire: 2269 return MO_ACQUIRE; 2270 case Release: 2271 return MO_RELEASE; 2272 case Volatile: 2273 return MO_SEQ_CST; 2274 default: 2275 ShouldNotReachHere(); 2276 return 0; 2277 } 2278 } 2279 2280 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) { 2281 if (callee()->is_static()) return false; // caller must have the capability! 2282 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2283 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2284 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2285 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2286 2287 if (is_reference_type(type)) { 2288 decorators |= ON_UNKNOWN_OOP_REF; 2289 } 2290 2291 if (unaligned) { 2292 decorators |= C2_UNALIGNED; 2293 } 2294 2295 #ifndef PRODUCT 2296 { 2297 ResourceMark rm; 2298 // Check the signatures. 2299 ciSignature* sig = callee()->signature(); 2300 #ifdef ASSERT 2301 if (!is_store) { 2302 // Object getReference(Object base, int/long offset), etc. 2303 BasicType rtype = sig->return_type()->basic_type(); 2304 assert(rtype == type, "getter must return the expected value"); 2305 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments"); 2306 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2307 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2308 } else { 2309 // void putReference(Object base, int/long offset, Object x), etc. 2310 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2311 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments"); 2312 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2313 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2314 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2315 assert(vtype == type, "putter must accept the expected value"); 2316 } 2317 #endif // ASSERT 2318 } 2319 #endif //PRODUCT 2320 2321 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2322 2323 Node* receiver = argument(0); // type: oop 2324 2325 // Build address expression. 2326 Node* heap_base_oop = top(); 2327 2328 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2329 Node* base = argument(1); // type: oop 2330 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2331 Node* offset = argument(2); // type: long 2332 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2333 // to be plain byte offsets, which are also the same as those accepted 2334 // by oopDesc::field_addr. 2335 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2336 "fieldOffset must be byte-scaled"); 2337 2338 ciInlineKlass* inline_klass = nullptr; 2339 if (is_flat) { 2340 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr(); 2341 if (cls == nullptr || cls->const_oop() == nullptr) { 2342 return false; 2343 } 2344 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type(); 2345 if (!mirror_type->is_inlinetype()) { 2346 return false; 2347 } 2348 inline_klass = mirror_type->as_inline_klass(); 2349 } 2350 2351 if (base->is_InlineType()) { 2352 InlineTypeNode* vt = base->as_InlineType(); 2353 if (is_store) { 2354 if (!vt->is_allocated(&_gvn)) { 2355 return false; 2356 } 2357 base = vt->get_oop(); 2358 } else { 2359 if (offset->is_Con()) { 2360 long off = find_long_con(offset, 0); 2361 ciInlineKlass* vk = vt->type()->inline_klass(); 2362 if ((long)(int)off != off || !vk->contains_field_offset(off)) { 2363 return false; 2364 } 2365 2366 ciField* field = vk->get_non_flat_field_by_offset(off); 2367 if (field != nullptr) { 2368 BasicType bt = type2field[field->type()->basic_type()]; 2369 if (bt == T_ARRAY || bt == T_NARROWOOP) { 2370 bt = T_OBJECT; 2371 } 2372 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) { 2373 Node* value = vt->field_value_by_offset(off, false); 2374 if (value->is_InlineType()) { 2375 value = value->as_InlineType()->adjust_scalarization_depth(this); 2376 } 2377 set_result(value); 2378 return true; 2379 } 2380 } 2381 } 2382 { 2383 // Re-execute the unsafe access if allocation triggers deoptimization. 2384 PreserveReexecuteState preexecs(this); 2385 jvms()->set_should_reexecute(true); 2386 vt = vt->buffer(this); 2387 } 2388 base = vt->get_oop(); 2389 } 2390 } 2391 2392 // 32-bit machines ignore the high half! 2393 offset = ConvL2X(offset); 2394 2395 // Save state and restore on bailout 2396 uint old_sp = sp(); 2397 SafePointNode* old_map = clone_map(); 2398 2399 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed); 2400 2401 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) { 2402 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) { 2403 decorators |= IN_NATIVE; // off-heap primitive access 2404 } else { 2405 set_map(old_map); 2406 set_sp(old_sp); 2407 return false; // off-heap oop accesses are not supported 2408 } 2409 } else { 2410 heap_base_oop = base; // on-heap or mixed access 2411 } 2412 2413 // Can base be null? Otherwise, always on-heap access. 2414 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base)); 2415 2416 if (!can_access_non_heap) { 2417 decorators |= IN_HEAP; 2418 } 2419 2420 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr; 2421 2422 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr(); 2423 if (adr_type == TypePtr::NULL_PTR) { 2424 set_map(old_map); 2425 set_sp(old_sp); 2426 return false; // off-heap access with zero address 2427 } 2428 2429 // Try to categorize the address. 2430 Compile::AliasType* alias_type = C->alias_type(adr_type); 2431 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2432 2433 if (alias_type->adr_type() == TypeInstPtr::KLASS || 2434 alias_type->adr_type() == TypeAryPtr::RANGE) { 2435 set_map(old_map); 2436 set_sp(old_sp); 2437 return false; // not supported 2438 } 2439 2440 bool mismatched = false; 2441 BasicType bt = T_ILLEGAL; 2442 ciField* field = nullptr; 2443 if (adr_type->isa_instptr()) { 2444 const TypeInstPtr* instptr = adr_type->is_instptr(); 2445 ciInstanceKlass* k = instptr->instance_klass(); 2446 int off = instptr->offset(); 2447 if (instptr->const_oop() != nullptr && 2448 k == ciEnv::current()->Class_klass() && 2449 instptr->offset() >= (k->size_helper() * wordSize)) { 2450 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass(); 2451 field = k->get_field_by_offset(off, true); 2452 } else { 2453 field = k->get_non_flat_field_by_offset(off); 2454 } 2455 if (field != nullptr) { 2456 bt = type2field[field->type()->basic_type()]; 2457 } 2458 assert(bt == alias_type->basic_type() || is_flat, "should match"); 2459 } else { 2460 bt = alias_type->basic_type(); 2461 } 2462 2463 if (bt != T_ILLEGAL) { 2464 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access"); 2465 if (bt == T_BYTE && adr_type->isa_aryptr()) { 2466 // Alias type doesn't differentiate between byte[] and boolean[]). 2467 // Use address type to get the element type. 2468 bt = adr_type->is_aryptr()->elem()->array_element_basic_type(); 2469 } 2470 if (is_reference_type(bt, true)) { 2471 // accessing an array field with getReference is not a mismatch 2472 bt = T_OBJECT; 2473 } 2474 if ((bt == T_OBJECT) != (type == T_OBJECT)) { 2475 // Don't intrinsify mismatched object accesses 2476 set_map(old_map); 2477 set_sp(old_sp); 2478 return false; 2479 } 2480 mismatched = (bt != type); 2481 } else if (alias_type->adr_type()->isa_oopptr()) { 2482 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched 2483 } 2484 2485 if (is_flat) { 2486 if (adr_type->isa_instptr()) { 2487 if (field == nullptr || field->type() != inline_klass) { 2488 mismatched = true; 2489 } 2490 } else if (adr_type->isa_aryptr()) { 2491 const Type* elem = adr_type->is_aryptr()->elem(); 2492 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) { 2493 mismatched = true; 2494 } 2495 } else { 2496 mismatched = true; 2497 } 2498 if (is_store) { 2499 const Type* val_t = _gvn.type(val); 2500 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) { 2501 set_map(old_map); 2502 set_sp(old_sp); 2503 return false; 2504 } 2505 } 2506 } 2507 2508 destruct_map_clone(old_map); 2509 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched"); 2510 2511 if (mismatched) { 2512 decorators |= C2_MISMATCHED; 2513 } 2514 2515 // First guess at the value type. 2516 const Type *value_type = Type::get_const_basic_type(type); 2517 2518 // Figure out the memory ordering. 2519 decorators |= mo_decorator_for_access_kind(kind); 2520 2521 if (!is_store) { 2522 if (type == T_OBJECT && !is_flat) { 2523 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2524 if (tjp != nullptr) { 2525 value_type = tjp; 2526 } 2527 } 2528 } 2529 2530 receiver = null_check(receiver); 2531 if (stopped()) { 2532 return true; 2533 } 2534 // Heap pointers get a null-check from the interpreter, 2535 // as a courtesy. However, this is not guaranteed by Unsafe, 2536 // and it is not possible to fully distinguish unintended nulls 2537 // from intended ones in this API. 2538 2539 if (!is_store) { 2540 Node* p = nullptr; 2541 // Try to constant fold a load from a constant field 2542 2543 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) { 2544 // final or stable field 2545 p = make_constant_from_field(field, heap_base_oop); 2546 } 2547 2548 if (p == nullptr) { // Could not constant fold the load 2549 if (is_flat) { 2550 if (adr_type->isa_instptr() && !mismatched) { 2551 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass(); 2552 int offset = adr_type->is_instptr()->offset(); 2553 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, holder, offset, decorators); 2554 } else { 2555 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, 0, decorators); 2556 } 2557 } else { 2558 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators); 2559 const TypeOopPtr* ptr = value_type->make_oopptr(); 2560 if (ptr != nullptr && ptr->is_inlinetypeptr()) { 2561 // Load a non-flattened inline type from memory 2562 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null()); 2563 } 2564 } 2565 // Normalize the value returned by getBoolean in the following cases 2566 if (type == T_BOOLEAN && 2567 (mismatched || 2568 heap_base_oop == top() || // - heap_base_oop is null or 2569 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null 2570 // and the unsafe access is made to large offset 2571 // (i.e., larger than the maximum offset necessary for any 2572 // field access) 2573 ) { 2574 IdealKit ideal = IdealKit(this); 2575 #define __ ideal. 2576 IdealVariable normalized_result(ideal); 2577 __ declarations_done(); 2578 __ set(normalized_result, p); 2579 __ if_then(p, BoolTest::ne, ideal.ConI(0)); 2580 __ set(normalized_result, ideal.ConI(1)); 2581 ideal.end_if(); 2582 final_sync(ideal); 2583 p = __ value(normalized_result); 2584 #undef __ 2585 } 2586 } 2587 if (type == T_ADDRESS) { 2588 p = gvn().transform(new CastP2XNode(nullptr, p)); 2589 p = ConvX2UL(p); 2590 } 2591 // The load node has the control of the preceding MemBarCPUOrder. All 2592 // following nodes will have the control of the MemBarCPUOrder inserted at 2593 // the end of this method. So, pushing the load onto the stack at a later 2594 // point is fine. 2595 set_result(p); 2596 } else { 2597 if (bt == T_ADDRESS) { 2598 // Repackage the long as a pointer. 2599 val = ConvL2X(val); 2600 val = gvn().transform(new CastX2PNode(val)); 2601 } 2602 if (is_flat) { 2603 if (adr_type->isa_instptr() && !mismatched) { 2604 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass(); 2605 int offset = adr_type->is_instptr()->offset(); 2606 val->as_InlineType()->store_flat(this, base, base, holder, offset, decorators); 2607 } else { 2608 val->as_InlineType()->store_flat(this, base, adr, nullptr, 0, decorators); 2609 } 2610 } else { 2611 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators); 2612 } 2613 } 2614 2615 if (argument(1)->is_InlineType() && is_store) { 2616 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass()); 2617 value = value->make_larval(this, false); 2618 replace_in_map(argument(1), value); 2619 } 2620 2621 return true; 2622 } 2623 2624 bool LibraryCallKit::inline_unsafe_make_private_buffer() { 2625 Node* receiver = argument(0); 2626 Node* value = argument(1); 2627 if (!value->is_InlineType()) { 2628 return false; 2629 } 2630 2631 receiver = null_check(receiver); 2632 if (stopped()) { 2633 return true; 2634 } 2635 2636 set_result(value->as_InlineType()->make_larval(this, true)); 2637 return true; 2638 } 2639 2640 bool LibraryCallKit::inline_unsafe_finish_private_buffer() { 2641 Node* receiver = argument(0); 2642 Node* buffer = argument(1); 2643 if (!buffer->is_InlineType()) { 2644 return false; 2645 } 2646 InlineTypeNode* vt = buffer->as_InlineType(); 2647 if (!vt->is_allocated(&_gvn)) { 2648 return false; 2649 } 2650 // TODO 8239003 Why is this needed? 2651 if (AllocateNode::Ideal_allocation(vt->get_oop()) == nullptr) { 2652 return false; 2653 } 2654 2655 receiver = null_check(receiver); 2656 if (stopped()) { 2657 return true; 2658 } 2659 2660 set_result(vt->finish_larval(this)); 2661 return true; 2662 } 2663 2664 //----------------------------inline_unsafe_load_store---------------------------- 2665 // This method serves a couple of different customers (depending on LoadStoreKind): 2666 // 2667 // LS_cmp_swap: 2668 // 2669 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x); 2670 // boolean compareAndSetInt( Object o, long offset, int expected, int x); 2671 // boolean compareAndSetLong( Object o, long offset, long expected, long x); 2672 // 2673 // LS_cmp_swap_weak: 2674 // 2675 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x); 2676 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x); 2677 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x); 2678 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x); 2679 // 2680 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x); 2681 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x); 2682 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x); 2683 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x); 2684 // 2685 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x); 2686 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x); 2687 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x); 2688 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x); 2689 // 2690 // LS_cmp_exchange: 2691 // 2692 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x); 2693 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x); 2694 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x); 2695 // 2696 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2697 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2698 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2699 // 2700 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2701 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2702 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2703 // 2704 // LS_get_add: 2705 // 2706 // int getAndAddInt( Object o, long offset, int delta) 2707 // long getAndAddLong(Object o, long offset, long delta) 2708 // 2709 // LS_get_set: 2710 // 2711 // int getAndSet(Object o, long offset, int newValue) 2712 // long getAndSet(Object o, long offset, long newValue) 2713 // Object getAndSet(Object o, long offset, Object newValue) 2714 // 2715 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2716 // This basic scheme here is the same as inline_unsafe_access, but 2717 // differs in enough details that combining them would make the code 2718 // overly confusing. (This is a true fact! I originally combined 2719 // them, but even I was confused by it!) As much code/comments as 2720 // possible are retained from inline_unsafe_access though to make 2721 // the correspondences clearer. - dl 2722 2723 if (callee()->is_static()) return false; // caller must have the capability! 2724 2725 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2726 decorators |= mo_decorator_for_access_kind(access_kind); 2727 2728 #ifndef PRODUCT 2729 BasicType rtype; 2730 { 2731 ResourceMark rm; 2732 // Check the signatures. 2733 ciSignature* sig = callee()->signature(); 2734 rtype = sig->return_type()->basic_type(); 2735 switch(kind) { 2736 case LS_get_add: 2737 case LS_get_set: { 2738 // Check the signatures. 2739 #ifdef ASSERT 2740 assert(rtype == type, "get and set must return the expected type"); 2741 assert(sig->count() == 3, "get and set has 3 arguments"); 2742 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2743 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2744 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2745 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation"); 2746 #endif // ASSERT 2747 break; 2748 } 2749 case LS_cmp_swap: 2750 case LS_cmp_swap_weak: { 2751 // Check the signatures. 2752 #ifdef ASSERT 2753 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2754 assert(sig->count() == 4, "CAS has 4 arguments"); 2755 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2756 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2757 #endif // ASSERT 2758 break; 2759 } 2760 case LS_cmp_exchange: { 2761 // Check the signatures. 2762 #ifdef ASSERT 2763 assert(rtype == type, "CAS must return the expected type"); 2764 assert(sig->count() == 4, "CAS has 4 arguments"); 2765 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2766 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2767 #endif // ASSERT 2768 break; 2769 } 2770 default: 2771 ShouldNotReachHere(); 2772 } 2773 } 2774 #endif //PRODUCT 2775 2776 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2777 2778 // Get arguments: 2779 Node* receiver = nullptr; 2780 Node* base = nullptr; 2781 Node* offset = nullptr; 2782 Node* oldval = nullptr; 2783 Node* newval = nullptr; 2784 switch(kind) { 2785 case LS_cmp_swap: 2786 case LS_cmp_swap_weak: 2787 case LS_cmp_exchange: { 2788 const bool two_slot_type = type2size[type] == 2; 2789 receiver = argument(0); // type: oop 2790 base = argument(1); // type: oop 2791 offset = argument(2); // type: long 2792 oldval = argument(4); // type: oop, int, or long 2793 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2794 break; 2795 } 2796 case LS_get_add: 2797 case LS_get_set: { 2798 receiver = argument(0); // type: oop 2799 base = argument(1); // type: oop 2800 offset = argument(2); // type: long 2801 oldval = nullptr; 2802 newval = argument(4); // type: oop, int, or long 2803 break; 2804 } 2805 default: 2806 ShouldNotReachHere(); 2807 } 2808 2809 // Build field offset expression. 2810 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2811 // to be plain byte offsets, which are also the same as those accepted 2812 // by oopDesc::field_addr. 2813 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2814 // 32-bit machines ignore the high half of long offsets 2815 offset = ConvL2X(offset); 2816 // Save state and restore on bailout 2817 uint old_sp = sp(); 2818 SafePointNode* old_map = clone_map(); 2819 Node* adr = make_unsafe_address(base, offset,type, false); 2820 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2821 2822 Compile::AliasType* alias_type = C->alias_type(adr_type); 2823 BasicType bt = alias_type->basic_type(); 2824 if (bt != T_ILLEGAL && 2825 (is_reference_type(bt) != (type == T_OBJECT))) { 2826 // Don't intrinsify mismatched object accesses. 2827 set_map(old_map); 2828 set_sp(old_sp); 2829 return false; 2830 } 2831 2832 destruct_map_clone(old_map); 2833 2834 // For CAS, unlike inline_unsafe_access, there seems no point in 2835 // trying to refine types. Just use the coarse types here. 2836 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2837 const Type *value_type = Type::get_const_basic_type(type); 2838 2839 switch (kind) { 2840 case LS_get_set: 2841 case LS_cmp_exchange: { 2842 if (type == T_OBJECT) { 2843 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2844 if (tjp != nullptr) { 2845 value_type = tjp; 2846 } 2847 } 2848 break; 2849 } 2850 case LS_cmp_swap: 2851 case LS_cmp_swap_weak: 2852 case LS_get_add: 2853 break; 2854 default: 2855 ShouldNotReachHere(); 2856 } 2857 2858 // Null check receiver. 2859 receiver = null_check(receiver); 2860 if (stopped()) { 2861 return true; 2862 } 2863 2864 int alias_idx = C->get_alias_index(adr_type); 2865 2866 if (is_reference_type(type)) { 2867 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF; 2868 2869 if (oldval != nullptr && oldval->is_InlineType()) { 2870 // Re-execute the unsafe access if allocation triggers deoptimization. 2871 PreserveReexecuteState preexecs(this); 2872 jvms()->set_should_reexecute(true); 2873 oldval = oldval->as_InlineType()->buffer(this)->get_oop(); 2874 } 2875 if (newval != nullptr && newval->is_InlineType()) { 2876 // Re-execute the unsafe access if allocation triggers deoptimization. 2877 PreserveReexecuteState preexecs(this); 2878 jvms()->set_should_reexecute(true); 2879 newval = newval->as_InlineType()->buffer(this)->get_oop(); 2880 } 2881 2882 // Transformation of a value which could be null pointer (CastPP #null) 2883 // could be delayed during Parse (for example, in adjust_map_after_if()). 2884 // Execute transformation here to avoid barrier generation in such case. 2885 if (_gvn.type(newval) == TypePtr::NULL_PTR) 2886 newval = _gvn.makecon(TypePtr::NULL_PTR); 2887 2888 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) { 2889 // Refine the value to a null constant, when it is known to be null 2890 oldval = _gvn.makecon(TypePtr::NULL_PTR); 2891 } 2892 } 2893 2894 Node* result = nullptr; 2895 switch (kind) { 2896 case LS_cmp_exchange: { 2897 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx, 2898 oldval, newval, value_type, type, decorators); 2899 break; 2900 } 2901 case LS_cmp_swap_weak: 2902 decorators |= C2_WEAK_CMPXCHG; 2903 case LS_cmp_swap: { 2904 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx, 2905 oldval, newval, value_type, type, decorators); 2906 break; 2907 } 2908 case LS_get_set: { 2909 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx, 2910 newval, value_type, type, decorators); 2911 break; 2912 } 2913 case LS_get_add: { 2914 result = access_atomic_add_at(base, adr, adr_type, alias_idx, 2915 newval, value_type, type, decorators); 2916 break; 2917 } 2918 default: 2919 ShouldNotReachHere(); 2920 } 2921 2922 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 2923 set_result(result); 2924 return true; 2925 } 2926 2927 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 2928 // Regardless of form, don't allow previous ld/st to move down, 2929 // then issue acquire, release, or volatile mem_bar. 2930 insert_mem_bar(Op_MemBarCPUOrder); 2931 switch(id) { 2932 case vmIntrinsics::_loadFence: 2933 insert_mem_bar(Op_LoadFence); 2934 return true; 2935 case vmIntrinsics::_storeFence: 2936 insert_mem_bar(Op_StoreFence); 2937 return true; 2938 case vmIntrinsics::_storeStoreFence: 2939 insert_mem_bar(Op_StoreStoreFence); 2940 return true; 2941 case vmIntrinsics::_fullFence: 2942 insert_mem_bar(Op_MemBarVolatile); 2943 return true; 2944 default: 2945 fatal_unexpected_iid(id); 2946 return false; 2947 } 2948 } 2949 2950 bool LibraryCallKit::inline_onspinwait() { 2951 insert_mem_bar(Op_OnSpinWait); 2952 return true; 2953 } 2954 2955 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 2956 if (!kls->is_Con()) { 2957 return true; 2958 } 2959 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr(); 2960 if (klsptr == nullptr) { 2961 return true; 2962 } 2963 ciInstanceKlass* ik = klsptr->instance_klass(); 2964 // don't need a guard for a klass that is already initialized 2965 return !ik->is_initialized(); 2966 } 2967 2968 //----------------------------inline_unsafe_writeback0------------------------- 2969 // public native void Unsafe.writeback0(long address) 2970 bool LibraryCallKit::inline_unsafe_writeback0() { 2971 if (!Matcher::has_match_rule(Op_CacheWB)) { 2972 return false; 2973 } 2974 #ifndef PRODUCT 2975 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync"); 2976 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync"); 2977 ciSignature* sig = callee()->signature(); 2978 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!"); 2979 #endif 2980 null_check_receiver(); // null-check, then ignore 2981 Node *addr = argument(1); 2982 addr = new CastX2PNode(addr); 2983 addr = _gvn.transform(addr); 2984 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr); 2985 flush = _gvn.transform(flush); 2986 set_memory(flush, TypeRawPtr::BOTTOM); 2987 return true; 2988 } 2989 2990 //----------------------------inline_unsafe_writeback0------------------------- 2991 // public native void Unsafe.writeback0(long address) 2992 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) { 2993 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) { 2994 return false; 2995 } 2996 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) { 2997 return false; 2998 } 2999 #ifndef PRODUCT 3000 assert(Matcher::has_match_rule(Op_CacheWB), 3001 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB" 3002 : "found match rule for CacheWBPostSync but not CacheWB")); 3003 3004 #endif 3005 null_check_receiver(); // null-check, then ignore 3006 Node *sync; 3007 if (is_pre) { 3008 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 3009 } else { 3010 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 3011 } 3012 sync = _gvn.transform(sync); 3013 set_memory(sync, TypeRawPtr::BOTTOM); 3014 return true; 3015 } 3016 3017 //----------------------------inline_unsafe_allocate--------------------------- 3018 // public native Object Unsafe.allocateInstance(Class<?> cls); 3019 bool LibraryCallKit::inline_unsafe_allocate() { 3020 3021 #if INCLUDE_JVMTI 3022 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 3023 return false; 3024 } 3025 #endif //INCLUDE_JVMTI 3026 3027 if (callee()->is_static()) return false; // caller must have the capability! 3028 3029 null_check_receiver(); // null-check, then ignore 3030 Node* cls = null_check(argument(1)); 3031 if (stopped()) return true; 3032 3033 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0); 3034 kls = null_check(kls); 3035 if (stopped()) return true; // argument was like int.class 3036 3037 #if INCLUDE_JVMTI 3038 // Don't try to access new allocated obj in the intrinsic. 3039 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled. 3040 // Deoptimize and allocate in interpreter instead. 3041 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc)); 3042 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered); 3043 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0))); 3044 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq)); 3045 { 3046 BuildCutout unless(this, tst, PROB_MAX); 3047 uncommon_trap(Deoptimization::Reason_intrinsic, 3048 Deoptimization::Action_make_not_entrant); 3049 } 3050 if (stopped()) { 3051 return true; 3052 } 3053 #endif //INCLUDE_JVMTI 3054 3055 Node* test = nullptr; 3056 if (LibraryCallKit::klass_needs_init_guard(kls)) { 3057 // Note: The argument might still be an illegal value like 3058 // Serializable.class or Object[].class. The runtime will handle it. 3059 // But we must make an explicit check for initialization. 3060 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 3061 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 3062 // can generate code to load it as unsigned byte. 3063 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered); 3064 Node* bits = intcon(InstanceKlass::fully_initialized); 3065 test = _gvn.transform(new SubINode(inst, bits)); 3066 // The 'test' is non-zero if we need to take a slow path. 3067 } 3068 Node* obj = nullptr; 3069 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr(); 3070 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) { 3071 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this); 3072 } else { 3073 obj = new_instance(kls, test); 3074 } 3075 set_result(obj); 3076 return true; 3077 } 3078 3079 //------------------------inline_native_time_funcs-------------- 3080 // inline code for System.currentTimeMillis() and System.nanoTime() 3081 // these have the same type and signature 3082 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 3083 const TypeFunc* tf = OptoRuntime::void_long_Type(); 3084 const TypePtr* no_memory_effects = nullptr; 3085 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 3086 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 3087 #ifdef ASSERT 3088 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 3089 assert(value_top == top(), "second value must be top"); 3090 #endif 3091 set_result(value); 3092 return true; 3093 } 3094 3095 3096 #if INCLUDE_JVMTI 3097 3098 // When notifications are disabled then just update the VTMS transition bit and return. 3099 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol. 3100 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) { 3101 if (!DoJVMTIVirtualThreadTransitions) { 3102 return true; 3103 } 3104 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument 3105 IdealKit ideal(this); 3106 3107 Node* ONE = ideal.ConI(1); 3108 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1))); 3109 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events)); 3110 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw); 3111 3112 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); { 3113 sync_kit(ideal); 3114 // if notifyJvmti enabled then make a call to the given SharedRuntime function 3115 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type(); 3116 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide); 3117 ideal.sync_kit(this); 3118 } ideal.else_(); { 3119 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object 3120 Node* thread = ideal.thread(); 3121 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset())); 3122 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset()); 3123 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr(); 3124 3125 sync_kit(ideal); 3126 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3127 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3128 3129 ideal.sync_kit(this); 3130 } ideal.end_if(); 3131 final_sync(ideal); 3132 3133 return true; 3134 } 3135 3136 // Always update the temporary VTMS transition bit. 3137 bool LibraryCallKit::inline_native_notify_jvmti_hide() { 3138 if (!DoJVMTIVirtualThreadTransitions) { 3139 return true; 3140 } 3141 IdealKit ideal(this); 3142 3143 { 3144 // unconditionally update the temporary VTMS transition bit in current JavaThread 3145 Node* thread = ideal.thread(); 3146 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification 3147 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset())); 3148 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr(); 3149 3150 sync_kit(ideal); 3151 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3152 ideal.sync_kit(this); 3153 } 3154 final_sync(ideal); 3155 3156 return true; 3157 } 3158 3159 // Always update the is_disable_suspend bit. 3160 bool LibraryCallKit::inline_native_notify_jvmti_sync() { 3161 if (!DoJVMTIVirtualThreadTransitions) { 3162 return true; 3163 } 3164 IdealKit ideal(this); 3165 3166 { 3167 // unconditionally update the is_disable_suspend bit in current JavaThread 3168 Node* thread = ideal.thread(); 3169 Node* arg = _gvn.transform(argument(0)); // argument for notification 3170 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset())); 3171 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr(); 3172 3173 sync_kit(ideal); 3174 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3175 ideal.sync_kit(this); 3176 } 3177 final_sync(ideal); 3178 3179 return true; 3180 } 3181 3182 #endif // INCLUDE_JVMTI 3183 3184 #ifdef JFR_HAVE_INTRINSICS 3185 3186 /** 3187 * if oop->klass != null 3188 * // normal class 3189 * epoch = _epoch_state ? 2 : 1 3190 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch { 3191 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts 3192 * } 3193 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path 3194 * else 3195 * // primitive class 3196 * if oop->array_klass != null 3197 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path 3198 * else 3199 * id = LAST_TYPE_ID + 1 // void class path 3200 * if (!signaled) 3201 * signaled = true 3202 */ 3203 bool LibraryCallKit::inline_native_classID() { 3204 Node* cls = argument(0); 3205 3206 IdealKit ideal(this); 3207 #define __ ideal. 3208 IdealVariable result(ideal); __ declarations_done(); 3209 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), 3210 basic_plus_adr(cls, java_lang_Class::klass_offset()), 3211 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 3212 3213 3214 __ if_then(kls, BoolTest::ne, null()); { 3215 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET)); 3216 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw); 3217 3218 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address())); 3219 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw); 3220 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch)); 3221 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT))); 3222 mask = _gvn.transform(new OrLNode(mask, epoch)); 3223 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask)); 3224 3225 float unlikely = PROB_UNLIKELY(0.999); 3226 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); { 3227 sync_kit(ideal); 3228 make_runtime_call(RC_LEAF, 3229 OptoRuntime::class_id_load_barrier_Type(), 3230 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier), 3231 "class id load barrier", 3232 TypePtr::BOTTOM, 3233 kls); 3234 ideal.sync_kit(this); 3235 } __ end_if(); 3236 3237 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)))); 3238 } __ else_(); { 3239 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), 3240 basic_plus_adr(cls, java_lang_Class::array_klass_offset()), 3241 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 3242 __ if_then(array_kls, BoolTest::ne, null()); { 3243 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET)); 3244 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw); 3245 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))); 3246 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1)))); 3247 } __ else_(); { 3248 // void class case 3249 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1))); 3250 } __ end_if(); 3251 3252 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address())); 3253 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire); 3254 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); { 3255 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true); 3256 } __ end_if(); 3257 } __ end_if(); 3258 3259 final_sync(ideal); 3260 set_result(ideal.value(result)); 3261 #undef __ 3262 return true; 3263 } 3264 3265 //------------------------inline_native_jvm_commit------------------ 3266 bool LibraryCallKit::inline_native_jvm_commit() { 3267 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3268 3269 // Save input memory and i_o state. 3270 Node* input_memory_state = reset_memory(); 3271 set_all_memory(input_memory_state); 3272 Node* input_io_state = i_o(); 3273 3274 // TLS. 3275 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3276 // Jfr java buffer. 3277 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))))); 3278 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered)); 3279 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))))); 3280 3281 // Load the current value of the notified field in the JfrThreadLocal. 3282 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR)); 3283 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered); 3284 3285 // Test for notification. 3286 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1))); 3287 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq)); 3288 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN); 3289 3290 // True branch, is notified. 3291 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified)); 3292 set_control(is_notified); 3293 3294 // Reset notified state. 3295 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered); 3296 3297 // 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. 3298 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered)); 3299 // Convert the machine-word to a long. 3300 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X)); 3301 3302 // False branch, not notified. 3303 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified)); 3304 set_control(not_notified); 3305 set_all_memory(input_memory_state); 3306 3307 // Arg is the next position as a long. 3308 Node* arg = argument(0); 3309 // Convert long to machine-word. 3310 Node* next_pos_X = _gvn.transform(ConvL2X(arg)); 3311 3312 // Store the next_position to the underlying jfr java buffer. 3313 Node* commit_memory; 3314 #ifdef _LP64 3315 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release); 3316 #else 3317 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release); 3318 #endif 3319 3320 // 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. 3321 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))))); 3322 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered); 3323 Node* lease_constant = _gvn.transform(_gvn.intcon(4)); 3324 3325 // And flags with lease constant. 3326 Node* lease = _gvn.transform(new AndINode(flags, lease_constant)); 3327 3328 // Branch on lease to conditionalize returning the leased java buffer. 3329 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant)); 3330 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq)); 3331 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN); 3332 3333 // False branch, not a lease. 3334 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease)); 3335 3336 // True branch, is lease. 3337 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease)); 3338 set_control(is_lease); 3339 3340 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop. 3341 Node* call_return_lease = make_runtime_call(RC_NO_LEAF, 3342 OptoRuntime::void_void_Type(), 3343 StubRoutines::jfr_return_lease(), 3344 "return_lease", TypePtr::BOTTOM); 3345 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control)); 3346 3347 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT); 3348 record_for_igvn(lease_compare_rgn); 3349 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3350 record_for_igvn(lease_compare_mem); 3351 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO); 3352 record_for_igvn(lease_compare_io); 3353 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG); 3354 record_for_igvn(lease_result_value); 3355 3356 // Update control and phi nodes. 3357 lease_compare_rgn->init_req(_true_path, call_return_lease_control); 3358 lease_compare_rgn->init_req(_false_path, not_lease); 3359 3360 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3361 lease_compare_mem->init_req(_false_path, commit_memory); 3362 3363 lease_compare_io->init_req(_true_path, i_o()); 3364 lease_compare_io->init_req(_false_path, input_io_state); 3365 3366 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0. 3367 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position. 3368 3369 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3370 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3371 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO); 3372 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG); 3373 3374 // Update control and phi nodes. 3375 result_rgn->init_req(_true_path, is_notified); 3376 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn)); 3377 3378 result_mem->init_req(_true_path, notified_reset_memory); 3379 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem)); 3380 3381 result_io->init_req(_true_path, input_io_state); 3382 result_io->init_req(_false_path, _gvn.transform(lease_compare_io)); 3383 3384 result_value->init_req(_true_path, current_pos); 3385 result_value->init_req(_false_path, _gvn.transform(lease_result_value)); 3386 3387 // Set output state. 3388 set_control(_gvn.transform(result_rgn)); 3389 set_all_memory(_gvn.transform(result_mem)); 3390 set_i_o(_gvn.transform(result_io)); 3391 set_result(result_rgn, result_value); 3392 return true; 3393 } 3394 3395 /* 3396 * The intrinsic is a model of this pseudo-code: 3397 * 3398 * JfrThreadLocal* const tl = Thread::jfr_thread_local() 3399 * jobject h_event_writer = tl->java_event_writer(); 3400 * if (h_event_writer == nullptr) { 3401 * return nullptr; 3402 * } 3403 * oop threadObj = Thread::threadObj(); 3404 * oop vthread = java_lang_Thread::vthread(threadObj); 3405 * traceid tid; 3406 * bool excluded; 3407 * if (vthread != threadObj) { // i.e. current thread is virtual 3408 * tid = java_lang_Thread::tid(vthread); 3409 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread); 3410 * excluded = vthread_epoch_raw & excluded_mask; 3411 * if (!excluded) { 3412 * traceid current_epoch = JfrTraceIdEpoch::current_generation(); 3413 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask; 3414 * if (vthread_epoch != current_epoch) { 3415 * write_checkpoint(); 3416 * } 3417 * } 3418 * } else { 3419 * tid = java_lang_Thread::tid(threadObj); 3420 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj); 3421 * excluded = thread_epoch_raw & excluded_mask; 3422 * } 3423 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer); 3424 * traceid tid_in_event_writer = getField(event_writer, "threadID"); 3425 * if (tid_in_event_writer != tid) { 3426 * setField(event_writer, "threadID", tid); 3427 * setField(event_writer, "excluded", excluded); 3428 * } 3429 * return event_writer 3430 */ 3431 bool LibraryCallKit::inline_native_getEventWriter() { 3432 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3433 3434 // Save input memory and i_o state. 3435 Node* input_memory_state = reset_memory(); 3436 set_all_memory(input_memory_state); 3437 Node* input_io_state = i_o(); 3438 3439 Node* excluded_mask = _gvn.intcon(32768); 3440 Node* epoch_mask = _gvn.intcon(32767); 3441 3442 // TLS 3443 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3444 3445 // Load the address of java event writer jobject handle from the jfr_thread_local structure. 3446 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)); 3447 3448 // Load the eventwriter jobject handle. 3449 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered); 3450 3451 // Null check the jobject handle. 3452 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null())); 3453 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne)); 3454 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN); 3455 3456 // False path, jobj is null. 3457 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null)); 3458 3459 // True path, jobj is not null. 3460 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null)); 3461 3462 set_control(jobj_is_not_null); 3463 3464 // Load the threadObj for the CarrierThread. 3465 Node* threadObj = generate_current_thread(tls_ptr); 3466 3467 // Load the vthread. 3468 Node* vthread = generate_virtual_thread(tls_ptr); 3469 3470 // If vthread != threadObj, this is a virtual thread. 3471 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj)); 3472 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne)); 3473 IfNode* iff_vthread_not_equal_threadObj = 3474 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN); 3475 3476 // False branch, fallback to threadObj. 3477 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj)); 3478 set_control(vthread_equal_threadObj); 3479 3480 // Load the tid field from the vthread object. 3481 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J"); 3482 3483 // Load the raw epoch value from the threadObj. 3484 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset()); 3485 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR, 3486 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3487 3488 // Mask off the excluded information from the epoch. 3489 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask)); 3490 3491 // True branch, this is a virtual thread. 3492 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj)); 3493 set_control(vthread_not_equal_threadObj); 3494 3495 // Load the tid field from the vthread object. 3496 Node* vthread_tid = load_field_from_object(vthread, "tid", "J"); 3497 3498 // Load the raw epoch value from the vthread. 3499 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset()); 3500 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR, 3501 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3502 3503 // Mask off the excluded information from the epoch. 3504 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask))); 3505 3506 // Branch on excluded to conditionalize updating the epoch for the virtual thread. 3507 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask))); 3508 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne)); 3509 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN); 3510 3511 // False branch, vthread is excluded, no need to write epoch info. 3512 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded)); 3513 3514 // True branch, vthread is included, update epoch info. 3515 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded)); 3516 set_control(included); 3517 3518 // Get epoch value. 3519 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask))); 3520 3521 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR. 3522 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address())); 3523 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered); 3524 3525 // Compare the epoch in the vthread to the current epoch generation. 3526 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch)); 3527 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne)); 3528 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN); 3529 3530 // False path, epoch is equal, checkpoint information is valid. 3531 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal)); 3532 3533 // True path, epoch is not equal, write a checkpoint for the vthread. 3534 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal)); 3535 3536 set_control(epoch_is_not_equal); 3537 3538 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch. 3539 // The call also updates the native thread local thread id and the vthread with the current epoch. 3540 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF, 3541 OptoRuntime::jfr_write_checkpoint_Type(), 3542 StubRoutines::jfr_write_checkpoint(), 3543 "write_checkpoint", TypePtr::BOTTOM); 3544 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control)); 3545 3546 // vthread epoch != current epoch 3547 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT); 3548 record_for_igvn(epoch_compare_rgn); 3549 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3550 record_for_igvn(epoch_compare_mem); 3551 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO); 3552 record_for_igvn(epoch_compare_io); 3553 3554 // Update control and phi nodes. 3555 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control); 3556 epoch_compare_rgn->init_req(_false_path, epoch_is_equal); 3557 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3558 epoch_compare_mem->init_req(_false_path, input_memory_state); 3559 epoch_compare_io->init_req(_true_path, i_o()); 3560 epoch_compare_io->init_req(_false_path, input_io_state); 3561 3562 // excluded != true 3563 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT); 3564 record_for_igvn(exclude_compare_rgn); 3565 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3566 record_for_igvn(exclude_compare_mem); 3567 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO); 3568 record_for_igvn(exclude_compare_io); 3569 3570 // Update control and phi nodes. 3571 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn)); 3572 exclude_compare_rgn->init_req(_false_path, excluded); 3573 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem)); 3574 exclude_compare_mem->init_req(_false_path, input_memory_state); 3575 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io)); 3576 exclude_compare_io->init_req(_false_path, input_io_state); 3577 3578 // vthread != threadObj 3579 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT); 3580 record_for_igvn(vthread_compare_rgn); 3581 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3582 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO); 3583 record_for_igvn(vthread_compare_io); 3584 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG); 3585 record_for_igvn(tid); 3586 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL); 3587 record_for_igvn(exclusion); 3588 3589 // Update control and phi nodes. 3590 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn)); 3591 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj); 3592 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem)); 3593 vthread_compare_mem->init_req(_false_path, input_memory_state); 3594 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io)); 3595 vthread_compare_io->init_req(_false_path, input_io_state); 3596 tid->init_req(_true_path, _gvn.transform(vthread_tid)); 3597 tid->init_req(_false_path, _gvn.transform(thread_obj_tid)); 3598 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded)); 3599 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded)); 3600 3601 // Update branch state. 3602 set_control(_gvn.transform(vthread_compare_rgn)); 3603 set_all_memory(_gvn.transform(vthread_compare_mem)); 3604 set_i_o(_gvn.transform(vthread_compare_io)); 3605 3606 // Load the event writer oop by dereferencing the jobject handle. 3607 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter")); 3608 assert(klass_EventWriter->is_loaded(), "invariant"); 3609 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass(); 3610 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter); 3611 const TypeOopPtr* const xtype = aklass->as_instance_type(); 3612 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global))); 3613 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD); 3614 3615 // Load the current thread id from the event writer object. 3616 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J"); 3617 // Get the field offset to, conditionally, store an updated tid value later. 3618 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false); 3619 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr(); 3620 // Get the field offset to, conditionally, store an updated exclusion value later. 3621 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false); 3622 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr(); 3623 3624 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT); 3625 record_for_igvn(event_writer_tid_compare_rgn); 3626 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3627 record_for_igvn(event_writer_tid_compare_mem); 3628 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO); 3629 record_for_igvn(event_writer_tid_compare_io); 3630 3631 // Compare the current tid from the thread object to what is currently stored in the event writer object. 3632 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid))); 3633 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne)); 3634 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN); 3635 3636 // False path, tids are the same. 3637 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal)); 3638 3639 // True path, tid is not equal, need to update the tid in the event writer. 3640 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal)); 3641 record_for_igvn(tid_is_not_equal); 3642 3643 // Store the exclusion state to the event writer. 3644 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered); 3645 3646 // Store the tid to the event writer. 3647 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered); 3648 3649 // Update control and phi nodes. 3650 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal); 3651 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal); 3652 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3653 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem)); 3654 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o())); 3655 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io)); 3656 3657 // Result of top level CFG, Memory, IO and Value. 3658 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3659 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3660 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO); 3661 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM); 3662 3663 // Result control. 3664 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn)); 3665 result_rgn->init_req(_false_path, jobj_is_null); 3666 3667 // Result memory. 3668 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem)); 3669 result_mem->init_req(_false_path, _gvn.transform(input_memory_state)); 3670 3671 // Result IO. 3672 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io)); 3673 result_io->init_req(_false_path, _gvn.transform(input_io_state)); 3674 3675 // Result value. 3676 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop 3677 result_value->init_req(_false_path, null()); // return null 3678 3679 // Set output state. 3680 set_control(_gvn.transform(result_rgn)); 3681 set_all_memory(_gvn.transform(result_mem)); 3682 set_i_o(_gvn.transform(result_io)); 3683 set_result(result_rgn, result_value); 3684 return true; 3685 } 3686 3687 /* 3688 * The intrinsic is a model of this pseudo-code: 3689 * 3690 * JfrThreadLocal* const tl = thread->jfr_thread_local(); 3691 * if (carrierThread != thread) { // is virtual thread 3692 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread); 3693 * bool excluded = vthread_epoch_raw & excluded_mask; 3694 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread)); 3695 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded); 3696 * if (!excluded) { 3697 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask; 3698 * Atomic::store(&tl->_vthread_epoch, vthread_epoch); 3699 * } 3700 * Atomic::release_store(&tl->_vthread, true); 3701 * return; 3702 * } 3703 * Atomic::release_store(&tl->_vthread, false); 3704 */ 3705 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) { 3706 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3707 3708 Node* input_memory_state = reset_memory(); 3709 set_all_memory(input_memory_state); 3710 3711 Node* excluded_mask = _gvn.intcon(32768); 3712 Node* epoch_mask = _gvn.intcon(32767); 3713 3714 Node* const carrierThread = generate_current_thread(jt); 3715 // If thread != carrierThread, this is a virtual thread. 3716 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread)); 3717 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne)); 3718 IfNode* iff_thread_not_equal_carrierThread = 3719 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN); 3720 3721 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR)); 3722 3723 // False branch, is carrierThread. 3724 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread)); 3725 // Store release 3726 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true); 3727 3728 set_all_memory(input_memory_state); 3729 3730 // True branch, is virtual thread. 3731 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread)); 3732 set_control(thread_not_equal_carrierThread); 3733 3734 // Load the raw epoch value from the vthread. 3735 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset()); 3736 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR, 3737 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3738 3739 // Mask off the excluded information from the epoch. 3740 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask))); 3741 3742 // Load the tid field from the thread. 3743 Node* tid = load_field_from_object(thread, "tid", "J"); 3744 3745 // Store the vthread tid to the jfr thread local. 3746 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR)); 3747 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true); 3748 3749 // Branch is_excluded to conditionalize updating the epoch . 3750 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask))); 3751 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq)); 3752 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN); 3753 3754 // True branch, vthread is excluded, no need to write epoch info. 3755 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded)); 3756 set_control(excluded); 3757 Node* vthread_is_excluded = _gvn.intcon(1); 3758 3759 // False branch, vthread is included, update epoch info. 3760 Node* included = _gvn.transform(new IfFalseNode(iff_excluded)); 3761 set_control(included); 3762 Node* vthread_is_included = _gvn.intcon(0); 3763 3764 // Get epoch value. 3765 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask))); 3766 3767 // Store the vthread epoch to the jfr thread local. 3768 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR)); 3769 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true); 3770 3771 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT); 3772 record_for_igvn(excluded_rgn); 3773 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM); 3774 record_for_igvn(excluded_mem); 3775 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL); 3776 record_for_igvn(exclusion); 3777 3778 // Merge the excluded control and memory. 3779 excluded_rgn->init_req(_true_path, excluded); 3780 excluded_rgn->init_req(_false_path, included); 3781 excluded_mem->init_req(_true_path, tid_memory); 3782 excluded_mem->init_req(_false_path, included_memory); 3783 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded)); 3784 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included)); 3785 3786 // Set intermediate state. 3787 set_control(_gvn.transform(excluded_rgn)); 3788 set_all_memory(excluded_mem); 3789 3790 // Store the vthread exclusion state to the jfr thread local. 3791 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR)); 3792 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true); 3793 3794 // Store release 3795 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true); 3796 3797 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT); 3798 record_for_igvn(thread_compare_rgn); 3799 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3800 record_for_igvn(thread_compare_mem); 3801 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL); 3802 record_for_igvn(vthread); 3803 3804 // Merge the thread_compare control and memory. 3805 thread_compare_rgn->init_req(_true_path, control()); 3806 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread); 3807 thread_compare_mem->init_req(_true_path, vthread_true_memory); 3808 thread_compare_mem->init_req(_false_path, vthread_false_memory); 3809 3810 // Set output state. 3811 set_control(_gvn.transform(thread_compare_rgn)); 3812 set_all_memory(_gvn.transform(thread_compare_mem)); 3813 } 3814 3815 #endif // JFR_HAVE_INTRINSICS 3816 3817 //------------------------inline_native_currentCarrierThread------------------ 3818 bool LibraryCallKit::inline_native_currentCarrierThread() { 3819 Node* junk = nullptr; 3820 set_result(generate_current_thread(junk)); 3821 return true; 3822 } 3823 3824 //------------------------inline_native_currentThread------------------ 3825 bool LibraryCallKit::inline_native_currentThread() { 3826 Node* junk = nullptr; 3827 set_result(generate_virtual_thread(junk)); 3828 return true; 3829 } 3830 3831 //------------------------inline_native_setVthread------------------ 3832 bool LibraryCallKit::inline_native_setCurrentThread() { 3833 assert(C->method()->changes_current_thread(), 3834 "method changes current Thread but is not annotated ChangesCurrentThread"); 3835 Node* arr = argument(1); 3836 Node* thread = _gvn.transform(new ThreadLocalNode()); 3837 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset())); 3838 Node* thread_obj_handle 3839 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered); 3840 thread_obj_handle = _gvn.transform(thread_obj_handle); 3841 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr(); 3842 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED); 3843 JFR_ONLY(extend_setCurrentThread(thread, arr);) 3844 return true; 3845 } 3846 3847 const Type* LibraryCallKit::scopedValueCache_type() { 3848 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass()); 3849 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass()); 3850 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true); 3851 3852 // Because we create the scopedValue cache lazily we have to make the 3853 // type of the result BotPTR. 3854 bool xk = etype->klass_is_exact(); 3855 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0)); 3856 return objects_type; 3857 } 3858 3859 Node* LibraryCallKit::scopedValueCache_helper() { 3860 Node* thread = _gvn.transform(new ThreadLocalNode()); 3861 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset())); 3862 // We cannot use immutable_memory() because we might flip onto a 3863 // different carrier thread, at which point we'll need to use that 3864 // carrier thread's cache. 3865 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(), 3866 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)); 3867 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered); 3868 } 3869 3870 //------------------------inline_native_scopedValueCache------------------ 3871 bool LibraryCallKit::inline_native_scopedValueCache() { 3872 Node* cache_obj_handle = scopedValueCache_helper(); 3873 const Type* objects_type = scopedValueCache_type(); 3874 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE)); 3875 3876 return true; 3877 } 3878 3879 //------------------------inline_native_setScopedValueCache------------------ 3880 bool LibraryCallKit::inline_native_setScopedValueCache() { 3881 Node* arr = argument(0); 3882 Node* cache_obj_handle = scopedValueCache_helper(); 3883 const Type* objects_type = scopedValueCache_type(); 3884 3885 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr(); 3886 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED); 3887 3888 return true; 3889 } 3890 3891 //-----------------------load_klass_from_mirror_common------------------------- 3892 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3893 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3894 // and branch to the given path on the region. 3895 // If never_see_null, take an uncommon trap on null, so we can optimistically 3896 // compile for the non-null case. 3897 // If the region is null, force never_see_null = true. 3898 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3899 bool never_see_null, 3900 RegionNode* region, 3901 int null_path, 3902 int offset) { 3903 if (region == nullptr) never_see_null = true; 3904 Node* p = basic_plus_adr(mirror, offset); 3905 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL; 3906 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3907 Node* null_ctl = top(); 3908 kls = null_check_oop(kls, &null_ctl, never_see_null); 3909 if (region != nullptr) { 3910 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3911 region->init_req(null_path, null_ctl); 3912 } else { 3913 assert(null_ctl == top(), "no loose ends"); 3914 } 3915 return kls; 3916 } 3917 3918 //--------------------(inline_native_Class_query helpers)--------------------- 3919 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER. 3920 // Fall through if (mods & mask) == bits, take the guard otherwise. 3921 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 3922 // Branch around if the given klass has the given modifier bit set. 3923 // Like generate_guard, adds a new path onto the region. 3924 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3925 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered); 3926 Node* mask = intcon(modifier_mask); 3927 Node* bits = intcon(modifier_bits); 3928 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3929 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3930 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3931 return generate_fair_guard(bol, region); 3932 } 3933 3934 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3935 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region); 3936 } 3937 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) { 3938 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region); 3939 } 3940 3941 //-------------------------inline_native_Class_query------------------- 3942 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 3943 const Type* return_type = TypeInt::BOOL; 3944 Node* prim_return_value = top(); // what happens if it's a primitive class? 3945 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3946 bool expect_prim = false; // most of these guys expect to work on refs 3947 3948 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 3949 3950 Node* mirror = argument(0); 3951 Node* obj = top(); 3952 3953 switch (id) { 3954 case vmIntrinsics::_isInstance: 3955 // nothing is an instance of a primitive type 3956 prim_return_value = intcon(0); 3957 obj = argument(1); 3958 break; 3959 case vmIntrinsics::_getModifiers: 3960 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3961 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line"); 3962 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin); 3963 break; 3964 case vmIntrinsics::_isInterface: 3965 prim_return_value = intcon(0); 3966 break; 3967 case vmIntrinsics::_isArray: 3968 prim_return_value = intcon(0); 3969 expect_prim = true; // cf. ObjectStreamClass.getClassSignature 3970 break; 3971 case vmIntrinsics::_isPrimitive: 3972 prim_return_value = intcon(1); 3973 expect_prim = true; // obviously 3974 break; 3975 case vmIntrinsics::_isHidden: 3976 prim_return_value = intcon(0); 3977 break; 3978 case vmIntrinsics::_getSuperclass: 3979 prim_return_value = null(); 3980 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 3981 break; 3982 case vmIntrinsics::_getClassAccessFlags: 3983 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3984 return_type = TypeInt::INT; // not bool! 6297094 3985 break; 3986 default: 3987 fatal_unexpected_iid(id); 3988 break; 3989 } 3990 3991 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3992 if (mirror_con == nullptr) return false; // cannot happen? 3993 3994 #ifndef PRODUCT 3995 if (C->print_intrinsics() || C->print_inlining()) { 3996 ciType* k = mirror_con->java_mirror_type(); 3997 if (k) { 3998 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 3999 k->print_name(); 4000 tty->cr(); 4001 } 4002 } 4003 #endif 4004 4005 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 4006 RegionNode* region = new RegionNode(PATH_LIMIT); 4007 record_for_igvn(region); 4008 PhiNode* phi = new PhiNode(region, return_type); 4009 4010 // The mirror will never be null of Reflection.getClassAccessFlags, however 4011 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 4012 // if it is. See bug 4774291. 4013 4014 // For Reflection.getClassAccessFlags(), the null check occurs in 4015 // the wrong place; see inline_unsafe_access(), above, for a similar 4016 // situation. 4017 mirror = null_check(mirror); 4018 // If mirror or obj is dead, only null-path is taken. 4019 if (stopped()) return true; 4020 4021 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 4022 4023 // Now load the mirror's klass metaobject, and null-check it. 4024 // Side-effects region with the control path if the klass is null. 4025 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 4026 // If kls is null, we have a primitive mirror. 4027 phi->init_req(_prim_path, prim_return_value); 4028 if (stopped()) { set_result(region, phi); return true; } 4029 bool safe_for_replace = (region->in(_prim_path) == top()); 4030 4031 Node* p; // handy temp 4032 Node* null_ctl; 4033 4034 // Now that we have the non-null klass, we can perform the real query. 4035 // For constant classes, the query will constant-fold in LoadNode::Value. 4036 Node* query_value = top(); 4037 switch (id) { 4038 case vmIntrinsics::_isInstance: 4039 // nothing is an instance of a primitive type 4040 query_value = gen_instanceof(obj, kls, safe_for_replace); 4041 break; 4042 4043 case vmIntrinsics::_getModifiers: 4044 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset())); 4045 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered); 4046 break; 4047 4048 case vmIntrinsics::_isInterface: 4049 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 4050 if (generate_interface_guard(kls, region) != nullptr) 4051 // A guard was added. If the guard is taken, it was an interface. 4052 phi->add_req(intcon(1)); 4053 // If we fall through, it's a plain class. 4054 query_value = intcon(0); 4055 break; 4056 4057 case vmIntrinsics::_isArray: 4058 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.) 4059 if (generate_array_guard(kls, region) != nullptr) 4060 // A guard was added. If the guard is taken, it was an array. 4061 phi->add_req(intcon(1)); 4062 // If we fall through, it's a plain class. 4063 query_value = intcon(0); 4064 break; 4065 4066 case vmIntrinsics::_isPrimitive: 4067 query_value = intcon(0); // "normal" path produces false 4068 break; 4069 4070 case vmIntrinsics::_isHidden: 4071 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.) 4072 if (generate_hidden_class_guard(kls, region) != nullptr) 4073 // A guard was added. If the guard is taken, it was an hidden class. 4074 phi->add_req(intcon(1)); 4075 // If we fall through, it's a plain class. 4076 query_value = intcon(0); 4077 break; 4078 4079 4080 case vmIntrinsics::_getSuperclass: 4081 // The rules here are somewhat unfortunate, but we can still do better 4082 // with random logic than with a JNI call. 4083 // Interfaces store null or Object as _super, but must report null. 4084 // Arrays store an intermediate super as _super, but must report Object. 4085 // Other types can report the actual _super. 4086 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 4087 if (generate_interface_guard(kls, region) != nullptr) 4088 // A guard was added. If the guard is taken, it was an interface. 4089 phi->add_req(null()); 4090 if (generate_array_guard(kls, region) != nullptr) 4091 // A guard was added. If the guard is taken, it was an array. 4092 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 4093 // If we fall through, it's a plain class. Get its _super. 4094 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 4095 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 4096 null_ctl = top(); 4097 kls = null_check_oop(kls, &null_ctl); 4098 if (null_ctl != top()) { 4099 // If the guard is taken, Object.superClass is null (both klass and mirror). 4100 region->add_req(null_ctl); 4101 phi ->add_req(null()); 4102 } 4103 if (!stopped()) { 4104 query_value = load_mirror_from_klass(kls); 4105 } 4106 break; 4107 4108 case vmIntrinsics::_getClassAccessFlags: 4109 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 4110 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered); 4111 break; 4112 4113 default: 4114 fatal_unexpected_iid(id); 4115 break; 4116 } 4117 4118 // Fall-through is the normal case of a query to a real class. 4119 phi->init_req(1, query_value); 4120 region->init_req(1, control()); 4121 4122 C->set_has_split_ifs(true); // Has chance for split-if optimization 4123 set_result(region, phi); 4124 return true; 4125 } 4126 4127 4128 //-------------------------inline_Class_cast------------------- 4129 bool LibraryCallKit::inline_Class_cast() { 4130 Node* mirror = argument(0); // Class 4131 Node* obj = argument(1); 4132 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 4133 if (mirror_con == nullptr) { 4134 return false; // dead path (mirror->is_top()). 4135 } 4136 if (obj == nullptr || obj->is_top()) { 4137 return false; // dead path 4138 } 4139 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr(); 4140 4141 // First, see if Class.cast() can be folded statically. 4142 // java_mirror_type() returns non-null for compile-time Class constants. 4143 bool is_null_free_array = false; 4144 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array); 4145 if (tm != nullptr && tm->is_klass() && 4146 tp != nullptr) { 4147 if (!tp->is_loaded()) { 4148 // Don't use intrinsic when class is not loaded. 4149 return false; 4150 } else { 4151 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces); 4152 if (is_null_free_array) { 4153 tklass = tklass->is_aryklassptr()->cast_to_null_free(); 4154 } 4155 int static_res = C->static_subtype_check(tklass, tp->as_klass_type()); 4156 if (static_res == Compile::SSC_always_true) { 4157 // isInstance() is true - fold the code. 4158 set_result(obj); 4159 return true; 4160 } else if (static_res == Compile::SSC_always_false) { 4161 // Don't use intrinsic, have to throw ClassCastException. 4162 // If the reference is null, the non-intrinsic bytecode will 4163 // be optimized appropriately. 4164 return false; 4165 } 4166 } 4167 } 4168 4169 // Bailout intrinsic and do normal inlining if exception path is frequent. 4170 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 4171 return false; 4172 } 4173 4174 // Generate dynamic checks. 4175 // Class.cast() is java implementation of _checkcast bytecode. 4176 // Do checkcast (Parse::do_checkcast()) optimizations here. 4177 4178 mirror = null_check(mirror); 4179 // If mirror is dead, only null-path is taken. 4180 if (stopped()) { 4181 return true; 4182 } 4183 4184 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive). 4185 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT }; 4186 RegionNode* region = new RegionNode(PATH_LIMIT); 4187 record_for_igvn(region); 4188 4189 // Now load the mirror's klass metaobject, and null-check it. 4190 // If kls is null, we have a primitive mirror and 4191 // nothing is an instance of a primitive type. 4192 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 4193 4194 Node* res = top(); 4195 Node* io = i_o(); 4196 Node* mem = merged_memory(); 4197 if (!stopped()) { 4198 4199 Node* bad_type_ctrl = top(); 4200 // Do checkcast optimizations. 4201 res = gen_checkcast(obj, kls, &bad_type_ctrl); 4202 region->init_req(_bad_type_path, bad_type_ctrl); 4203 } 4204 if (region->in(_prim_path) != top() || 4205 region->in(_bad_type_path) != top() || 4206 region->in(_npe_path) != top()) { 4207 // Let Interpreter throw ClassCastException. 4208 PreserveJVMState pjvms(this); 4209 set_control(_gvn.transform(region)); 4210 // Set IO and memory because gen_checkcast may override them when buffering inline types 4211 set_i_o(io); 4212 set_all_memory(mem); 4213 uncommon_trap(Deoptimization::Reason_intrinsic, 4214 Deoptimization::Action_maybe_recompile); 4215 } 4216 if (!stopped()) { 4217 set_result(res); 4218 } 4219 return true; 4220 } 4221 4222 4223 //--------------------------inline_native_subtype_check------------------------ 4224 // This intrinsic takes the JNI calls out of the heart of 4225 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 4226 bool LibraryCallKit::inline_native_subtype_check() { 4227 // Pull both arguments off the stack. 4228 Node* args[2]; // two java.lang.Class mirrors: superc, subc 4229 args[0] = argument(0); 4230 args[1] = argument(1); 4231 Node* klasses[2]; // corresponding Klasses: superk, subk 4232 klasses[0] = klasses[1] = top(); 4233 4234 enum { 4235 // A full decision tree on {superc is prim, subc is prim}: 4236 _prim_0_path = 1, // {P,N} => false 4237 // {P,P} & superc!=subc => false 4238 _prim_same_path, // {P,P} & superc==subc => true 4239 _prim_1_path, // {N,P} => false 4240 _ref_subtype_path, // {N,N} & subtype check wins => true 4241 _both_ref_path, // {N,N} & subtype check loses => false 4242 PATH_LIMIT 4243 }; 4244 4245 RegionNode* region = new RegionNode(PATH_LIMIT); 4246 RegionNode* prim_region = new RegionNode(2); 4247 Node* phi = new PhiNode(region, TypeInt::BOOL); 4248 record_for_igvn(region); 4249 record_for_igvn(prim_region); 4250 4251 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 4252 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL; 4253 int class_klass_offset = java_lang_Class::klass_offset(); 4254 4255 // First null-check both mirrors and load each mirror's klass metaobject. 4256 int which_arg; 4257 for (which_arg = 0; which_arg <= 1; which_arg++) { 4258 Node* arg = args[which_arg]; 4259 arg = null_check(arg); 4260 if (stopped()) break; 4261 args[which_arg] = arg; 4262 4263 Node* p = basic_plus_adr(arg, class_klass_offset); 4264 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type); 4265 klasses[which_arg] = _gvn.transform(kls); 4266 } 4267 4268 // Having loaded both klasses, test each for null. 4269 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 4270 for (which_arg = 0; which_arg <= 1; which_arg++) { 4271 Node* kls = klasses[which_arg]; 4272 Node* null_ctl = top(); 4273 kls = null_check_oop(kls, &null_ctl, never_see_null); 4274 if (which_arg == 0) { 4275 prim_region->init_req(1, null_ctl); 4276 } else { 4277 region->init_req(_prim_1_path, null_ctl); 4278 } 4279 if (stopped()) break; 4280 klasses[which_arg] = kls; 4281 } 4282 4283 if (!stopped()) { 4284 // now we have two reference types, in klasses[0..1] 4285 Node* subk = klasses[1]; // the argument to isAssignableFrom 4286 Node* superk = klasses[0]; // the receiver 4287 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 4288 region->set_req(_ref_subtype_path, control()); 4289 } 4290 4291 // If both operands are primitive (both klasses null), then 4292 // we must return true when they are identical primitives. 4293 // It is convenient to test this after the first null klass check. 4294 // This path is also used if superc is a value mirror. 4295 set_control(_gvn.transform(prim_region)); 4296 if (!stopped()) { 4297 // Since superc is primitive, make a guard for the superc==subc case. 4298 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 4299 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 4300 generate_fair_guard(bol_eq, region); 4301 if (region->req() == PATH_LIMIT+1) { 4302 // A guard was added. If the added guard is taken, superc==subc. 4303 region->swap_edges(PATH_LIMIT, _prim_same_path); 4304 region->del_req(PATH_LIMIT); 4305 } 4306 region->set_req(_prim_0_path, control()); // Not equal after all. 4307 } 4308 4309 // these are the only paths that produce 'true': 4310 phi->set_req(_prim_same_path, intcon(1)); 4311 phi->set_req(_ref_subtype_path, intcon(1)); 4312 4313 // pull together the cases: 4314 assert(region->req() == PATH_LIMIT, "sane region"); 4315 for (uint i = 1; i < region->req(); i++) { 4316 Node* ctl = region->in(i); 4317 if (ctl == nullptr || ctl == top()) { 4318 region->set_req(i, top()); 4319 phi ->set_req(i, top()); 4320 } else if (phi->in(i) == nullptr) { 4321 phi->set_req(i, intcon(0)); // all other paths produce 'false' 4322 } 4323 } 4324 4325 set_control(_gvn.transform(region)); 4326 set_result(_gvn.transform(phi)); 4327 return true; 4328 } 4329 4330 //---------------------generate_array_guard_common------------------------ 4331 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) { 4332 4333 if (stopped()) { 4334 return nullptr; 4335 } 4336 4337 // Like generate_guard, adds a new path onto the region. 4338 jint layout_con = 0; 4339 Node* layout_val = get_layout_helper(kls, layout_con); 4340 if (layout_val == nullptr) { 4341 bool query = 0; 4342 switch(kind) { 4343 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break; 4344 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break; 4345 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break; 4346 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break; 4347 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break; 4348 default: 4349 ShouldNotReachHere(); 4350 } 4351 if (!query) { 4352 return nullptr; // never a branch 4353 } else { // always a branch 4354 Node* always_branch = control(); 4355 if (region != nullptr) 4356 region->add_req(always_branch); 4357 set_control(top()); 4358 return always_branch; 4359 } 4360 } 4361 unsigned int value = 0; 4362 BoolTest::mask btest = BoolTest::illegal; 4363 switch(kind) { 4364 case ObjectArray: 4365 case NonObjectArray: { 4366 value = Klass::_lh_array_tag_obj_value; 4367 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 4368 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne; 4369 break; 4370 } 4371 case TypeArray: { 4372 value = Klass::_lh_array_tag_type_value; 4373 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 4374 btest = BoolTest::eq; 4375 break; 4376 } 4377 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break; 4378 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break; 4379 default: 4380 ShouldNotReachHere(); 4381 } 4382 // Now test the correct condition. 4383 jint nval = (jint)value; 4384 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 4385 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 4386 return generate_fair_guard(bol, region); 4387 } 4388 4389 //-----------------------inline_newNullRestrictedArray-------------------------- 4390 // public static native Object[] newNullRestrictedArray(Class<?> componentType, int length); 4391 bool LibraryCallKit::inline_newNullRestrictedArray() { 4392 Node* componentType = argument(0); 4393 Node* length = argument(1); 4394 4395 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr(); 4396 if (tp != nullptr) { 4397 ciInstanceKlass* ik = tp->instance_klass(); 4398 if (ik == C->env()->Class_klass()) { 4399 ciType* t = tp->java_mirror_type(); 4400 if (t != nullptr && t->is_inlinetype()) { 4401 ciArrayKlass* array_klass = ciArrayKlass::make(t, true); 4402 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) { 4403 const TypeAryKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass, Type::trust_interfaces)->is_aryklassptr(); 4404 array_klass_type = array_klass_type->cast_to_null_free(); 4405 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false); // no arguments to push 4406 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj); 4407 alloc->set_null_free(); 4408 set_result(obj); 4409 assert(gvn().type(obj)->is_aryptr()->is_null_free(), "must be null-free"); 4410 return true; 4411 } 4412 } 4413 } 4414 } 4415 return false; 4416 } 4417 4418 //-----------------------inline_native_newArray-------------------------- 4419 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length); 4420 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size); 4421 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) { 4422 Node* mirror; 4423 Node* count_val; 4424 if (uninitialized) { 4425 null_check_receiver(); 4426 mirror = argument(1); 4427 count_val = argument(2); 4428 } else { 4429 mirror = argument(0); 4430 count_val = argument(1); 4431 } 4432 4433 mirror = null_check(mirror); 4434 // If mirror or obj is dead, only null-path is taken. 4435 if (stopped()) return true; 4436 4437 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 4438 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4439 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4440 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4441 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4442 4443 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 4444 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 4445 result_reg, _slow_path); 4446 Node* normal_ctl = control(); 4447 Node* no_array_ctl = result_reg->in(_slow_path); 4448 4449 // Generate code for the slow case. We make a call to newArray(). 4450 set_control(no_array_ctl); 4451 if (!stopped()) { 4452 // Either the input type is void.class, or else the 4453 // array klass has not yet been cached. Either the 4454 // ensuing call will throw an exception, or else it 4455 // will cache the array klass for next time. 4456 PreserveJVMState pjvms(this); 4457 CallJavaNode* slow_call = nullptr; 4458 if (uninitialized) { 4459 // Generate optimized virtual call (holder class 'Unsafe' is final) 4460 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true); 4461 } else { 4462 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true); 4463 } 4464 Node* slow_result = set_results_for_java_call(slow_call); 4465 // this->control() comes from set_results_for_java_call 4466 result_reg->set_req(_slow_path, control()); 4467 result_val->set_req(_slow_path, slow_result); 4468 result_io ->set_req(_slow_path, i_o()); 4469 result_mem->set_req(_slow_path, reset_memory()); 4470 } 4471 4472 set_control(normal_ctl); 4473 if (!stopped()) { 4474 // Normal case: The array type has been cached in the java.lang.Class. 4475 // The following call works fine even if the array type is polymorphic. 4476 // It could be a dynamic mix of int[], boolean[], Object[], etc. 4477 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push 4478 result_reg->init_req(_normal_path, control()); 4479 result_val->init_req(_normal_path, obj); 4480 result_io ->init_req(_normal_path, i_o()); 4481 result_mem->init_req(_normal_path, reset_memory()); 4482 4483 if (uninitialized) { 4484 // Mark the allocation so that zeroing is skipped 4485 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj); 4486 alloc->maybe_set_complete(&_gvn); 4487 } 4488 } 4489 4490 // Return the combined state. 4491 set_i_o( _gvn.transform(result_io) ); 4492 set_all_memory( _gvn.transform(result_mem)); 4493 4494 C->set_has_split_ifs(true); // Has chance for split-if optimization 4495 set_result(result_reg, result_val); 4496 return true; 4497 } 4498 4499 //----------------------inline_native_getLength-------------------------- 4500 // public static native int java.lang.reflect.Array.getLength(Object array); 4501 bool LibraryCallKit::inline_native_getLength() { 4502 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 4503 4504 Node* array = null_check(argument(0)); 4505 // If array is dead, only null-path is taken. 4506 if (stopped()) return true; 4507 4508 // Deoptimize if it is a non-array. 4509 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr); 4510 4511 if (non_array != nullptr) { 4512 PreserveJVMState pjvms(this); 4513 set_control(non_array); 4514 uncommon_trap(Deoptimization::Reason_intrinsic, 4515 Deoptimization::Action_maybe_recompile); 4516 } 4517 4518 // If control is dead, only non-array-path is taken. 4519 if (stopped()) return true; 4520 4521 // The works fine even if the array type is polymorphic. 4522 // It could be a dynamic mix of int[], boolean[], Object[], etc. 4523 Node* result = load_array_length(array); 4524 4525 C->set_has_split_ifs(true); // Has chance for split-if optimization 4526 set_result(result); 4527 return true; 4528 } 4529 4530 //------------------------inline_array_copyOf---------------------------- 4531 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 4532 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 4533 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 4534 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 4535 4536 // Get the arguments. 4537 Node* original = argument(0); 4538 Node* start = is_copyOfRange? argument(1): intcon(0); 4539 Node* end = is_copyOfRange? argument(2): argument(1); 4540 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 4541 4542 Node* newcopy = nullptr; 4543 4544 // Set the original stack and the reexecute bit for the interpreter to reexecute 4545 // the bytecode that invokes Arrays.copyOf if deoptimization happens. 4546 { PreserveReexecuteState preexecs(this); 4547 jvms()->set_should_reexecute(true); 4548 4549 array_type_mirror = null_check(array_type_mirror); 4550 original = null_check(original); 4551 4552 // Check if a null path was taken unconditionally. 4553 if (stopped()) return true; 4554 4555 Node* orig_length = load_array_length(original); 4556 4557 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0); 4558 klass_node = null_check(klass_node); 4559 4560 RegionNode* bailout = new RegionNode(1); 4561 record_for_igvn(bailout); 4562 4563 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc. 4564 // Bail out if that is so. 4565 // Inline type array may have object field that would require a 4566 // write barrier. Conservatively, go to slow path. 4567 // TODO 8251971: Optimize for the case when flat src/dst are later found 4568 // to not contain oops (i.e., move this check to the macro expansion phase). 4569 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4570 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr(); 4571 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr(); 4572 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) && 4573 // Can src array be flat and contain oops? 4574 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) && 4575 // Can dest array be flat and contain oops? 4576 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops()); 4577 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout); 4578 if (not_objArray != nullptr) { 4579 // Improve the klass node's type from the new optimistic assumption: 4580 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 4581 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0)); 4582 Node* cast = new CastPPNode(control(), klass_node, akls); 4583 klass_node = _gvn.transform(cast); 4584 } 4585 4586 // Bail out if either start or end is negative. 4587 generate_negative_guard(start, bailout, &start); 4588 generate_negative_guard(end, bailout, &end); 4589 4590 Node* length = end; 4591 if (_gvn.type(start) != TypeInt::ZERO) { 4592 length = _gvn.transform(new SubINode(end, start)); 4593 } 4594 4595 // Bail out if length is negative. 4596 // Without this the new_array would throw 4597 // NegativeArraySizeException but IllegalArgumentException is what 4598 // should be thrown 4599 generate_negative_guard(length, bailout, &length); 4600 4601 // Handle inline type arrays 4602 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check); 4603 if (!stopped()) { 4604 // TODO JDK-8329224 4605 if (!orig_t->is_null_free()) { 4606 // Not statically known to be null free, add a check 4607 generate_fair_guard(null_free_array_test(original), bailout); 4608 } 4609 orig_t = _gvn.type(original)->isa_aryptr(); 4610 if (orig_t != nullptr && orig_t->is_flat()) { 4611 // Src is flat, check that dest is flat as well 4612 if (exclude_flat) { 4613 // Dest can't be flat, bail out 4614 bailout->add_req(control()); 4615 set_control(top()); 4616 } else { 4617 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout); 4618 } 4619 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) && 4620 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated). 4621 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) { 4622 // Src might be flat and dest might not be flat. Go to the slow path if src is flat. 4623 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat. 4624 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout); 4625 if (orig_t != nullptr) { 4626 orig_t = orig_t->cast_to_not_flat(); 4627 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t)); 4628 } 4629 } 4630 if (!can_validate) { 4631 // No validation. The subtype check emitted at macro expansion time will not go to the slow 4632 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays. 4633 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free. 4634 generate_fair_guard(flat_array_test(klass_node), bailout); 4635 generate_fair_guard(null_free_array_test(original), bailout); 4636 } 4637 } 4638 4639 if (bailout->req() > 1) { 4640 PreserveJVMState pjvms(this); 4641 set_control(_gvn.transform(bailout)); 4642 uncommon_trap(Deoptimization::Reason_intrinsic, 4643 Deoptimization::Action_maybe_recompile); 4644 } 4645 4646 if (!stopped()) { 4647 // How many elements will we copy from the original? 4648 // The answer is MinI(orig_length - start, length). 4649 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 4650 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length); 4651 4652 // Generate a direct call to the right arraycopy function(s). 4653 // We know the copy is disjoint but we might not know if the 4654 // oop stores need checking. 4655 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 4656 // This will fail a store-check if x contains any non-nulls. 4657 4658 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 4659 // loads/stores but it is legal only if we're sure the 4660 // Arrays.copyOf would succeed. So we need all input arguments 4661 // to the copyOf to be validated, including that the copy to the 4662 // new array won't trigger an ArrayStoreException. That subtype 4663 // check can be optimized if we know something on the type of 4664 // the input array from type speculation. 4665 if (_gvn.type(klass_node)->singleton()) { 4666 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr(); 4667 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr(); 4668 4669 int test = C->static_subtype_check(superk, subk); 4670 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 4671 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 4672 if (t_original->speculative_type() != nullptr) { 4673 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 4674 } 4675 } 4676 } 4677 4678 bool validated = false; 4679 // Reason_class_check rather than Reason_intrinsic because we 4680 // want to intrinsify even if this traps. 4681 if (can_validate) { 4682 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node); 4683 4684 if (not_subtype_ctrl != top()) { 4685 PreserveJVMState pjvms(this); 4686 set_control(not_subtype_ctrl); 4687 uncommon_trap(Deoptimization::Reason_class_check, 4688 Deoptimization::Action_make_not_entrant); 4689 assert(stopped(), "Should be stopped"); 4690 } 4691 validated = true; 4692 } 4693 4694 if (!stopped()) { 4695 newcopy = new_array(klass_node, length, 0); // no arguments to push 4696 4697 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false, 4698 load_object_klass(original), klass_node); 4699 if (!is_copyOfRange) { 4700 ac->set_copyof(validated); 4701 } else { 4702 ac->set_copyofrange(validated); 4703 } 4704 Node* n = _gvn.transform(ac); 4705 if (n == ac) { 4706 ac->connect_outputs(this); 4707 } else { 4708 assert(validated, "shouldn't transform if all arguments not validated"); 4709 set_all_memory(n); 4710 } 4711 } 4712 } 4713 } // original reexecute is set back here 4714 4715 C->set_has_split_ifs(true); // Has chance for split-if optimization 4716 if (!stopped()) { 4717 set_result(newcopy); 4718 } 4719 return true; 4720 } 4721 4722 4723 //----------------------generate_virtual_guard--------------------------- 4724 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 4725 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 4726 RegionNode* slow_region) { 4727 ciMethod* method = callee(); 4728 int vtable_index = method->vtable_index(); 4729 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4730 "bad index %d", vtable_index); 4731 // Get the Method* out of the appropriate vtable entry. 4732 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 4733 vtable_index*vtableEntry::size_in_bytes() + 4734 in_bytes(vtableEntry::method_offset()); 4735 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 4736 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 4737 4738 // Compare the target method with the expected method (e.g., Object.hashCode). 4739 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 4740 4741 Node* native_call = makecon(native_call_addr); 4742 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 4743 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 4744 4745 return generate_slow_guard(test_native, slow_region); 4746 } 4747 4748 //-----------------------generate_method_call---------------------------- 4749 // Use generate_method_call to make a slow-call to the real 4750 // method if the fast path fails. An alternative would be to 4751 // use a stub like OptoRuntime::slow_arraycopy_Java. 4752 // This only works for expanding the current library call, 4753 // not another intrinsic. (E.g., don't use this for making an 4754 // arraycopy call inside of the copyOf intrinsic.) 4755 CallJavaNode* 4756 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) { 4757 // When compiling the intrinsic method itself, do not use this technique. 4758 guarantee(callee() != C->method(), "cannot make slow-call to self"); 4759 4760 ciMethod* method = callee(); 4761 // ensure the JVMS we have will be correct for this call 4762 guarantee(method_id == method->intrinsic_id(), "must match"); 4763 4764 const TypeFunc* tf = TypeFunc::make(method); 4765 if (res_not_null) { 4766 assert(tf->return_type() == T_OBJECT, ""); 4767 const TypeTuple* range = tf->range_cc(); 4768 const Type** fields = TypeTuple::fields(range->cnt()); 4769 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL); 4770 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields); 4771 tf = TypeFunc::make(tf->domain_cc(), new_range); 4772 } 4773 CallJavaNode* slow_call; 4774 if (is_static) { 4775 assert(!is_virtual, ""); 4776 slow_call = new CallStaticJavaNode(C, tf, 4777 SharedRuntime::get_resolve_static_call_stub(), method); 4778 } else if (is_virtual) { 4779 assert(!gvn().type(argument(0))->maybe_null(), "should not be null"); 4780 int vtable_index = Method::invalid_vtable_index; 4781 if (UseInlineCaches) { 4782 // Suppress the vtable call 4783 } else { 4784 // hashCode and clone are not a miranda methods, 4785 // so the vtable index is fixed. 4786 // No need to use the linkResolver to get it. 4787 vtable_index = method->vtable_index(); 4788 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4789 "bad index %d", vtable_index); 4790 } 4791 slow_call = new CallDynamicJavaNode(tf, 4792 SharedRuntime::get_resolve_virtual_call_stub(), 4793 method, vtable_index); 4794 } else { // neither virtual nor static: opt_virtual 4795 assert(!gvn().type(argument(0))->maybe_null(), "should not be null"); 4796 slow_call = new CallStaticJavaNode(C, tf, 4797 SharedRuntime::get_resolve_opt_virtual_call_stub(), method); 4798 slow_call->set_optimized_virtual(true); 4799 } 4800 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) { 4801 // To be able to issue a direct call (optimized virtual or virtual) 4802 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information 4803 // about the method being invoked should be attached to the call site to 4804 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C). 4805 slow_call->set_override_symbolic_info(true); 4806 } 4807 set_arguments_for_java_call(slow_call); 4808 set_edges_for_java_call(slow_call); 4809 return slow_call; 4810 } 4811 4812 4813 /** 4814 * Build special case code for calls to hashCode on an object. This call may 4815 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 4816 * slightly different code. 4817 */ 4818 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 4819 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 4820 assert(!(is_virtual && is_static), "either virtual, special, or static"); 4821 4822 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 4823 4824 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4825 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 4826 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4827 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4828 Node* obj = argument(0); 4829 4830 if (gvn().type(obj)->is_inlinetypeptr()) { 4831 return false; 4832 } 4833 4834 if (!is_static) { 4835 // Check for hashing null object 4836 obj = null_check_receiver(); 4837 if (stopped()) return true; // unconditionally null 4838 result_reg->init_req(_null_path, top()); 4839 result_val->init_req(_null_path, top()); 4840 } else { 4841 // Do a null check, and return zero if null. 4842 // System.identityHashCode(null) == 0 4843 Node* null_ctl = top(); 4844 obj = null_check_oop(obj, &null_ctl); 4845 result_reg->init_req(_null_path, null_ctl); 4846 result_val->init_req(_null_path, _gvn.intcon(0)); 4847 } 4848 4849 // Unconditionally null? Then return right away. 4850 if (stopped()) { 4851 set_control( result_reg->in(_null_path)); 4852 if (!stopped()) 4853 set_result(result_val->in(_null_path)); 4854 return true; 4855 } 4856 4857 // We only go to the fast case code if we pass a number of guards. The 4858 // paths which do not pass are accumulated in the slow_region. 4859 RegionNode* slow_region = new RegionNode(1); 4860 record_for_igvn(slow_region); 4861 4862 // If this is a virtual call, we generate a funny guard. We pull out 4863 // the vtable entry corresponding to hashCode() from the target object. 4864 // If the target method which we are calling happens to be the native 4865 // Object hashCode() method, we pass the guard. We do not need this 4866 // guard for non-virtual calls -- the caller is known to be the native 4867 // Object hashCode(). 4868 if (is_virtual) { 4869 // After null check, get the object's klass. 4870 Node* obj_klass = load_object_klass(obj); 4871 generate_virtual_guard(obj_klass, slow_region); 4872 } 4873 4874 // Get the header out of the object, use LoadMarkNode when available 4875 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 4876 // The control of the load must be null. Otherwise, the load can move before 4877 // the null check after castPP removal. 4878 Node* no_ctrl = nullptr; 4879 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 4880 4881 // Test the header to see if it is safe to read w.r.t. locking. 4882 // This also serves as guard against inline types 4883 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place); 4884 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 4885 if (LockingMode == LM_LIGHTWEIGHT) { 4886 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value); 4887 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val)); 4888 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq)); 4889 4890 generate_slow_guard(test_monitor, slow_region); 4891 } else { 4892 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value); 4893 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val)); 4894 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne)); 4895 4896 generate_slow_guard(test_not_unlocked, slow_region); 4897 } 4898 4899 // Get the hash value and check to see that it has been properly assigned. 4900 // We depend on hash_mask being at most 32 bits and avoid the use of 4901 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 4902 // vm: see markWord.hpp. 4903 Node *hash_mask = _gvn.intcon(markWord::hash_mask); 4904 Node *hash_shift = _gvn.intcon(markWord::hash_shift); 4905 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 4906 // This hack lets the hash bits live anywhere in the mark object now, as long 4907 // as the shift drops the relevant bits into the low 32 bits. Note that 4908 // Java spec says that HashCode is an int so there's no point in capturing 4909 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 4910 hshifted_header = ConvX2I(hshifted_header); 4911 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 4912 4913 Node *no_hash_val = _gvn.intcon(markWord::no_hash); 4914 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 4915 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 4916 4917 generate_slow_guard(test_assigned, slow_region); 4918 4919 Node* init_mem = reset_memory(); 4920 // fill in the rest of the null path: 4921 result_io ->init_req(_null_path, i_o()); 4922 result_mem->init_req(_null_path, init_mem); 4923 4924 result_val->init_req(_fast_path, hash_val); 4925 result_reg->init_req(_fast_path, control()); 4926 result_io ->init_req(_fast_path, i_o()); 4927 result_mem->init_req(_fast_path, init_mem); 4928 4929 // Generate code for the slow case. We make a call to hashCode(). 4930 set_control(_gvn.transform(slow_region)); 4931 if (!stopped()) { 4932 // No need for PreserveJVMState, because we're using up the present state. 4933 set_all_memory(init_mem); 4934 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 4935 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false); 4936 Node* slow_result = set_results_for_java_call(slow_call); 4937 // this->control() comes from set_results_for_java_call 4938 result_reg->init_req(_slow_path, control()); 4939 result_val->init_req(_slow_path, slow_result); 4940 result_io ->set_req(_slow_path, i_o()); 4941 result_mem ->set_req(_slow_path, reset_memory()); 4942 } 4943 4944 // Return the combined state. 4945 set_i_o( _gvn.transform(result_io) ); 4946 set_all_memory( _gvn.transform(result_mem)); 4947 4948 set_result(result_reg, result_val); 4949 return true; 4950 } 4951 4952 //---------------------------inline_native_getClass---------------------------- 4953 // public final native Class<?> java.lang.Object.getClass(); 4954 // 4955 // Build special case code for calls to getClass on an object. 4956 bool LibraryCallKit::inline_native_getClass() { 4957 Node* obj = argument(0); 4958 if (obj->is_InlineType()) { 4959 const Type* t = _gvn.type(obj); 4960 if (t->maybe_null()) { 4961 null_check(obj); 4962 } 4963 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror()))); 4964 return true; 4965 } 4966 obj = null_check_receiver(); 4967 if (stopped()) return true; 4968 set_result(load_mirror_from_klass(load_object_klass(obj))); 4969 return true; 4970 } 4971 4972 //-----------------inline_native_Reflection_getCallerClass--------------------- 4973 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 4974 // 4975 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 4976 // 4977 // NOTE: This code must perform the same logic as JVM_GetCallerClass 4978 // in that it must skip particular security frames and checks for 4979 // caller sensitive methods. 4980 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 4981 #ifndef PRODUCT 4982 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4983 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 4984 } 4985 #endif 4986 4987 if (!jvms()->has_method()) { 4988 #ifndef PRODUCT 4989 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4990 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 4991 } 4992 #endif 4993 return false; 4994 } 4995 4996 // Walk back up the JVM state to find the caller at the required 4997 // depth. 4998 JVMState* caller_jvms = jvms(); 4999 5000 // Cf. JVM_GetCallerClass 5001 // NOTE: Start the loop at depth 1 because the current JVM state does 5002 // not include the Reflection.getCallerClass() frame. 5003 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) { 5004 ciMethod* m = caller_jvms->method(); 5005 switch (n) { 5006 case 0: 5007 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 5008 break; 5009 case 1: 5010 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 5011 if (!m->caller_sensitive()) { 5012 #ifndef PRODUCT 5013 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 5014 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 5015 } 5016 #endif 5017 return false; // bail-out; let JVM_GetCallerClass do the work 5018 } 5019 break; 5020 default: 5021 if (!m->is_ignored_by_security_stack_walk()) { 5022 // We have reached the desired frame; return the holder class. 5023 // Acquire method holder as java.lang.Class and push as constant. 5024 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 5025 ciInstance* caller_mirror = caller_klass->java_mirror(); 5026 set_result(makecon(TypeInstPtr::make(caller_mirror))); 5027 5028 #ifndef PRODUCT 5029 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 5030 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()); 5031 tty->print_cr(" JVM state at this point:"); 5032 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 5033 ciMethod* m = jvms()->of_depth(i)->method(); 5034 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 5035 } 5036 } 5037 #endif 5038 return true; 5039 } 5040 break; 5041 } 5042 } 5043 5044 #ifndef PRODUCT 5045 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 5046 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 5047 tty->print_cr(" JVM state at this point:"); 5048 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 5049 ciMethod* m = jvms()->of_depth(i)->method(); 5050 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 5051 } 5052 } 5053 #endif 5054 5055 return false; // bail-out; let JVM_GetCallerClass do the work 5056 } 5057 5058 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 5059 Node* arg = argument(0); 5060 Node* result = nullptr; 5061 5062 switch (id) { 5063 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 5064 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 5065 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 5066 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 5067 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break; 5068 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break; 5069 5070 case vmIntrinsics::_doubleToLongBits: { 5071 // two paths (plus control) merge in a wood 5072 RegionNode *r = new RegionNode(3); 5073 Node *phi = new PhiNode(r, TypeLong::LONG); 5074 5075 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 5076 // Build the boolean node 5077 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 5078 5079 // Branch either way. 5080 // NaN case is less traveled, which makes all the difference. 5081 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 5082 Node *opt_isnan = _gvn.transform(ifisnan); 5083 assert( opt_isnan->is_If(), "Expect an IfNode"); 5084 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 5085 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 5086 5087 set_control(iftrue); 5088 5089 static const jlong nan_bits = CONST64(0x7ff8000000000000); 5090 Node *slow_result = longcon(nan_bits); // return NaN 5091 phi->init_req(1, _gvn.transform( slow_result )); 5092 r->init_req(1, iftrue); 5093 5094 // Else fall through 5095 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 5096 set_control(iffalse); 5097 5098 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 5099 r->init_req(2, iffalse); 5100 5101 // Post merge 5102 set_control(_gvn.transform(r)); 5103 record_for_igvn(r); 5104 5105 C->set_has_split_ifs(true); // Has chance for split-if optimization 5106 result = phi; 5107 assert(result->bottom_type()->isa_long(), "must be"); 5108 break; 5109 } 5110 5111 case vmIntrinsics::_floatToIntBits: { 5112 // two paths (plus control) merge in a wood 5113 RegionNode *r = new RegionNode(3); 5114 Node *phi = new PhiNode(r, TypeInt::INT); 5115 5116 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 5117 // Build the boolean node 5118 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 5119 5120 // Branch either way. 5121 // NaN case is less traveled, which makes all the difference. 5122 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 5123 Node *opt_isnan = _gvn.transform(ifisnan); 5124 assert( opt_isnan->is_If(), "Expect an IfNode"); 5125 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 5126 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 5127 5128 set_control(iftrue); 5129 5130 static const jint nan_bits = 0x7fc00000; 5131 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 5132 phi->init_req(1, _gvn.transform( slow_result )); 5133 r->init_req(1, iftrue); 5134 5135 // Else fall through 5136 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 5137 set_control(iffalse); 5138 5139 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 5140 r->init_req(2, iffalse); 5141 5142 // Post merge 5143 set_control(_gvn.transform(r)); 5144 record_for_igvn(r); 5145 5146 C->set_has_split_ifs(true); // Has chance for split-if optimization 5147 result = phi; 5148 assert(result->bottom_type()->isa_int(), "must be"); 5149 break; 5150 } 5151 5152 default: 5153 fatal_unexpected_iid(id); 5154 break; 5155 } 5156 set_result(_gvn.transform(result)); 5157 return true; 5158 } 5159 5160 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) { 5161 Node* arg = argument(0); 5162 Node* result = nullptr; 5163 5164 switch (id) { 5165 case vmIntrinsics::_floatIsInfinite: 5166 result = new IsInfiniteFNode(arg); 5167 break; 5168 case vmIntrinsics::_floatIsFinite: 5169 result = new IsFiniteFNode(arg); 5170 break; 5171 case vmIntrinsics::_doubleIsInfinite: 5172 result = new IsInfiniteDNode(arg); 5173 break; 5174 case vmIntrinsics::_doubleIsFinite: 5175 result = new IsFiniteDNode(arg); 5176 break; 5177 default: 5178 fatal_unexpected_iid(id); 5179 break; 5180 } 5181 set_result(_gvn.transform(result)); 5182 return true; 5183 } 5184 5185 //----------------------inline_unsafe_copyMemory------------------------- 5186 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 5187 5188 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) { 5189 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr(); 5190 const Type* base_t = gvn.type(base); 5191 5192 bool in_native = (base_t == TypePtr::NULL_PTR); 5193 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t); 5194 bool is_mixed = !in_heap && !in_native; 5195 5196 if (is_mixed) { 5197 return true; // mixed accesses can touch both on-heap and off-heap memory 5198 } 5199 if (in_heap) { 5200 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM); 5201 if (!is_prim_array) { 5202 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array, 5203 // there's not enough type information available to determine proper memory slice for it. 5204 return true; 5205 } 5206 } 5207 return false; 5208 } 5209 5210 bool LibraryCallKit::inline_unsafe_copyMemory() { 5211 if (callee()->is_static()) return false; // caller must have the capability! 5212 null_check_receiver(); // null-check receiver 5213 if (stopped()) return true; 5214 5215 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 5216 5217 Node* src_base = argument(1); // type: oop 5218 Node* src_off = ConvL2X(argument(2)); // type: long 5219 Node* dst_base = argument(4); // type: oop 5220 Node* dst_off = ConvL2X(argument(5)); // type: long 5221 Node* size = ConvL2X(argument(7)); // type: long 5222 5223 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 5224 "fieldOffset must be byte-scaled"); 5225 5226 Node* src_addr = make_unsafe_address(src_base, src_off); 5227 Node* dst_addr = make_unsafe_address(dst_base, dst_off); 5228 5229 Node* thread = _gvn.transform(new ThreadLocalNode()); 5230 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 5231 BasicType doing_unsafe_access_bt = T_BYTE; 5232 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 5233 5234 // update volatile field 5235 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 5236 5237 int flags = RC_LEAF | RC_NO_FP; 5238 5239 const TypePtr* dst_type = TypePtr::BOTTOM; 5240 5241 // Adjust memory effects of the runtime call based on input values. 5242 if (!has_wide_mem(_gvn, src_addr, src_base) && 5243 !has_wide_mem(_gvn, dst_addr, dst_base)) { 5244 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory 5245 5246 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr(); 5247 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) { 5248 flags |= RC_NARROW_MEM; // narrow in memory 5249 } 5250 } 5251 5252 // Call it. Note that the length argument is not scaled. 5253 make_runtime_call(flags, 5254 OptoRuntime::fast_arraycopy_Type(), 5255 StubRoutines::unsafe_arraycopy(), 5256 "unsafe_arraycopy", 5257 dst_type, 5258 src_addr, dst_addr, size XTOP); 5259 5260 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 5261 5262 return true; 5263 } 5264 5265 #undef XTOP 5266 5267 // TODO 8325106 Remove this and corresponding tests. Flatness is not a property of the Class anymore with JEP 401. 5268 //----------------------inline_unsafe_isFlatArray------------------------ 5269 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass); 5270 // This intrinsic exploits assumptions made by the native implementation 5271 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks. 5272 bool LibraryCallKit::inline_unsafe_isFlatArray() { 5273 Node* cls = argument(1); 5274 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset()); 5275 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, 5276 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT)); 5277 Node* result = flat_array_test(kls); 5278 set_result(result); 5279 return true; 5280 } 5281 5282 //------------------------clone_coping----------------------------------- 5283 // Helper function for inline_native_clone. 5284 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) { 5285 assert(obj_size != nullptr, ""); 5286 Node* raw_obj = alloc_obj->in(1); 5287 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 5288 5289 AllocateNode* alloc = nullptr; 5290 if (ReduceBulkZeroing) { 5291 // We will be completely responsible for initializing this object - 5292 // mark Initialize node as complete. 5293 alloc = AllocateNode::Ideal_allocation(alloc_obj); 5294 // The object was just allocated - there should be no any stores! 5295 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), ""); 5296 // Mark as complete_with_arraycopy so that on AllocateNode 5297 // expansion, we know this AllocateNode is initialized by an array 5298 // copy and a StoreStore barrier exists after the array copy. 5299 alloc->initialization()->set_complete_with_arraycopy(); 5300 } 5301 5302 Node* size = _gvn.transform(obj_size); 5303 access_clone(obj, alloc_obj, size, is_array); 5304 5305 // Do not let reads from the cloned object float above the arraycopy. 5306 if (alloc != nullptr) { 5307 // Do not let stores that initialize this object be reordered with 5308 // a subsequent store that would make this object accessible by 5309 // other threads. 5310 // Record what AllocateNode this StoreStore protects so that 5311 // escape analysis can go from the MemBarStoreStoreNode to the 5312 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 5313 // based on the escape status of the AllocateNode. 5314 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 5315 } else { 5316 insert_mem_bar(Op_MemBarCPUOrder); 5317 } 5318 } 5319 5320 //------------------------inline_native_clone---------------------------- 5321 // protected native Object java.lang.Object.clone(); 5322 // 5323 // Here are the simple edge cases: 5324 // null receiver => normal trap 5325 // virtual and clone was overridden => slow path to out-of-line clone 5326 // not cloneable or finalizer => slow path to out-of-line Object.clone 5327 // 5328 // The general case has two steps, allocation and copying. 5329 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 5330 // 5331 // Copying also has two cases, oop arrays and everything else. 5332 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 5333 // Everything else uses the tight inline loop supplied by CopyArrayNode. 5334 // 5335 // These steps fold up nicely if and when the cloned object's klass 5336 // can be sharply typed as an object array, a type array, or an instance. 5337 // 5338 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 5339 PhiNode* result_val; 5340 5341 // Set the reexecute bit for the interpreter to reexecute 5342 // the bytecode that invokes Object.clone if deoptimization happens. 5343 { PreserveReexecuteState preexecs(this); 5344 jvms()->set_should_reexecute(true); 5345 5346 Node* obj = argument(0); 5347 obj = null_check_receiver(); 5348 if (stopped()) return true; 5349 5350 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 5351 if (obj_type->is_inlinetypeptr()) { 5352 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have 5353 // no identity. 5354 set_result(obj); 5355 return true; 5356 } 5357 5358 // If we are going to clone an instance, we need its exact type to 5359 // know the number and types of fields to convert the clone to 5360 // loads/stores. Maybe a speculative type can help us. 5361 if (!obj_type->klass_is_exact() && 5362 obj_type->speculative_type() != nullptr && 5363 obj_type->speculative_type()->is_instance_klass() && 5364 !obj_type->speculative_type()->is_inlinetype()) { 5365 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 5366 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 5367 !spec_ik->has_injected_fields()) { 5368 if (!obj_type->isa_instptr() || 5369 obj_type->is_instptr()->instance_klass()->has_subklass()) { 5370 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 5371 } 5372 } 5373 } 5374 5375 // Conservatively insert a memory barrier on all memory slices. 5376 // Do not let writes into the original float below the clone. 5377 insert_mem_bar(Op_MemBarCPUOrder); 5378 5379 // paths into result_reg: 5380 enum { 5381 _slow_path = 1, // out-of-line call to clone method (virtual or not) 5382 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 5383 _array_path, // plain array allocation, plus arrayof_long_arraycopy 5384 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 5385 PATH_LIMIT 5386 }; 5387 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 5388 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 5389 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 5390 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 5391 record_for_igvn(result_reg); 5392 5393 Node* obj_klass = load_object_klass(obj); 5394 // We only go to the fast case code if we pass a number of guards. 5395 // The paths which do not pass are accumulated in the slow_region. 5396 RegionNode* slow_region = new RegionNode(1); 5397 record_for_igvn(slow_region); 5398 5399 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr); 5400 if (array_ctl != nullptr) { 5401 // It's an array. 5402 PreserveJVMState pjvms(this); 5403 set_control(array_ctl); 5404 5405 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 5406 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr(); 5407 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) && 5408 obj_type->can_be_inline_array() && 5409 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) { 5410 // Flat inline type array may have object field that would require a 5411 // write barrier. Conservatively, go to slow path. 5412 generate_fair_guard(flat_array_test(obj_klass), slow_region); 5413 } 5414 5415 if (!stopped()) { 5416 Node* obj_length = load_array_length(obj); 5417 Node* array_size = nullptr; // Size of the array without object alignment padding. 5418 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true); 5419 5420 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 5421 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) { 5422 // If it is an oop array, it requires very special treatment, 5423 // because gc barriers are required when accessing the array. 5424 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr); 5425 if (is_obja != nullptr) { 5426 PreserveJVMState pjvms2(this); 5427 set_control(is_obja); 5428 // Generate a direct call to the right arraycopy function(s). 5429 // Clones are always tightly coupled. 5430 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false); 5431 ac->set_clone_oop_array(); 5432 Node* n = _gvn.transform(ac); 5433 assert(n == ac, "cannot disappear"); 5434 ac->connect_outputs(this, /*deoptimize_on_exception=*/true); 5435 5436 result_reg->init_req(_objArray_path, control()); 5437 result_val->init_req(_objArray_path, alloc_obj); 5438 result_i_o ->set_req(_objArray_path, i_o()); 5439 result_mem ->set_req(_objArray_path, reset_memory()); 5440 } 5441 } 5442 // Otherwise, there are no barriers to worry about. 5443 // (We can dispense with card marks if we know the allocation 5444 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 5445 // causes the non-eden paths to take compensating steps to 5446 // simulate a fresh allocation, so that no further 5447 // card marks are required in compiled code to initialize 5448 // the object.) 5449 5450 if (!stopped()) { 5451 copy_to_clone(obj, alloc_obj, array_size, true); 5452 5453 // Present the results of the copy. 5454 result_reg->init_req(_array_path, control()); 5455 result_val->init_req(_array_path, alloc_obj); 5456 result_i_o ->set_req(_array_path, i_o()); 5457 result_mem ->set_req(_array_path, reset_memory()); 5458 } 5459 } 5460 } 5461 5462 if (!stopped()) { 5463 // It's an instance (we did array above). Make the slow-path tests. 5464 // If this is a virtual call, we generate a funny guard. We grab 5465 // the vtable entry corresponding to clone() from the target object. 5466 // If the target method which we are calling happens to be the 5467 // Object clone() method, we pass the guard. We do not need this 5468 // guard for non-virtual calls; the caller is known to be the native 5469 // Object clone(). 5470 if (is_virtual) { 5471 generate_virtual_guard(obj_klass, slow_region); 5472 } 5473 5474 // The object must be easily cloneable and must not have a finalizer. 5475 // Both of these conditions may be checked in a single test. 5476 // We could optimize the test further, but we don't care. 5477 generate_access_flags_guard(obj_klass, 5478 // Test both conditions: 5479 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER, 5480 // Must be cloneable but not finalizer: 5481 JVM_ACC_IS_CLONEABLE_FAST, 5482 slow_region); 5483 } 5484 5485 if (!stopped()) { 5486 // It's an instance, and it passed the slow-path tests. 5487 PreserveJVMState pjvms(this); 5488 Node* obj_size = nullptr; // Total object size, including object alignment padding. 5489 // Need to deoptimize on exception from allocation since Object.clone intrinsic 5490 // is reexecuted if deoptimization occurs and there could be problems when merging 5491 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 5492 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true); 5493 5494 copy_to_clone(obj, alloc_obj, obj_size, false); 5495 5496 // Present the results of the slow call. 5497 result_reg->init_req(_instance_path, control()); 5498 result_val->init_req(_instance_path, alloc_obj); 5499 result_i_o ->set_req(_instance_path, i_o()); 5500 result_mem ->set_req(_instance_path, reset_memory()); 5501 } 5502 5503 // Generate code for the slow case. We make a call to clone(). 5504 set_control(_gvn.transform(slow_region)); 5505 if (!stopped()) { 5506 PreserveJVMState pjvms(this); 5507 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true); 5508 // We need to deoptimize on exception (see comment above) 5509 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true); 5510 // this->control() comes from set_results_for_java_call 5511 result_reg->init_req(_slow_path, control()); 5512 result_val->init_req(_slow_path, slow_result); 5513 result_i_o ->set_req(_slow_path, i_o()); 5514 result_mem ->set_req(_slow_path, reset_memory()); 5515 } 5516 5517 // Return the combined state. 5518 set_control( _gvn.transform(result_reg)); 5519 set_i_o( _gvn.transform(result_i_o)); 5520 set_all_memory( _gvn.transform(result_mem)); 5521 } // original reexecute is set back here 5522 5523 set_result(_gvn.transform(result_val)); 5524 return true; 5525 } 5526 5527 // If we have a tightly coupled allocation, the arraycopy may take care 5528 // of the array initialization. If one of the guards we insert between 5529 // the allocation and the arraycopy causes a deoptimization, an 5530 // uninitialized array will escape the compiled method. To prevent that 5531 // we set the JVM state for uncommon traps between the allocation and 5532 // the arraycopy to the state before the allocation so, in case of 5533 // deoptimization, we'll reexecute the allocation and the 5534 // initialization. 5535 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 5536 if (alloc != nullptr) { 5537 ciMethod* trap_method = alloc->jvms()->method(); 5538 int trap_bci = alloc->jvms()->bci(); 5539 5540 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5541 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 5542 // Make sure there's no store between the allocation and the 5543 // arraycopy otherwise visible side effects could be rexecuted 5544 // in case of deoptimization and cause incorrect execution. 5545 bool no_interfering_store = true; 5546 Node* mem = alloc->in(TypeFunc::Memory); 5547 if (mem->is_MergeMem()) { 5548 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 5549 Node* n = mms.memory(); 5550 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 5551 assert(n->is_Store(), "what else?"); 5552 no_interfering_store = false; 5553 break; 5554 } 5555 } 5556 } else { 5557 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 5558 Node* n = mms.memory(); 5559 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 5560 assert(n->is_Store(), "what else?"); 5561 no_interfering_store = false; 5562 break; 5563 } 5564 } 5565 } 5566 5567 if (no_interfering_store) { 5568 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc); 5569 5570 JVMState* saved_jvms = jvms(); 5571 saved_reexecute_sp = _reexecute_sp; 5572 5573 set_jvms(sfpt->jvms()); 5574 _reexecute_sp = jvms()->sp(); 5575 5576 return saved_jvms; 5577 } 5578 } 5579 } 5580 return nullptr; 5581 } 5582 5583 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack 5584 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter. 5585 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const { 5586 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 5587 uint size = alloc->req(); 5588 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 5589 old_jvms->set_map(sfpt); 5590 for (uint i = 0; i < size; i++) { 5591 sfpt->init_req(i, alloc->in(i)); 5592 } 5593 int adjustment = 1; 5594 // TODO 8325106 why can't we check via the type of the const klass node? 5595 if (alloc->is_null_free()) { 5596 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newNullRestrictedArray 5597 // which requires both the component type and the array length on stack for re-execution. Re-create and push 5598 // the component type. 5599 ciArrayKlass* klass = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr()->exact_klass()->as_array_klass(); 5600 ciInstance* instance = klass->component_mirror_instance(); 5601 const TypeInstPtr* t_instance = TypeInstPtr::make(instance); 5602 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance)); 5603 adjustment++; 5604 } 5605 // re-push array length for deoptimization 5606 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength)); 5607 old_jvms->set_sp(old_jvms->sp() + adjustment); 5608 old_jvms->set_monoff(old_jvms->monoff() + adjustment); 5609 old_jvms->set_scloff(old_jvms->scloff() + adjustment); 5610 old_jvms->set_endoff(old_jvms->endoff() + adjustment); 5611 old_jvms->set_should_reexecute(true); 5612 5613 sfpt->set_i_o(map()->i_o()); 5614 sfpt->set_memory(map()->memory()); 5615 sfpt->set_control(map()->control()); 5616 return sfpt; 5617 } 5618 5619 // In case of a deoptimization, we restart execution at the 5620 // allocation, allocating a new array. We would leave an uninitialized 5621 // array in the heap that GCs wouldn't expect. Move the allocation 5622 // after the traps so we don't allocate the array if we 5623 // deoptimize. This is possible because tightly_coupled_allocation() 5624 // guarantees there's no observer of the allocated array at this point 5625 // and the control flow is simple enough. 5626 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards, 5627 int saved_reexecute_sp, uint new_idx) { 5628 if (saved_jvms_before_guards != nullptr && !stopped()) { 5629 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards); 5630 5631 assert(alloc != nullptr, "only with a tightly coupled allocation"); 5632 // restore JVM state to the state at the arraycopy 5633 saved_jvms_before_guards->map()->set_control(map()->control()); 5634 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?"); 5635 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?"); 5636 // If we've improved the types of some nodes (null check) while 5637 // emitting the guards, propagate them to the current state 5638 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx); 5639 set_jvms(saved_jvms_before_guards); 5640 _reexecute_sp = saved_reexecute_sp; 5641 5642 // Remove the allocation from above the guards 5643 CallProjections* callprojs = alloc->extract_projections(true); 5644 InitializeNode* init = alloc->initialization(); 5645 Node* alloc_mem = alloc->in(TypeFunc::Memory); 5646 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 5647 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 5648 5649 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below 5650 // the allocation (i.e. is only valid if the allocation succeeds): 5651 // 1) replace CastIINode with AllocateArrayNode's length here 5652 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method 5653 // 5654 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate 5655 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy) 5656 Node* init_control = init->proj_out(TypeFunc::Control); 5657 Node* alloc_length = alloc->Ideal_length(); 5658 #ifdef ASSERT 5659 Node* prev_cast = nullptr; 5660 #endif 5661 for (uint i = 0; i < init_control->outcnt(); i++) { 5662 Node* init_out = init_control->raw_out(i); 5663 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) { 5664 #ifdef ASSERT 5665 if (prev_cast == nullptr) { 5666 prev_cast = init_out; 5667 } else { 5668 if (prev_cast->cmp(*init_out) == false) { 5669 prev_cast->dump(); 5670 init_out->dump(); 5671 assert(false, "not equal CastIINode"); 5672 } 5673 } 5674 #endif 5675 C->gvn_replace_by(init_out, alloc_length); 5676 } 5677 } 5678 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 5679 5680 // move the allocation here (after the guards) 5681 _gvn.hash_delete(alloc); 5682 alloc->set_req(TypeFunc::Control, control()); 5683 alloc->set_req(TypeFunc::I_O, i_o()); 5684 Node *mem = reset_memory(); 5685 set_all_memory(mem); 5686 alloc->set_req(TypeFunc::Memory, mem); 5687 set_control(init->proj_out_or_null(TypeFunc::Control)); 5688 set_i_o(callprojs->fallthrough_ioproj); 5689 5690 // Update memory as done in GraphKit::set_output_for_allocation() 5691 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 5692 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 5693 if (ary_type->isa_aryptr() && length_type != nullptr) { 5694 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 5695 } 5696 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 5697 int elemidx = C->get_alias_index(telemref); 5698 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw); 5699 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx); 5700 5701 Node* allocx = _gvn.transform(alloc); 5702 assert(allocx == alloc, "where has the allocation gone?"); 5703 assert(dest->is_CheckCastPP(), "not an allocation result?"); 5704 5705 _gvn.hash_delete(dest); 5706 dest->set_req(0, control()); 5707 Node* destx = _gvn.transform(dest); 5708 assert(destx == dest, "where has the allocation result gone?"); 5709 5710 array_ideal_length(alloc, ary_type, true); 5711 } 5712 } 5713 5714 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(), 5715 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary 5716 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array 5717 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter, 5718 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in 5719 // the interpreter similar to what we are doing for the newly emitted guards for the array copy. 5720 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc, 5721 JVMState* saved_jvms_before_guards) { 5722 if (saved_jvms_before_guards->map()->control()->is_IfProj()) { 5723 // There is at least one unrelated uncommon trap which needs to be replaced. 5724 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc); 5725 5726 JVMState* saved_jvms = jvms(); 5727 const int saved_reexecute_sp = _reexecute_sp; 5728 set_jvms(sfpt->jvms()); 5729 _reexecute_sp = jvms()->sp(); 5730 5731 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards); 5732 5733 // Restore state 5734 set_jvms(saved_jvms); 5735 _reexecute_sp = saved_reexecute_sp; 5736 } 5737 } 5738 5739 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon 5740 // traps will have the state of the array allocation. Let the old uncommon trap nodes die. 5741 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) { 5742 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards 5743 while (if_proj->is_IfProj()) { 5744 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj); 5745 if (uncommon_trap != nullptr) { 5746 create_new_uncommon_trap(uncommon_trap); 5747 } 5748 assert(if_proj->in(0)->is_If(), "must be If"); 5749 if_proj = if_proj->in(0)->in(0); 5750 } 5751 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(), 5752 "must have reached control projection of init node"); 5753 } 5754 5755 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) { 5756 const int trap_request = uncommon_trap_call->uncommon_trap_request(); 5757 assert(trap_request != 0, "no valid UCT trap request"); 5758 PreserveJVMState pjvms(this); 5759 set_control(uncommon_trap_call->in(0)); 5760 uncommon_trap(Deoptimization::trap_request_reason(trap_request), 5761 Deoptimization::trap_request_action(trap_request)); 5762 assert(stopped(), "Should be stopped"); 5763 _gvn.hash_delete(uncommon_trap_call); 5764 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it 5765 } 5766 5767 //------------------------------inline_array_partition----------------------- 5768 bool LibraryCallKit::inline_array_partition() { 5769 5770 Node* elementType = null_check(argument(0)); 5771 Node* obj = argument(1); 5772 Node* offset = argument(2); 5773 Node* fromIndex = argument(4); 5774 Node* toIndex = argument(5); 5775 Node* indexPivot1 = argument(6); 5776 Node* indexPivot2 = argument(7); 5777 5778 Node* pivotIndices = nullptr; 5779 5780 // Set the original stack and the reexecute bit for the interpreter to reexecute 5781 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens. 5782 { PreserveReexecuteState preexecs(this); 5783 jvms()->set_should_reexecute(true); 5784 5785 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr(); 5786 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type(); 5787 BasicType bt = elem_type->basic_type(); 5788 // Disable the intrinsic if the CPU does not support SIMD sort 5789 if (!Matcher::supports_simd_sort(bt)) { 5790 return false; 5791 } 5792 address stubAddr = nullptr; 5793 stubAddr = StubRoutines::select_array_partition_function(); 5794 // stub not loaded 5795 if (stubAddr == nullptr) { 5796 return false; 5797 } 5798 // get the address of the array 5799 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr(); 5800 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) { 5801 return false; // failed input validation 5802 } 5803 Node* obj_adr = make_unsafe_address(obj, offset); 5804 5805 // create the pivotIndices array of type int and size = 2 5806 Node* size = intcon(2); 5807 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT))); 5808 pivotIndices = new_array(klass_node, size, 0); // no arguments to push 5809 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices); 5810 guarantee(alloc != nullptr, "created above"); 5811 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT)); 5812 5813 // pass the basic type enum to the stub 5814 Node* elemType = intcon(bt); 5815 5816 // Call the stub 5817 const char *stubName = "array_partition_stub"; 5818 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(), 5819 stubAddr, stubName, TypePtr::BOTTOM, 5820 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr, 5821 indexPivot1, indexPivot2); 5822 5823 } // original reexecute is set back here 5824 5825 if (!stopped()) { 5826 set_result(pivotIndices); 5827 } 5828 5829 return true; 5830 } 5831 5832 5833 //------------------------------inline_array_sort----------------------- 5834 bool LibraryCallKit::inline_array_sort() { 5835 5836 Node* elementType = null_check(argument(0)); 5837 Node* obj = argument(1); 5838 Node* offset = argument(2); 5839 Node* fromIndex = argument(4); 5840 Node* toIndex = argument(5); 5841 5842 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr(); 5843 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type(); 5844 BasicType bt = elem_type->basic_type(); 5845 // Disable the intrinsic if the CPU does not support SIMD sort 5846 if (!Matcher::supports_simd_sort(bt)) { 5847 return false; 5848 } 5849 address stubAddr = nullptr; 5850 stubAddr = StubRoutines::select_arraysort_function(); 5851 //stub not loaded 5852 if (stubAddr == nullptr) { 5853 return false; 5854 } 5855 5856 // get address of the array 5857 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr(); 5858 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) { 5859 return false; // failed input validation 5860 } 5861 Node* obj_adr = make_unsafe_address(obj, offset); 5862 5863 // pass the basic type enum to the stub 5864 Node* elemType = intcon(bt); 5865 5866 // Call the stub. 5867 const char *stubName = "arraysort_stub"; 5868 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(), 5869 stubAddr, stubName, TypePtr::BOTTOM, 5870 obj_adr, elemType, fromIndex, toIndex); 5871 5872 return true; 5873 } 5874 5875 5876 //------------------------------inline_arraycopy----------------------- 5877 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 5878 // Object dest, int destPos, 5879 // int length); 5880 bool LibraryCallKit::inline_arraycopy() { 5881 // Get the arguments. 5882 Node* src = argument(0); // type: oop 5883 Node* src_offset = argument(1); // type: int 5884 Node* dest = argument(2); // type: oop 5885 Node* dest_offset = argument(3); // type: int 5886 Node* length = argument(4); // type: int 5887 5888 uint new_idx = C->unique(); 5889 5890 // Check for allocation before we add nodes that would confuse 5891 // tightly_coupled_allocation() 5892 AllocateArrayNode* alloc = tightly_coupled_allocation(dest); 5893 5894 int saved_reexecute_sp = -1; 5895 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 5896 // See arraycopy_restore_alloc_state() comment 5897 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards 5898 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation 5899 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards 5900 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr); 5901 5902 // The following tests must be performed 5903 // (1) src and dest are arrays. 5904 // (2) src and dest arrays must have elements of the same BasicType 5905 // (3) src and dest must not be null. 5906 // (4) src_offset must not be negative. 5907 // (5) dest_offset must not be negative. 5908 // (6) length must not be negative. 5909 // (7) src_offset + length must not exceed length of src. 5910 // (8) dest_offset + length must not exceed length of dest. 5911 // (9) each element of an oop array must be assignable 5912 5913 // (3) src and dest must not be null. 5914 // always do this here because we need the JVM state for uncommon traps 5915 Node* null_ctl = top(); 5916 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 5917 assert(null_ctl->is_top(), "no null control here"); 5918 dest = null_check(dest, T_ARRAY); 5919 5920 if (!can_emit_guards) { 5921 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any 5922 // guards but the arraycopy node could still take advantage of a 5923 // tightly allocated allocation. tightly_coupled_allocation() is 5924 // called again to make sure it takes the null check above into 5925 // account: the null check is mandatory and if it caused an 5926 // uncommon trap to be emitted then the allocation can't be 5927 // considered tightly coupled in this context. 5928 alloc = tightly_coupled_allocation(dest); 5929 } 5930 5931 bool validated = false; 5932 5933 const Type* src_type = _gvn.type(src); 5934 const Type* dest_type = _gvn.type(dest); 5935 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5936 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5937 5938 // Do we have the type of src? 5939 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM); 5940 // Do we have the type of dest? 5941 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM); 5942 // Is the type for src from speculation? 5943 bool src_spec = false; 5944 // Is the type for dest from speculation? 5945 bool dest_spec = false; 5946 5947 if ((!has_src || !has_dest) && can_emit_guards) { 5948 // We don't have sufficient type information, let's see if 5949 // speculative types can help. We need to have types for both src 5950 // and dest so that it pays off. 5951 5952 // Do we already have or could we have type information for src 5953 bool could_have_src = has_src; 5954 // Do we already have or could we have type information for dest 5955 bool could_have_dest = has_dest; 5956 5957 ciKlass* src_k = nullptr; 5958 if (!has_src) { 5959 src_k = src_type->speculative_type_not_null(); 5960 if (src_k != nullptr && src_k->is_array_klass()) { 5961 could_have_src = true; 5962 } 5963 } 5964 5965 ciKlass* dest_k = nullptr; 5966 if (!has_dest) { 5967 dest_k = dest_type->speculative_type_not_null(); 5968 if (dest_k != nullptr && dest_k->is_array_klass()) { 5969 could_have_dest = true; 5970 } 5971 } 5972 5973 if (could_have_src && could_have_dest) { 5974 // This is going to pay off so emit the required guards 5975 if (!has_src) { 5976 src = maybe_cast_profiled_obj(src, src_k, true); 5977 src_type = _gvn.type(src); 5978 top_src = src_type->isa_aryptr(); 5979 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM); 5980 src_spec = true; 5981 } 5982 if (!has_dest) { 5983 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5984 dest_type = _gvn.type(dest); 5985 top_dest = dest_type->isa_aryptr(); 5986 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM); 5987 dest_spec = true; 5988 } 5989 } 5990 } 5991 5992 if (has_src && has_dest && can_emit_guards) { 5993 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type(); 5994 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type(); 5995 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT; 5996 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT; 5997 5998 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) { 5999 // If both arrays are object arrays then having the exact types 6000 // for both will remove the need for a subtype check at runtime 6001 // before the call and may make it possible to pick a faster copy 6002 // routine (without a subtype check on every element) 6003 // Do we have the exact type of src? 6004 bool could_have_src = src_spec; 6005 // Do we have the exact type of dest? 6006 bool could_have_dest = dest_spec; 6007 ciKlass* src_k = nullptr; 6008 ciKlass* dest_k = nullptr; 6009 if (!src_spec) { 6010 src_k = src_type->speculative_type_not_null(); 6011 if (src_k != nullptr && src_k->is_array_klass()) { 6012 could_have_src = true; 6013 } 6014 } 6015 if (!dest_spec) { 6016 dest_k = dest_type->speculative_type_not_null(); 6017 if (dest_k != nullptr && dest_k->is_array_klass()) { 6018 could_have_dest = true; 6019 } 6020 } 6021 if (could_have_src && could_have_dest) { 6022 // If we can have both exact types, emit the missing guards 6023 if (could_have_src && !src_spec) { 6024 src = maybe_cast_profiled_obj(src, src_k, true); 6025 src_type = _gvn.type(src); 6026 top_src = src_type->isa_aryptr(); 6027 } 6028 if (could_have_dest && !dest_spec) { 6029 dest = maybe_cast_profiled_obj(dest, dest_k, true); 6030 dest_type = _gvn.type(dest); 6031 top_dest = dest_type->isa_aryptr(); 6032 } 6033 } 6034 } 6035 } 6036 6037 ciMethod* trap_method = method(); 6038 int trap_bci = bci(); 6039 if (saved_jvms_before_guards != nullptr) { 6040 trap_method = alloc->jvms()->method(); 6041 trap_bci = alloc->jvms()->bci(); 6042 } 6043 6044 bool negative_length_guard_generated = false; 6045 6046 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 6047 can_emit_guards && !src->is_top() && !dest->is_top()) { 6048 // validate arguments: enables transformation the ArrayCopyNode 6049 validated = true; 6050 6051 RegionNode* slow_region = new RegionNode(1); 6052 record_for_igvn(slow_region); 6053 6054 // (1) src and dest are arrays. 6055 generate_non_array_guard(load_object_klass(src), slow_region); 6056 generate_non_array_guard(load_object_klass(dest), slow_region); 6057 6058 // (2) src and dest arrays must have elements of the same BasicType 6059 // done at macro expansion or at Ideal transformation time 6060 6061 // (4) src_offset must not be negative. 6062 generate_negative_guard(src_offset, slow_region); 6063 6064 // (5) dest_offset must not be negative. 6065 generate_negative_guard(dest_offset, slow_region); 6066 6067 // (7) src_offset + length must not exceed length of src. 6068 generate_limit_guard(src_offset, length, 6069 load_array_length(src), 6070 slow_region); 6071 6072 // (8) dest_offset + length must not exceed length of dest. 6073 generate_limit_guard(dest_offset, length, 6074 load_array_length(dest), 6075 slow_region); 6076 6077 // (6) length must not be negative. 6078 // This is also checked in generate_arraycopy() during macro expansion, but 6079 // we also have to check it here for the case where the ArrayCopyNode will 6080 // be eliminated by Escape Analysis. 6081 if (EliminateAllocations) { 6082 generate_negative_guard(length, slow_region); 6083 negative_length_guard_generated = true; 6084 } 6085 6086 // (9) each element of an oop array must be assignable 6087 Node* dest_klass = load_object_klass(dest); 6088 if (src != dest) { 6089 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass); 6090 slow_region->add_req(not_subtype_ctrl); 6091 } 6092 6093 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr(); 6094 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type(); 6095 src = _gvn.transform(new CheckCastPPNode(control(), src, toop)); 6096 src_type = _gvn.type(src); 6097 top_src = src_type->isa_aryptr(); 6098 6099 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above) 6100 if (!stopped() && UseFlatArray) { 6101 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here. 6102 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat"); 6103 if (top_src != nullptr && top_src->is_flat()) { 6104 // Src is flat, check that dest is flat as well 6105 if (top_dest != nullptr && !top_dest->is_flat()) { 6106 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region); 6107 // Since dest is flat and src <: dest, dest must have the same type as src. 6108 top_dest = top_src->cast_to_exactness(false); 6109 assert(top_dest->is_flat(), "dest must be flat"); 6110 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest)); 6111 } 6112 } else if (top_src == nullptr || !top_src->is_not_flat()) { 6113 // Src might be flat and dest might not be flat. Go to the slow path if src is flat. 6114 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat. 6115 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat"); 6116 generate_fair_guard(flat_array_test(src), slow_region); 6117 if (top_src != nullptr) { 6118 top_src = top_src->cast_to_not_flat(); 6119 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src)); 6120 } 6121 } 6122 } 6123 6124 { 6125 PreserveJVMState pjvms(this); 6126 set_control(_gvn.transform(slow_region)); 6127 uncommon_trap(Deoptimization::Reason_intrinsic, 6128 Deoptimization::Action_make_not_entrant); 6129 assert(stopped(), "Should be stopped"); 6130 } 6131 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx); 6132 } 6133 6134 if (stopped()) { 6135 return true; 6136 } 6137 6138 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated, 6139 // Create LoadRange and LoadKlass nodes for use during macro expansion here 6140 // so the compiler has a chance to eliminate them: during macro expansion, 6141 // we have to set their control (CastPP nodes are eliminated). 6142 load_object_klass(src), load_object_klass(dest), 6143 load_array_length(src), load_array_length(dest)); 6144 6145 ac->set_arraycopy(validated); 6146 6147 Node* n = _gvn.transform(ac); 6148 if (n == ac) { 6149 ac->connect_outputs(this); 6150 } else { 6151 assert(validated, "shouldn't transform if all arguments not validated"); 6152 set_all_memory(n); 6153 } 6154 clear_upper_avx(); 6155 6156 6157 return true; 6158 } 6159 6160 6161 // Helper function which determines if an arraycopy immediately follows 6162 // an allocation, with no intervening tests or other escapes for the object. 6163 AllocateArrayNode* 6164 LibraryCallKit::tightly_coupled_allocation(Node* ptr) { 6165 if (stopped()) return nullptr; // no fast path 6166 if (!C->do_aliasing()) return nullptr; // no MergeMems around 6167 6168 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr); 6169 if (alloc == nullptr) return nullptr; 6170 6171 Node* rawmem = memory(Compile::AliasIdxRaw); 6172 // Is the allocation's memory state untouched? 6173 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 6174 // Bail out if there have been raw-memory effects since the allocation. 6175 // (Example: There might have been a call or safepoint.) 6176 return nullptr; 6177 } 6178 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 6179 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 6180 return nullptr; 6181 } 6182 6183 // There must be no unexpected observers of this allocation. 6184 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 6185 Node* obs = ptr->fast_out(i); 6186 if (obs != this->map()) { 6187 return nullptr; 6188 } 6189 } 6190 6191 // This arraycopy must unconditionally follow the allocation of the ptr. 6192 Node* alloc_ctl = ptr->in(0); 6193 Node* ctl = control(); 6194 while (ctl != alloc_ctl) { 6195 // There may be guards which feed into the slow_region. 6196 // Any other control flow means that we might not get a chance 6197 // to finish initializing the allocated object. 6198 // Various low-level checks bottom out in uncommon traps. These 6199 // are considered safe since we've already checked above that 6200 // there is no unexpected observer of this allocation. 6201 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) { 6202 assert(ctl->in(0)->is_If(), "must be If"); 6203 ctl = ctl->in(0)->in(0); 6204 } else { 6205 return nullptr; 6206 } 6207 } 6208 6209 // If we get this far, we have an allocation which immediately 6210 // precedes the arraycopy, and we can take over zeroing the new object. 6211 // The arraycopy will finish the initialization, and provide 6212 // a new control state to which we will anchor the destination pointer. 6213 6214 return alloc; 6215 } 6216 6217 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) { 6218 if (node->is_IfProj()) { 6219 Node* other_proj = node->as_IfProj()->other_if_proj(); 6220 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) { 6221 Node* obs = other_proj->fast_out(j); 6222 if (obs->in(0) == other_proj && obs->is_CallStaticJava() && 6223 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) { 6224 return obs->as_CallStaticJava(); 6225 } 6226 } 6227 } 6228 return nullptr; 6229 } 6230 6231 //-------------inline_encodeISOArray----------------------------------- 6232 // encode char[] to byte[] in ISO_8859_1 or ASCII 6233 bool LibraryCallKit::inline_encodeISOArray(bool ascii) { 6234 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 6235 // no receiver since it is static method 6236 Node *src = argument(0); 6237 Node *src_offset = argument(1); 6238 Node *dst = argument(2); 6239 Node *dst_offset = argument(3); 6240 Node *length = argument(4); 6241 6242 src = must_be_not_null(src, true); 6243 dst = must_be_not_null(dst, true); 6244 6245 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6246 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr(); 6247 if (src_type == nullptr || src_type->elem() == Type::BOTTOM || 6248 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) { 6249 // failed array check 6250 return false; 6251 } 6252 6253 // Figure out the size and type of the elements we will be copying. 6254 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6255 BasicType dst_elem = dst_type->elem()->array_element_basic_type(); 6256 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 6257 return false; 6258 } 6259 6260 Node* src_start = array_element_address(src, src_offset, T_CHAR); 6261 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 6262 // 'src_start' points to src array + scaled offset 6263 // 'dst_start' points to dst array + scaled offset 6264 6265 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 6266 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii); 6267 enc = _gvn.transform(enc); 6268 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 6269 set_memory(res_mem, mtype); 6270 set_result(enc); 6271 clear_upper_avx(); 6272 6273 return true; 6274 } 6275 6276 //-------------inline_multiplyToLen----------------------------------- 6277 bool LibraryCallKit::inline_multiplyToLen() { 6278 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 6279 6280 address stubAddr = StubRoutines::multiplyToLen(); 6281 if (stubAddr == nullptr) { 6282 return false; // Intrinsic's stub is not implemented on this platform 6283 } 6284 const char* stubName = "multiplyToLen"; 6285 6286 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 6287 6288 // no receiver because it is a static method 6289 Node* x = argument(0); 6290 Node* xlen = argument(1); 6291 Node* y = argument(2); 6292 Node* ylen = argument(3); 6293 Node* z = argument(4); 6294 6295 x = must_be_not_null(x, true); 6296 y = must_be_not_null(y, true); 6297 6298 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr(); 6299 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr(); 6300 if (x_type == nullptr || x_type->elem() == Type::BOTTOM || 6301 y_type == nullptr || y_type->elem() == Type::BOTTOM) { 6302 // failed array check 6303 return false; 6304 } 6305 6306 BasicType x_elem = x_type->elem()->array_element_basic_type(); 6307 BasicType y_elem = y_type->elem()->array_element_basic_type(); 6308 if (x_elem != T_INT || y_elem != T_INT) { 6309 return false; 6310 } 6311 6312 // Set the original stack and the reexecute bit for the interpreter to reexecute 6313 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens 6314 // on the return from z array allocation in runtime. 6315 { PreserveReexecuteState preexecs(this); 6316 jvms()->set_should_reexecute(true); 6317 6318 Node* x_start = array_element_address(x, intcon(0), x_elem); 6319 Node* y_start = array_element_address(y, intcon(0), y_elem); 6320 // 'x_start' points to x array + scaled xlen 6321 // 'y_start' points to y array + scaled ylen 6322 6323 // Allocate the result array 6324 Node* zlen = _gvn.transform(new AddINode(xlen, ylen)); 6325 ciKlass* klass = ciTypeArrayKlass::make(T_INT); 6326 Node* klass_node = makecon(TypeKlassPtr::make(klass)); 6327 6328 IdealKit ideal(this); 6329 6330 #define __ ideal. 6331 Node* one = __ ConI(1); 6332 Node* zero = __ ConI(0); 6333 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done(); 6334 __ set(need_alloc, zero); 6335 __ set(z_alloc, z); 6336 __ if_then(z, BoolTest::eq, null()); { 6337 __ increment (need_alloc, one); 6338 } __ else_(); { 6339 // Update graphKit memory and control from IdealKit. 6340 sync_kit(ideal); 6341 Node* cast = new CastPPNode(control(), z, TypePtr::NOTNULL); 6342 _gvn.set_type(cast, cast->bottom_type()); 6343 C->record_for_igvn(cast); 6344 6345 Node* zlen_arg = load_array_length(cast); 6346 // Update IdealKit memory and control from graphKit. 6347 __ sync_kit(this); 6348 __ if_then(zlen_arg, BoolTest::lt, zlen); { 6349 __ increment (need_alloc, one); 6350 } __ end_if(); 6351 } __ end_if(); 6352 6353 __ if_then(__ value(need_alloc), BoolTest::ne, zero); { 6354 // Update graphKit memory and control from IdealKit. 6355 sync_kit(ideal); 6356 Node * narr = new_array(klass_node, zlen, 1); 6357 // Update IdealKit memory and control from graphKit. 6358 __ sync_kit(this); 6359 __ set(z_alloc, narr); 6360 } __ end_if(); 6361 6362 sync_kit(ideal); 6363 z = __ value(z_alloc); 6364 // Can't use TypeAryPtr::INTS which uses Bottom offset. 6365 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass)); 6366 // Final sync IdealKit and GraphKit. 6367 final_sync(ideal); 6368 #undef __ 6369 6370 Node* z_start = array_element_address(z, intcon(0), T_INT); 6371 6372 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6373 OptoRuntime::multiplyToLen_Type(), 6374 stubAddr, stubName, TypePtr::BOTTOM, 6375 x_start, xlen, y_start, ylen, z_start, zlen); 6376 } // original reexecute is set back here 6377 6378 C->set_has_split_ifs(true); // Has chance for split-if optimization 6379 set_result(z); 6380 return true; 6381 } 6382 6383 //-------------inline_squareToLen------------------------------------ 6384 bool LibraryCallKit::inline_squareToLen() { 6385 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 6386 6387 address stubAddr = StubRoutines::squareToLen(); 6388 if (stubAddr == nullptr) { 6389 return false; // Intrinsic's stub is not implemented on this platform 6390 } 6391 const char* stubName = "squareToLen"; 6392 6393 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 6394 6395 Node* x = argument(0); 6396 Node* len = argument(1); 6397 Node* z = argument(2); 6398 Node* zlen = argument(3); 6399 6400 x = must_be_not_null(x, true); 6401 z = must_be_not_null(z, true); 6402 6403 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr(); 6404 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr(); 6405 if (x_type == nullptr || x_type->elem() == Type::BOTTOM || 6406 z_type == nullptr || z_type->elem() == Type::BOTTOM) { 6407 // failed array check 6408 return false; 6409 } 6410 6411 BasicType x_elem = x_type->elem()->array_element_basic_type(); 6412 BasicType z_elem = z_type->elem()->array_element_basic_type(); 6413 if (x_elem != T_INT || z_elem != T_INT) { 6414 return false; 6415 } 6416 6417 6418 Node* x_start = array_element_address(x, intcon(0), x_elem); 6419 Node* z_start = array_element_address(z, intcon(0), z_elem); 6420 6421 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6422 OptoRuntime::squareToLen_Type(), 6423 stubAddr, stubName, TypePtr::BOTTOM, 6424 x_start, len, z_start, zlen); 6425 6426 set_result(z); 6427 return true; 6428 } 6429 6430 //-------------inline_mulAdd------------------------------------------ 6431 bool LibraryCallKit::inline_mulAdd() { 6432 assert(UseMulAddIntrinsic, "not implemented on this platform"); 6433 6434 address stubAddr = StubRoutines::mulAdd(); 6435 if (stubAddr == nullptr) { 6436 return false; // Intrinsic's stub is not implemented on this platform 6437 } 6438 const char* stubName = "mulAdd"; 6439 6440 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 6441 6442 Node* out = argument(0); 6443 Node* in = argument(1); 6444 Node* offset = argument(2); 6445 Node* len = argument(3); 6446 Node* k = argument(4); 6447 6448 in = must_be_not_null(in, true); 6449 out = must_be_not_null(out, true); 6450 6451 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr(); 6452 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr(); 6453 if (out_type == nullptr || out_type->elem() == Type::BOTTOM || 6454 in_type == nullptr || in_type->elem() == Type::BOTTOM) { 6455 // failed array check 6456 return false; 6457 } 6458 6459 BasicType out_elem = out_type->elem()->array_element_basic_type(); 6460 BasicType in_elem = in_type->elem()->array_element_basic_type(); 6461 if (out_elem != T_INT || in_elem != T_INT) { 6462 return false; 6463 } 6464 6465 Node* outlen = load_array_length(out); 6466 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 6467 Node* out_start = array_element_address(out, intcon(0), out_elem); 6468 Node* in_start = array_element_address(in, intcon(0), in_elem); 6469 6470 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6471 OptoRuntime::mulAdd_Type(), 6472 stubAddr, stubName, TypePtr::BOTTOM, 6473 out_start,in_start, new_offset, len, k); 6474 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6475 set_result(result); 6476 return true; 6477 } 6478 6479 //-------------inline_montgomeryMultiply----------------------------------- 6480 bool LibraryCallKit::inline_montgomeryMultiply() { 6481 address stubAddr = StubRoutines::montgomeryMultiply(); 6482 if (stubAddr == nullptr) { 6483 return false; // Intrinsic's stub is not implemented on this platform 6484 } 6485 6486 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 6487 const char* stubName = "montgomery_multiply"; 6488 6489 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 6490 6491 Node* a = argument(0); 6492 Node* b = argument(1); 6493 Node* n = argument(2); 6494 Node* len = argument(3); 6495 Node* inv = argument(4); 6496 Node* m = argument(6); 6497 6498 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr(); 6499 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr(); 6500 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr(); 6501 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr(); 6502 if (a_type == nullptr || a_type->elem() == Type::BOTTOM || 6503 b_type == nullptr || b_type->elem() == Type::BOTTOM || 6504 n_type == nullptr || n_type->elem() == Type::BOTTOM || 6505 m_type == nullptr || m_type->elem() == Type::BOTTOM) { 6506 // failed array check 6507 return false; 6508 } 6509 6510 BasicType a_elem = a_type->elem()->array_element_basic_type(); 6511 BasicType b_elem = b_type->elem()->array_element_basic_type(); 6512 BasicType n_elem = n_type->elem()->array_element_basic_type(); 6513 BasicType m_elem = m_type->elem()->array_element_basic_type(); 6514 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 6515 return false; 6516 } 6517 6518 // Make the call 6519 { 6520 Node* a_start = array_element_address(a, intcon(0), a_elem); 6521 Node* b_start = array_element_address(b, intcon(0), b_elem); 6522 Node* n_start = array_element_address(n, intcon(0), n_elem); 6523 Node* m_start = array_element_address(m, intcon(0), m_elem); 6524 6525 Node* call = make_runtime_call(RC_LEAF, 6526 OptoRuntime::montgomeryMultiply_Type(), 6527 stubAddr, stubName, TypePtr::BOTTOM, 6528 a_start, b_start, n_start, len, inv, top(), 6529 m_start); 6530 set_result(m); 6531 } 6532 6533 return true; 6534 } 6535 6536 bool LibraryCallKit::inline_montgomerySquare() { 6537 address stubAddr = StubRoutines::montgomerySquare(); 6538 if (stubAddr == nullptr) { 6539 return false; // Intrinsic's stub is not implemented on this platform 6540 } 6541 6542 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 6543 const char* stubName = "montgomery_square"; 6544 6545 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 6546 6547 Node* a = argument(0); 6548 Node* n = argument(1); 6549 Node* len = argument(2); 6550 Node* inv = argument(3); 6551 Node* m = argument(5); 6552 6553 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr(); 6554 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr(); 6555 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr(); 6556 if (a_type == nullptr || a_type->elem() == Type::BOTTOM || 6557 n_type == nullptr || n_type->elem() == Type::BOTTOM || 6558 m_type == nullptr || m_type->elem() == Type::BOTTOM) { 6559 // failed array check 6560 return false; 6561 } 6562 6563 BasicType a_elem = a_type->elem()->array_element_basic_type(); 6564 BasicType n_elem = n_type->elem()->array_element_basic_type(); 6565 BasicType m_elem = m_type->elem()->array_element_basic_type(); 6566 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 6567 return false; 6568 } 6569 6570 // Make the call 6571 { 6572 Node* a_start = array_element_address(a, intcon(0), a_elem); 6573 Node* n_start = array_element_address(n, intcon(0), n_elem); 6574 Node* m_start = array_element_address(m, intcon(0), m_elem); 6575 6576 Node* call = make_runtime_call(RC_LEAF, 6577 OptoRuntime::montgomerySquare_Type(), 6578 stubAddr, stubName, TypePtr::BOTTOM, 6579 a_start, n_start, len, inv, top(), 6580 m_start); 6581 set_result(m); 6582 } 6583 6584 return true; 6585 } 6586 6587 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) { 6588 address stubAddr = nullptr; 6589 const char* stubName = nullptr; 6590 6591 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift(); 6592 if (stubAddr == nullptr) { 6593 return false; // Intrinsic's stub is not implemented on this platform 6594 } 6595 6596 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker"; 6597 6598 assert(callee()->signature()->size() == 5, "expected 5 arguments"); 6599 6600 Node* newArr = argument(0); 6601 Node* oldArr = argument(1); 6602 Node* newIdx = argument(2); 6603 Node* shiftCount = argument(3); 6604 Node* numIter = argument(4); 6605 6606 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr(); 6607 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr(); 6608 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM || 6609 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) { 6610 return false; 6611 } 6612 6613 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type(); 6614 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type(); 6615 if (newArr_elem != T_INT || oldArr_elem != T_INT) { 6616 return false; 6617 } 6618 6619 // Make the call 6620 { 6621 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem); 6622 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem); 6623 6624 Node* call = make_runtime_call(RC_LEAF, 6625 OptoRuntime::bigIntegerShift_Type(), 6626 stubAddr, 6627 stubName, 6628 TypePtr::BOTTOM, 6629 newArr_start, 6630 oldArr_start, 6631 newIdx, 6632 shiftCount, 6633 numIter); 6634 } 6635 6636 return true; 6637 } 6638 6639 //-------------inline_vectorizedMismatch------------------------------ 6640 bool LibraryCallKit::inline_vectorizedMismatch() { 6641 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform"); 6642 6643 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 6644 Node* obja = argument(0); // Object 6645 Node* aoffset = argument(1); // long 6646 Node* objb = argument(3); // Object 6647 Node* boffset = argument(4); // long 6648 Node* length = argument(6); // int 6649 Node* scale = argument(7); // int 6650 6651 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr(); 6652 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr(); 6653 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM || 6654 objb_t == nullptr || objb_t->elem() == Type::BOTTOM || 6655 scale == top()) { 6656 return false; // failed input validation 6657 } 6658 6659 Node* obja_adr = make_unsafe_address(obja, aoffset); 6660 Node* objb_adr = make_unsafe_address(objb, boffset); 6661 6662 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size. 6663 // 6664 // inline_limit = ArrayOperationPartialInlineSize / element_size; 6665 // if (length <= inline_limit) { 6666 // inline_path: 6667 // vmask = VectorMaskGen length 6668 // vload1 = LoadVectorMasked obja, vmask 6669 // vload2 = LoadVectorMasked objb, vmask 6670 // result1 = VectorCmpMasked vload1, vload2, vmask 6671 // } else { 6672 // call_stub_path: 6673 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale) 6674 // } 6675 // exit_block: 6676 // return Phi(result1, result2); 6677 // 6678 enum { inline_path = 1, // input is small enough to process it all at once 6679 stub_path = 2, // input is too large; call into the VM 6680 PATH_LIMIT = 3 6681 }; 6682 6683 Node* exit_block = new RegionNode(PATH_LIMIT); 6684 Node* result_phi = new PhiNode(exit_block, TypeInt::INT); 6685 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM); 6686 6687 Node* call_stub_path = control(); 6688 6689 BasicType elem_bt = T_ILLEGAL; 6690 6691 const TypeInt* scale_t = _gvn.type(scale)->is_int(); 6692 if (scale_t->is_con()) { 6693 switch (scale_t->get_con()) { 6694 case 0: elem_bt = T_BYTE; break; 6695 case 1: elem_bt = T_SHORT; break; 6696 case 2: elem_bt = T_INT; break; 6697 case 3: elem_bt = T_LONG; break; 6698 6699 default: elem_bt = T_ILLEGAL; break; // not supported 6700 } 6701 } 6702 6703 int inline_limit = 0; 6704 bool do_partial_inline = false; 6705 6706 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) { 6707 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt); 6708 do_partial_inline = inline_limit >= 16; 6709 } 6710 6711 if (do_partial_inline) { 6712 assert(elem_bt != T_ILLEGAL, "sanity"); 6713 6714 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) && 6715 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) && 6716 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) { 6717 6718 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit); 6719 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit))); 6720 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt)); 6721 6722 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN); 6723 6724 if (!stopped()) { 6725 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin))); 6726 6727 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr(); 6728 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr(); 6729 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t)); 6730 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t)); 6731 6732 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt)); 6733 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask)); 6734 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask)); 6735 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT)); 6736 6737 exit_block->init_req(inline_path, control()); 6738 memory_phi->init_req(inline_path, map()->memory()); 6739 result_phi->init_req(inline_path, result); 6740 6741 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size())); 6742 clear_upper_avx(); 6743 } 6744 } 6745 } 6746 6747 if (call_stub_path != nullptr) { 6748 set_control(call_stub_path); 6749 6750 Node* call = make_runtime_call(RC_LEAF, 6751 OptoRuntime::vectorizedMismatch_Type(), 6752 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM, 6753 obja_adr, objb_adr, length, scale); 6754 6755 exit_block->init_req(stub_path, control()); 6756 memory_phi->init_req(stub_path, map()->memory()); 6757 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms))); 6758 } 6759 6760 exit_block = _gvn.transform(exit_block); 6761 memory_phi = _gvn.transform(memory_phi); 6762 result_phi = _gvn.transform(result_phi); 6763 6764 set_control(exit_block); 6765 set_all_memory(memory_phi); 6766 set_result(result_phi); 6767 6768 return true; 6769 } 6770 6771 //------------------------------inline_vectorizedHashcode---------------------------- 6772 bool LibraryCallKit::inline_vectorizedHashCode() { 6773 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform"); 6774 6775 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters"); 6776 Node* array = argument(0); 6777 Node* offset = argument(1); 6778 Node* length = argument(2); 6779 Node* initialValue = argument(3); 6780 Node* basic_type = argument(4); 6781 6782 if (basic_type == top()) { 6783 return false; // failed input validation 6784 } 6785 6786 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int(); 6787 if (!basic_type_t->is_con()) { 6788 return false; // Only intrinsify if mode argument is constant 6789 } 6790 6791 array = must_be_not_null(array, true); 6792 6793 BasicType bt = (BasicType)basic_type_t->get_con(); 6794 6795 // Resolve address of first element 6796 Node* array_start = array_element_address(array, offset, bt); 6797 6798 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)), 6799 array_start, length, initialValue, basic_type))); 6800 clear_upper_avx(); 6801 6802 return true; 6803 } 6804 6805 /** 6806 * Calculate CRC32 for byte. 6807 * int java.util.zip.CRC32.update(int crc, int b) 6808 */ 6809 bool LibraryCallKit::inline_updateCRC32() { 6810 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 6811 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 6812 // no receiver since it is static method 6813 Node* crc = argument(0); // type: int 6814 Node* b = argument(1); // type: int 6815 6816 /* 6817 * int c = ~ crc; 6818 * b = timesXtoThe32[(b ^ c) & 0xFF]; 6819 * b = b ^ (c >>> 8); 6820 * crc = ~b; 6821 */ 6822 6823 Node* M1 = intcon(-1); 6824 crc = _gvn.transform(new XorINode(crc, M1)); 6825 Node* result = _gvn.transform(new XorINode(crc, b)); 6826 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 6827 6828 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 6829 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 6830 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 6831 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 6832 6833 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 6834 result = _gvn.transform(new XorINode(crc, result)); 6835 result = _gvn.transform(new XorINode(result, M1)); 6836 set_result(result); 6837 return true; 6838 } 6839 6840 /** 6841 * Calculate CRC32 for byte[] array. 6842 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 6843 */ 6844 bool LibraryCallKit::inline_updateBytesCRC32() { 6845 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 6846 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6847 // no receiver since it is static method 6848 Node* crc = argument(0); // type: int 6849 Node* src = argument(1); // type: oop 6850 Node* offset = argument(2); // type: int 6851 Node* length = argument(3); // type: int 6852 6853 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6854 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6855 // failed array check 6856 return false; 6857 } 6858 6859 // Figure out the size and type of the elements we will be copying. 6860 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6861 if (src_elem != T_BYTE) { 6862 return false; 6863 } 6864 6865 // 'src_start' points to src array + scaled offset 6866 src = must_be_not_null(src, true); 6867 Node* src_start = array_element_address(src, offset, src_elem); 6868 6869 // We assume that range check is done by caller. 6870 // TODO: generate range check (offset+length < src.length) in debug VM. 6871 6872 // Call the stub. 6873 address stubAddr = StubRoutines::updateBytesCRC32(); 6874 const char *stubName = "updateBytesCRC32"; 6875 6876 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 6877 stubAddr, stubName, TypePtr::BOTTOM, 6878 crc, src_start, length); 6879 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6880 set_result(result); 6881 return true; 6882 } 6883 6884 /** 6885 * Calculate CRC32 for ByteBuffer. 6886 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 6887 */ 6888 bool LibraryCallKit::inline_updateByteBufferCRC32() { 6889 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 6890 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 6891 // no receiver since it is static method 6892 Node* crc = argument(0); // type: int 6893 Node* src = argument(1); // type: long 6894 Node* offset = argument(3); // type: int 6895 Node* length = argument(4); // type: int 6896 6897 src = ConvL2X(src); // adjust Java long to machine word 6898 Node* base = _gvn.transform(new CastX2PNode(src)); 6899 offset = ConvI2X(offset); 6900 6901 // 'src_start' points to src array + scaled offset 6902 Node* src_start = basic_plus_adr(top(), base, offset); 6903 6904 // Call the stub. 6905 address stubAddr = StubRoutines::updateBytesCRC32(); 6906 const char *stubName = "updateBytesCRC32"; 6907 6908 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 6909 stubAddr, stubName, TypePtr::BOTTOM, 6910 crc, src_start, length); 6911 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6912 set_result(result); 6913 return true; 6914 } 6915 6916 //------------------------------get_table_from_crc32c_class----------------------- 6917 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 6918 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class); 6919 assert (table != nullptr, "wrong version of java.util.zip.CRC32C"); 6920 6921 return table; 6922 } 6923 6924 //------------------------------inline_updateBytesCRC32C----------------------- 6925 // 6926 // Calculate CRC32C for byte[] array. 6927 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 6928 // 6929 bool LibraryCallKit::inline_updateBytesCRC32C() { 6930 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6931 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6932 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6933 // no receiver since it is a static method 6934 Node* crc = argument(0); // type: int 6935 Node* src = argument(1); // type: oop 6936 Node* offset = argument(2); // type: int 6937 Node* end = argument(3); // type: int 6938 6939 Node* length = _gvn.transform(new SubINode(end, offset)); 6940 6941 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6942 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6943 // failed array check 6944 return false; 6945 } 6946 6947 // Figure out the size and type of the elements we will be copying. 6948 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6949 if (src_elem != T_BYTE) { 6950 return false; 6951 } 6952 6953 // 'src_start' points to src array + scaled offset 6954 src = must_be_not_null(src, true); 6955 Node* src_start = array_element_address(src, offset, src_elem); 6956 6957 // static final int[] byteTable in class CRC32C 6958 Node* table = get_table_from_crc32c_class(callee()->holder()); 6959 table = must_be_not_null(table, true); 6960 Node* table_start = array_element_address(table, intcon(0), T_INT); 6961 6962 // We assume that range check is done by caller. 6963 // TODO: generate range check (offset+length < src.length) in debug VM. 6964 6965 // Call the stub. 6966 address stubAddr = StubRoutines::updateBytesCRC32C(); 6967 const char *stubName = "updateBytesCRC32C"; 6968 6969 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 6970 stubAddr, stubName, TypePtr::BOTTOM, 6971 crc, src_start, length, table_start); 6972 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6973 set_result(result); 6974 return true; 6975 } 6976 6977 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 6978 // 6979 // Calculate CRC32C for DirectByteBuffer. 6980 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 6981 // 6982 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 6983 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6984 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 6985 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6986 // no receiver since it is a static method 6987 Node* crc = argument(0); // type: int 6988 Node* src = argument(1); // type: long 6989 Node* offset = argument(3); // type: int 6990 Node* end = argument(4); // type: int 6991 6992 Node* length = _gvn.transform(new SubINode(end, offset)); 6993 6994 src = ConvL2X(src); // adjust Java long to machine word 6995 Node* base = _gvn.transform(new CastX2PNode(src)); 6996 offset = ConvI2X(offset); 6997 6998 // 'src_start' points to src array + scaled offset 6999 Node* src_start = basic_plus_adr(top(), base, offset); 7000 7001 // static final int[] byteTable in class CRC32C 7002 Node* table = get_table_from_crc32c_class(callee()->holder()); 7003 table = must_be_not_null(table, true); 7004 Node* table_start = array_element_address(table, intcon(0), T_INT); 7005 7006 // Call the stub. 7007 address stubAddr = StubRoutines::updateBytesCRC32C(); 7008 const char *stubName = "updateBytesCRC32C"; 7009 7010 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 7011 stubAddr, stubName, TypePtr::BOTTOM, 7012 crc, src_start, length, table_start); 7013 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 7014 set_result(result); 7015 return true; 7016 } 7017 7018 //------------------------------inline_updateBytesAdler32---------------------- 7019 // 7020 // Calculate Adler32 checksum for byte[] array. 7021 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 7022 // 7023 bool LibraryCallKit::inline_updateBytesAdler32() { 7024 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one 7025 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 7026 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 7027 // no receiver since it is static method 7028 Node* crc = argument(0); // type: int 7029 Node* src = argument(1); // type: oop 7030 Node* offset = argument(2); // type: int 7031 Node* length = argument(3); // type: int 7032 7033 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7034 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 7035 // failed array check 7036 return false; 7037 } 7038 7039 // Figure out the size and type of the elements we will be copying. 7040 BasicType src_elem = src_type->elem()->array_element_basic_type(); 7041 if (src_elem != T_BYTE) { 7042 return false; 7043 } 7044 7045 // 'src_start' points to src array + scaled offset 7046 Node* src_start = array_element_address(src, offset, src_elem); 7047 7048 // We assume that range check is done by caller. 7049 // TODO: generate range check (offset+length < src.length) in debug VM. 7050 7051 // Call the stub. 7052 address stubAddr = StubRoutines::updateBytesAdler32(); 7053 const char *stubName = "updateBytesAdler32"; 7054 7055 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 7056 stubAddr, stubName, TypePtr::BOTTOM, 7057 crc, src_start, length); 7058 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 7059 set_result(result); 7060 return true; 7061 } 7062 7063 //------------------------------inline_updateByteBufferAdler32--------------- 7064 // 7065 // Calculate Adler32 checksum for DirectByteBuffer. 7066 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 7067 // 7068 bool LibraryCallKit::inline_updateByteBufferAdler32() { 7069 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one 7070 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 7071 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 7072 // no receiver since it is static method 7073 Node* crc = argument(0); // type: int 7074 Node* src = argument(1); // type: long 7075 Node* offset = argument(3); // type: int 7076 Node* length = argument(4); // type: int 7077 7078 src = ConvL2X(src); // adjust Java long to machine word 7079 Node* base = _gvn.transform(new CastX2PNode(src)); 7080 offset = ConvI2X(offset); 7081 7082 // 'src_start' points to src array + scaled offset 7083 Node* src_start = basic_plus_adr(top(), base, offset); 7084 7085 // Call the stub. 7086 address stubAddr = StubRoutines::updateBytesAdler32(); 7087 const char *stubName = "updateBytesAdler32"; 7088 7089 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 7090 stubAddr, stubName, TypePtr::BOTTOM, 7091 crc, src_start, length); 7092 7093 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 7094 set_result(result); 7095 return true; 7096 } 7097 7098 //----------------------------inline_reference_get---------------------------- 7099 // public T java.lang.ref.Reference.get(); 7100 bool LibraryCallKit::inline_reference_get() { 7101 const int referent_offset = java_lang_ref_Reference::referent_offset(); 7102 7103 // Get the argument: 7104 Node* reference_obj = null_check_receiver(); 7105 if (stopped()) return true; 7106 7107 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF; 7108 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;", 7109 decorators, /*is_static*/ false, nullptr); 7110 if (result == nullptr) return false; 7111 7112 // Add memory barrier to prevent commoning reads from this field 7113 // across safepoint since GC can change its value. 7114 insert_mem_bar(Op_MemBarCPUOrder); 7115 7116 set_result(result); 7117 return true; 7118 } 7119 7120 //----------------------------inline_reference_refersTo0---------------------------- 7121 // bool java.lang.ref.Reference.refersTo0(); 7122 // bool java.lang.ref.PhantomReference.refersTo0(); 7123 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) { 7124 // Get arguments: 7125 Node* reference_obj = null_check_receiver(); 7126 Node* other_obj = argument(1); 7127 if (stopped()) return true; 7128 7129 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE; 7130 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF); 7131 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;", 7132 decorators, /*is_static*/ false, nullptr); 7133 if (referent == nullptr) return false; 7134 7135 // Add memory barrier to prevent commoning reads from this field 7136 // across safepoint since GC can change its value. 7137 insert_mem_bar(Op_MemBarCPUOrder); 7138 7139 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj)); 7140 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 7141 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN); 7142 7143 RegionNode* region = new RegionNode(3); 7144 PhiNode* phi = new PhiNode(region, TypeInt::BOOL); 7145 7146 Node* if_true = _gvn.transform(new IfTrueNode(if_node)); 7147 region->init_req(1, if_true); 7148 phi->init_req(1, intcon(1)); 7149 7150 Node* if_false = _gvn.transform(new IfFalseNode(if_node)); 7151 region->init_req(2, if_false); 7152 phi->init_req(2, intcon(0)); 7153 7154 set_control(_gvn.transform(region)); 7155 record_for_igvn(region); 7156 set_result(_gvn.transform(phi)); 7157 return true; 7158 } 7159 7160 7161 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString, 7162 DecoratorSet decorators, bool is_static, 7163 ciInstanceKlass* fromKls) { 7164 if (fromKls == nullptr) { 7165 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 7166 assert(tinst != nullptr, "obj is null"); 7167 assert(tinst->is_loaded(), "obj is not loaded"); 7168 fromKls = tinst->instance_klass(); 7169 } else { 7170 assert(is_static, "only for static field access"); 7171 } 7172 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 7173 ciSymbol::make(fieldTypeString), 7174 is_static); 7175 7176 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName); 7177 if (field == nullptr) return (Node *) nullptr; 7178 7179 if (is_static) { 7180 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 7181 fromObj = makecon(tip); 7182 } 7183 7184 // Next code copied from Parse::do_get_xxx(): 7185 7186 // Compute address and memory type. 7187 int offset = field->offset_in_bytes(); 7188 bool is_vol = field->is_volatile(); 7189 ciType* field_klass = field->type(); 7190 assert(field_klass->is_loaded(), "should be loaded"); 7191 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 7192 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 7193 BasicType bt = field->layout_type(); 7194 7195 // Build the resultant type of the load 7196 const Type *type; 7197 if (bt == T_OBJECT) { 7198 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 7199 } else { 7200 type = Type::get_const_basic_type(bt); 7201 } 7202 7203 if (is_vol) { 7204 decorators |= MO_SEQ_CST; 7205 } 7206 7207 return access_load_at(fromObj, adr, adr_type, type, bt, decorators); 7208 } 7209 7210 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 7211 bool is_exact /* true */, bool is_static /* false */, 7212 ciInstanceKlass * fromKls /* nullptr */) { 7213 if (fromKls == nullptr) { 7214 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 7215 assert(tinst != nullptr, "obj is null"); 7216 assert(tinst->is_loaded(), "obj is not loaded"); 7217 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 7218 fromKls = tinst->instance_klass(); 7219 } 7220 else { 7221 assert(is_static, "only for static field access"); 7222 } 7223 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 7224 ciSymbol::make(fieldTypeString), 7225 is_static); 7226 7227 assert(field != nullptr, "undefined field"); 7228 assert(!field->is_volatile(), "not defined for volatile fields"); 7229 7230 if (is_static) { 7231 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 7232 fromObj = makecon(tip); 7233 } 7234 7235 // Next code copied from Parse::do_get_xxx(): 7236 7237 // Compute address and memory type. 7238 int offset = field->offset_in_bytes(); 7239 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 7240 7241 return adr; 7242 } 7243 7244 //------------------------------inline_aescrypt_Block----------------------- 7245 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 7246 address stubAddr = nullptr; 7247 const char *stubName; 7248 assert(UseAES, "need AES instruction support"); 7249 7250 switch(id) { 7251 case vmIntrinsics::_aescrypt_encryptBlock: 7252 stubAddr = StubRoutines::aescrypt_encryptBlock(); 7253 stubName = "aescrypt_encryptBlock"; 7254 break; 7255 case vmIntrinsics::_aescrypt_decryptBlock: 7256 stubAddr = StubRoutines::aescrypt_decryptBlock(); 7257 stubName = "aescrypt_decryptBlock"; 7258 break; 7259 default: 7260 break; 7261 } 7262 if (stubAddr == nullptr) return false; 7263 7264 Node* aescrypt_object = argument(0); 7265 Node* src = argument(1); 7266 Node* src_offset = argument(2); 7267 Node* dest = argument(3); 7268 Node* dest_offset = argument(4); 7269 7270 src = must_be_not_null(src, true); 7271 dest = must_be_not_null(dest, true); 7272 7273 // (1) src and dest are arrays. 7274 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7275 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7276 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7277 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7278 7279 // for the quick and dirty code we will skip all the checks. 7280 // we are just trying to get the call to be generated. 7281 Node* src_start = src; 7282 Node* dest_start = dest; 7283 if (src_offset != nullptr || dest_offset != nullptr) { 7284 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7285 src_start = array_element_address(src, src_offset, T_BYTE); 7286 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7287 } 7288 7289 // now need to get the start of its expanded key array 7290 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7291 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7292 if (k_start == nullptr) return false; 7293 7294 // Call the stub. 7295 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 7296 stubAddr, stubName, TypePtr::BOTTOM, 7297 src_start, dest_start, k_start); 7298 7299 return true; 7300 } 7301 7302 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 7303 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 7304 address stubAddr = nullptr; 7305 const char *stubName = nullptr; 7306 7307 assert(UseAES, "need AES instruction support"); 7308 7309 switch(id) { 7310 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 7311 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 7312 stubName = "cipherBlockChaining_encryptAESCrypt"; 7313 break; 7314 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 7315 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 7316 stubName = "cipherBlockChaining_decryptAESCrypt"; 7317 break; 7318 default: 7319 break; 7320 } 7321 if (stubAddr == nullptr) return false; 7322 7323 Node* cipherBlockChaining_object = argument(0); 7324 Node* src = argument(1); 7325 Node* src_offset = argument(2); 7326 Node* len = argument(3); 7327 Node* dest = argument(4); 7328 Node* dest_offset = argument(5); 7329 7330 src = must_be_not_null(src, false); 7331 dest = must_be_not_null(dest, false); 7332 7333 // (1) src and dest are arrays. 7334 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7335 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7336 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7337 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7338 7339 // checks are the responsibility of the caller 7340 Node* src_start = src; 7341 Node* dest_start = dest; 7342 if (src_offset != nullptr || dest_offset != nullptr) { 7343 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7344 src_start = array_element_address(src, src_offset, T_BYTE); 7345 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7346 } 7347 7348 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7349 // (because of the predicated logic executed earlier). 7350 // so we cast it here safely. 7351 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7352 7353 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7354 if (embeddedCipherObj == nullptr) return false; 7355 7356 // cast it to what we know it will be at runtime 7357 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 7358 assert(tinst != nullptr, "CBC obj is null"); 7359 assert(tinst->is_loaded(), "CBC obj is not loaded"); 7360 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7361 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7362 7363 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7364 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7365 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7366 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7367 aescrypt_object = _gvn.transform(aescrypt_object); 7368 7369 // we need to get the start of the aescrypt_object's expanded key array 7370 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7371 if (k_start == nullptr) return false; 7372 7373 // similarly, get the start address of the r vector 7374 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B"); 7375 if (objRvec == nullptr) return false; 7376 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 7377 7378 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7379 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 7380 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 7381 stubAddr, stubName, TypePtr::BOTTOM, 7382 src_start, dest_start, k_start, r_start, len); 7383 7384 // return cipher length (int) 7385 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 7386 set_result(retvalue); 7387 return true; 7388 } 7389 7390 //------------------------------inline_electronicCodeBook_AESCrypt----------------------- 7391 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) { 7392 address stubAddr = nullptr; 7393 const char *stubName = nullptr; 7394 7395 assert(UseAES, "need AES instruction support"); 7396 7397 switch (id) { 7398 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 7399 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt(); 7400 stubName = "electronicCodeBook_encryptAESCrypt"; 7401 break; 7402 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 7403 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt(); 7404 stubName = "electronicCodeBook_decryptAESCrypt"; 7405 break; 7406 default: 7407 break; 7408 } 7409 7410 if (stubAddr == nullptr) return false; 7411 7412 Node* electronicCodeBook_object = argument(0); 7413 Node* src = argument(1); 7414 Node* src_offset = argument(2); 7415 Node* len = argument(3); 7416 Node* dest = argument(4); 7417 Node* dest_offset = argument(5); 7418 7419 // (1) src and dest are arrays. 7420 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7421 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7422 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7423 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7424 7425 // checks are the responsibility of the caller 7426 Node* src_start = src; 7427 Node* dest_start = dest; 7428 if (src_offset != nullptr || dest_offset != nullptr) { 7429 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7430 src_start = array_element_address(src, src_offset, T_BYTE); 7431 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7432 } 7433 7434 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7435 // (because of the predicated logic executed earlier). 7436 // so we cast it here safely. 7437 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7438 7439 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7440 if (embeddedCipherObj == nullptr) return false; 7441 7442 // cast it to what we know it will be at runtime 7443 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr(); 7444 assert(tinst != nullptr, "ECB obj is null"); 7445 assert(tinst->is_loaded(), "ECB obj is not loaded"); 7446 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7447 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7448 7449 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7450 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7451 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7452 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7453 aescrypt_object = _gvn.transform(aescrypt_object); 7454 7455 // we need to get the start of the aescrypt_object's expanded key array 7456 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7457 if (k_start == nullptr) return false; 7458 7459 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7460 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP, 7461 OptoRuntime::electronicCodeBook_aescrypt_Type(), 7462 stubAddr, stubName, TypePtr::BOTTOM, 7463 src_start, dest_start, k_start, len); 7464 7465 // return cipher length (int) 7466 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms)); 7467 set_result(retvalue); 7468 return true; 7469 } 7470 7471 //------------------------------inline_counterMode_AESCrypt----------------------- 7472 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 7473 assert(UseAES, "need AES instruction support"); 7474 if (!UseAESCTRIntrinsics) return false; 7475 7476 address stubAddr = nullptr; 7477 const char *stubName = nullptr; 7478 if (id == vmIntrinsics::_counterMode_AESCrypt) { 7479 stubAddr = StubRoutines::counterMode_AESCrypt(); 7480 stubName = "counterMode_AESCrypt"; 7481 } 7482 if (stubAddr == nullptr) return false; 7483 7484 Node* counterMode_object = argument(0); 7485 Node* src = argument(1); 7486 Node* src_offset = argument(2); 7487 Node* len = argument(3); 7488 Node* dest = argument(4); 7489 Node* dest_offset = argument(5); 7490 7491 // (1) src and dest are arrays. 7492 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7493 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7494 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7495 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7496 7497 // checks are the responsibility of the caller 7498 Node* src_start = src; 7499 Node* dest_start = dest; 7500 if (src_offset != nullptr || dest_offset != nullptr) { 7501 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7502 src_start = array_element_address(src, src_offset, T_BYTE); 7503 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7504 } 7505 7506 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7507 // (because of the predicated logic executed earlier). 7508 // so we cast it here safely. 7509 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7510 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7511 if (embeddedCipherObj == nullptr) return false; 7512 // cast it to what we know it will be at runtime 7513 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 7514 assert(tinst != nullptr, "CTR obj is null"); 7515 assert(tinst->is_loaded(), "CTR obj is not loaded"); 7516 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7517 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7518 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7519 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7520 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7521 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7522 aescrypt_object = _gvn.transform(aescrypt_object); 7523 // we need to get the start of the aescrypt_object's expanded key array 7524 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7525 if (k_start == nullptr) return false; 7526 // similarly, get the start address of the r vector 7527 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B"); 7528 if (obj_counter == nullptr) return false; 7529 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 7530 7531 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B"); 7532 if (saved_encCounter == nullptr) return false; 7533 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 7534 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 7535 7536 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7537 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 7538 OptoRuntime::counterMode_aescrypt_Type(), 7539 stubAddr, stubName, TypePtr::BOTTOM, 7540 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 7541 7542 // return cipher length (int) 7543 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 7544 set_result(retvalue); 7545 return true; 7546 } 7547 7548 //------------------------------get_key_start_from_aescrypt_object----------------------- 7549 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 7550 #if defined(PPC64) || defined(S390) 7551 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys. 7552 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns. 7553 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption. 7554 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]). 7555 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I"); 7556 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt"); 7557 if (objSessionK == nullptr) { 7558 return (Node *) nullptr; 7559 } 7560 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true); 7561 #else 7562 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I"); 7563 #endif // PPC64 7564 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt"); 7565 if (objAESCryptKey == nullptr) return (Node *) nullptr; 7566 7567 // now have the array, need to get the start address of the K array 7568 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 7569 return k_start; 7570 } 7571 7572 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 7573 // Return node representing slow path of predicate check. 7574 // the pseudo code we want to emulate with this predicate is: 7575 // for encryption: 7576 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7577 // for decryption: 7578 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7579 // note cipher==plain is more conservative than the original java code but that's OK 7580 // 7581 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 7582 // The receiver was checked for null already. 7583 Node* objCBC = argument(0); 7584 7585 Node* src = argument(1); 7586 Node* dest = argument(4); 7587 7588 // Load embeddedCipher field of CipherBlockChaining object. 7589 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7590 7591 // get AESCrypt klass for instanceOf check 7592 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7593 // will have same classloader as CipherBlockChaining object 7594 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 7595 assert(tinst != nullptr, "CBCobj is null"); 7596 assert(tinst->is_loaded(), "CBCobj is not loaded"); 7597 7598 // we want to do an instanceof comparison against the AESCrypt class 7599 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7600 if (!klass_AESCrypt->is_loaded()) { 7601 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7602 Node* ctrl = control(); 7603 set_control(top()); // no regular fast path 7604 return ctrl; 7605 } 7606 7607 src = must_be_not_null(src, true); 7608 dest = must_be_not_null(dest, true); 7609 7610 // Resolve oops to stable for CmpP below. 7611 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7612 7613 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7614 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7615 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7616 7617 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7618 7619 // for encryption, we are done 7620 if (!decrypting) 7621 return instof_false; // even if it is null 7622 7623 // for decryption, we need to add a further check to avoid 7624 // taking the intrinsic path when cipher and plain are the same 7625 // see the original java code for why. 7626 RegionNode* region = new RegionNode(3); 7627 region->init_req(1, instof_false); 7628 7629 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 7630 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 7631 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN); 7632 region->init_req(2, src_dest_conjoint); 7633 7634 record_for_igvn(region); 7635 return _gvn.transform(region); 7636 } 7637 7638 //----------------------------inline_electronicCodeBook_AESCrypt_predicate---------------------------- 7639 // Return node representing slow path of predicate check. 7640 // the pseudo code we want to emulate with this predicate is: 7641 // for encryption: 7642 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7643 // for decryption: 7644 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7645 // note cipher==plain is more conservative than the original java code but that's OK 7646 // 7647 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) { 7648 // The receiver was checked for null already. 7649 Node* objECB = argument(0); 7650 7651 // Load embeddedCipher field of ElectronicCodeBook object. 7652 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7653 7654 // get AESCrypt klass for instanceOf check 7655 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7656 // will have same classloader as ElectronicCodeBook object 7657 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr(); 7658 assert(tinst != nullptr, "ECBobj is null"); 7659 assert(tinst->is_loaded(), "ECBobj is not loaded"); 7660 7661 // we want to do an instanceof comparison against the AESCrypt class 7662 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7663 if (!klass_AESCrypt->is_loaded()) { 7664 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7665 Node* ctrl = control(); 7666 set_control(top()); // no regular fast path 7667 return ctrl; 7668 } 7669 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7670 7671 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7672 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7673 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7674 7675 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7676 7677 // for encryption, we are done 7678 if (!decrypting) 7679 return instof_false; // even if it is null 7680 7681 // for decryption, we need to add a further check to avoid 7682 // taking the intrinsic path when cipher and plain are the same 7683 // see the original java code for why. 7684 RegionNode* region = new RegionNode(3); 7685 region->init_req(1, instof_false); 7686 Node* src = argument(1); 7687 Node* dest = argument(4); 7688 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 7689 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 7690 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN); 7691 region->init_req(2, src_dest_conjoint); 7692 7693 record_for_igvn(region); 7694 return _gvn.transform(region); 7695 } 7696 7697 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 7698 // Return node representing slow path of predicate check. 7699 // the pseudo code we want to emulate with this predicate is: 7700 // for encryption: 7701 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7702 // for decryption: 7703 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7704 // note cipher==plain is more conservative than the original java code but that's OK 7705 // 7706 7707 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 7708 // The receiver was checked for null already. 7709 Node* objCTR = argument(0); 7710 7711 // Load embeddedCipher field of CipherBlockChaining object. 7712 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7713 7714 // get AESCrypt klass for instanceOf check 7715 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7716 // will have same classloader as CipherBlockChaining object 7717 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 7718 assert(tinst != nullptr, "CTRobj is null"); 7719 assert(tinst->is_loaded(), "CTRobj is not loaded"); 7720 7721 // we want to do an instanceof comparison against the AESCrypt class 7722 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7723 if (!klass_AESCrypt->is_loaded()) { 7724 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7725 Node* ctrl = control(); 7726 set_control(top()); // no regular fast path 7727 return ctrl; 7728 } 7729 7730 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7731 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7732 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7733 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7734 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7735 7736 return instof_false; // even if it is null 7737 } 7738 7739 //------------------------------inline_ghash_processBlocks 7740 bool LibraryCallKit::inline_ghash_processBlocks() { 7741 address stubAddr; 7742 const char *stubName; 7743 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 7744 7745 stubAddr = StubRoutines::ghash_processBlocks(); 7746 stubName = "ghash_processBlocks"; 7747 7748 Node* data = argument(0); 7749 Node* offset = argument(1); 7750 Node* len = argument(2); 7751 Node* state = argument(3); 7752 Node* subkeyH = argument(4); 7753 7754 state = must_be_not_null(state, true); 7755 subkeyH = must_be_not_null(subkeyH, true); 7756 data = must_be_not_null(data, true); 7757 7758 Node* state_start = array_element_address(state, intcon(0), T_LONG); 7759 assert(state_start, "state is null"); 7760 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 7761 assert(subkeyH_start, "subkeyH is null"); 7762 Node* data_start = array_element_address(data, offset, T_BYTE); 7763 assert(data_start, "data is null"); 7764 7765 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 7766 OptoRuntime::ghash_processBlocks_Type(), 7767 stubAddr, stubName, TypePtr::BOTTOM, 7768 state_start, subkeyH_start, data_start, len); 7769 return true; 7770 } 7771 7772 //------------------------------inline_chacha20Block 7773 bool LibraryCallKit::inline_chacha20Block() { 7774 address stubAddr; 7775 const char *stubName; 7776 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support"); 7777 7778 stubAddr = StubRoutines::chacha20Block(); 7779 stubName = "chacha20Block"; 7780 7781 Node* state = argument(0); 7782 Node* result = argument(1); 7783 7784 state = must_be_not_null(state, true); 7785 result = must_be_not_null(result, true); 7786 7787 Node* state_start = array_element_address(state, intcon(0), T_INT); 7788 assert(state_start, "state is null"); 7789 Node* result_start = array_element_address(result, intcon(0), T_BYTE); 7790 assert(result_start, "result is null"); 7791 7792 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP, 7793 OptoRuntime::chacha20Block_Type(), 7794 stubAddr, stubName, TypePtr::BOTTOM, 7795 state_start, result_start); 7796 // return key stream length (int) 7797 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms)); 7798 set_result(retvalue); 7799 return true; 7800 } 7801 7802 bool LibraryCallKit::inline_base64_encodeBlock() { 7803 address stubAddr; 7804 const char *stubName; 7805 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 7806 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters"); 7807 stubAddr = StubRoutines::base64_encodeBlock(); 7808 stubName = "encodeBlock"; 7809 7810 if (!stubAddr) return false; 7811 Node* base64obj = argument(0); 7812 Node* src = argument(1); 7813 Node* offset = argument(2); 7814 Node* len = argument(3); 7815 Node* dest = argument(4); 7816 Node* dp = argument(5); 7817 Node* isURL = argument(6); 7818 7819 src = must_be_not_null(src, true); 7820 dest = must_be_not_null(dest, true); 7821 7822 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 7823 assert(src_start, "source array is null"); 7824 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 7825 assert(dest_start, "destination array is null"); 7826 7827 Node* base64 = make_runtime_call(RC_LEAF, 7828 OptoRuntime::base64_encodeBlock_Type(), 7829 stubAddr, stubName, TypePtr::BOTTOM, 7830 src_start, offset, len, dest_start, dp, isURL); 7831 return true; 7832 } 7833 7834 bool LibraryCallKit::inline_base64_decodeBlock() { 7835 address stubAddr; 7836 const char *stubName; 7837 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 7838 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters"); 7839 stubAddr = StubRoutines::base64_decodeBlock(); 7840 stubName = "decodeBlock"; 7841 7842 if (!stubAddr) return false; 7843 Node* base64obj = argument(0); 7844 Node* src = argument(1); 7845 Node* src_offset = argument(2); 7846 Node* len = argument(3); 7847 Node* dest = argument(4); 7848 Node* dest_offset = argument(5); 7849 Node* isURL = argument(6); 7850 Node* isMIME = argument(7); 7851 7852 src = must_be_not_null(src, true); 7853 dest = must_be_not_null(dest, true); 7854 7855 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 7856 assert(src_start, "source array is null"); 7857 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 7858 assert(dest_start, "destination array is null"); 7859 7860 Node* call = make_runtime_call(RC_LEAF, 7861 OptoRuntime::base64_decodeBlock_Type(), 7862 stubAddr, stubName, TypePtr::BOTTOM, 7863 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME); 7864 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 7865 set_result(result); 7866 return true; 7867 } 7868 7869 bool LibraryCallKit::inline_poly1305_processBlocks() { 7870 address stubAddr; 7871 const char *stubName; 7872 assert(UsePoly1305Intrinsics, "need Poly intrinsics support"); 7873 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size()); 7874 stubAddr = StubRoutines::poly1305_processBlocks(); 7875 stubName = "poly1305_processBlocks"; 7876 7877 if (!stubAddr) return false; 7878 null_check_receiver(); // null-check receiver 7879 if (stopped()) return true; 7880 7881 Node* input = argument(1); 7882 Node* input_offset = argument(2); 7883 Node* len = argument(3); 7884 Node* alimbs = argument(4); 7885 Node* rlimbs = argument(5); 7886 7887 input = must_be_not_null(input, true); 7888 alimbs = must_be_not_null(alimbs, true); 7889 rlimbs = must_be_not_null(rlimbs, true); 7890 7891 Node* input_start = array_element_address(input, input_offset, T_BYTE); 7892 assert(input_start, "input array is null"); 7893 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG); 7894 assert(acc_start, "acc array is null"); 7895 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG); 7896 assert(r_start, "r array is null"); 7897 7898 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP, 7899 OptoRuntime::poly1305_processBlocks_Type(), 7900 stubAddr, stubName, TypePtr::BOTTOM, 7901 input_start, len, acc_start, r_start); 7902 return true; 7903 } 7904 7905 //------------------------------inline_digestBase_implCompress----------------------- 7906 // 7907 // Calculate MD5 for single-block byte[] array. 7908 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs) 7909 // 7910 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 7911 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 7912 // 7913 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 7914 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 7915 // 7916 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 7917 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 7918 // 7919 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array. 7920 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs) 7921 // 7922 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) { 7923 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 7924 7925 Node* digestBase_obj = argument(0); 7926 Node* src = argument(1); // type oop 7927 Node* ofs = argument(2); // type int 7928 7929 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7930 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 7931 // failed array check 7932 return false; 7933 } 7934 // Figure out the size and type of the elements we will be copying. 7935 BasicType src_elem = src_type->elem()->array_element_basic_type(); 7936 if (src_elem != T_BYTE) { 7937 return false; 7938 } 7939 // 'src_start' points to src array + offset 7940 src = must_be_not_null(src, true); 7941 Node* src_start = array_element_address(src, ofs, src_elem); 7942 Node* state = nullptr; 7943 Node* block_size = nullptr; 7944 address stubAddr; 7945 const char *stubName; 7946 7947 switch(id) { 7948 case vmIntrinsics::_md5_implCompress: 7949 assert(UseMD5Intrinsics, "need MD5 instruction support"); 7950 state = get_state_from_digest_object(digestBase_obj, T_INT); 7951 stubAddr = StubRoutines::md5_implCompress(); 7952 stubName = "md5_implCompress"; 7953 break; 7954 case vmIntrinsics::_sha_implCompress: 7955 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 7956 state = get_state_from_digest_object(digestBase_obj, T_INT); 7957 stubAddr = StubRoutines::sha1_implCompress(); 7958 stubName = "sha1_implCompress"; 7959 break; 7960 case vmIntrinsics::_sha2_implCompress: 7961 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 7962 state = get_state_from_digest_object(digestBase_obj, T_INT); 7963 stubAddr = StubRoutines::sha256_implCompress(); 7964 stubName = "sha256_implCompress"; 7965 break; 7966 case vmIntrinsics::_sha5_implCompress: 7967 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 7968 state = get_state_from_digest_object(digestBase_obj, T_LONG); 7969 stubAddr = StubRoutines::sha512_implCompress(); 7970 stubName = "sha512_implCompress"; 7971 break; 7972 case vmIntrinsics::_sha3_implCompress: 7973 assert(UseSHA3Intrinsics, "need SHA3 instruction support"); 7974 state = get_state_from_digest_object(digestBase_obj, T_BYTE); 7975 stubAddr = StubRoutines::sha3_implCompress(); 7976 stubName = "sha3_implCompress"; 7977 block_size = get_block_size_from_digest_object(digestBase_obj); 7978 if (block_size == nullptr) return false; 7979 break; 7980 default: 7981 fatal_unexpected_iid(id); 7982 return false; 7983 } 7984 if (state == nullptr) return false; 7985 7986 assert(stubAddr != nullptr, "Stub %s is not generated", stubName); 7987 if (stubAddr == nullptr) return false; 7988 7989 // Call the stub. 7990 Node* call; 7991 if (block_size == nullptr) { 7992 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false), 7993 stubAddr, stubName, TypePtr::BOTTOM, 7994 src_start, state); 7995 } else { 7996 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true), 7997 stubAddr, stubName, TypePtr::BOTTOM, 7998 src_start, state, block_size); 7999 } 8000 8001 return true; 8002 } 8003 8004 //------------------------------inline_digestBase_implCompressMB----------------------- 8005 // 8006 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array. 8007 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 8008 // 8009 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 8010 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics, 8011 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support"); 8012 assert((uint)predicate < 5, "sanity"); 8013 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 8014 8015 Node* digestBase_obj = argument(0); // The receiver was checked for null already. 8016 Node* src = argument(1); // byte[] array 8017 Node* ofs = argument(2); // type int 8018 Node* limit = argument(3); // type int 8019 8020 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 8021 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 8022 // failed array check 8023 return false; 8024 } 8025 // Figure out the size and type of the elements we will be copying. 8026 BasicType src_elem = src_type->elem()->array_element_basic_type(); 8027 if (src_elem != T_BYTE) { 8028 return false; 8029 } 8030 // 'src_start' points to src array + offset 8031 src = must_be_not_null(src, false); 8032 Node* src_start = array_element_address(src, ofs, src_elem); 8033 8034 const char* klass_digestBase_name = nullptr; 8035 const char* stub_name = nullptr; 8036 address stub_addr = nullptr; 8037 BasicType elem_type = T_INT; 8038 8039 switch (predicate) { 8040 case 0: 8041 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) { 8042 klass_digestBase_name = "sun/security/provider/MD5"; 8043 stub_name = "md5_implCompressMB"; 8044 stub_addr = StubRoutines::md5_implCompressMB(); 8045 } 8046 break; 8047 case 1: 8048 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) { 8049 klass_digestBase_name = "sun/security/provider/SHA"; 8050 stub_name = "sha1_implCompressMB"; 8051 stub_addr = StubRoutines::sha1_implCompressMB(); 8052 } 8053 break; 8054 case 2: 8055 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) { 8056 klass_digestBase_name = "sun/security/provider/SHA2"; 8057 stub_name = "sha256_implCompressMB"; 8058 stub_addr = StubRoutines::sha256_implCompressMB(); 8059 } 8060 break; 8061 case 3: 8062 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) { 8063 klass_digestBase_name = "sun/security/provider/SHA5"; 8064 stub_name = "sha512_implCompressMB"; 8065 stub_addr = StubRoutines::sha512_implCompressMB(); 8066 elem_type = T_LONG; 8067 } 8068 break; 8069 case 4: 8070 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) { 8071 klass_digestBase_name = "sun/security/provider/SHA3"; 8072 stub_name = "sha3_implCompressMB"; 8073 stub_addr = StubRoutines::sha3_implCompressMB(); 8074 elem_type = T_BYTE; 8075 } 8076 break; 8077 default: 8078 fatal("unknown DigestBase intrinsic predicate: %d", predicate); 8079 } 8080 if (klass_digestBase_name != nullptr) { 8081 assert(stub_addr != nullptr, "Stub is generated"); 8082 if (stub_addr == nullptr) return false; 8083 8084 // get DigestBase klass to lookup for SHA klass 8085 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 8086 assert(tinst != nullptr, "digestBase_obj is not instance???"); 8087 assert(tinst->is_loaded(), "DigestBase is not loaded"); 8088 8089 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name)); 8090 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded"); 8091 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass(); 8092 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit); 8093 } 8094 return false; 8095 } 8096 8097 //------------------------------inline_digestBase_implCompressMB----------------------- 8098 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase, 8099 BasicType elem_type, address stubAddr, const char *stubName, 8100 Node* src_start, Node* ofs, Node* limit) { 8101 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase); 8102 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 8103 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 8104 digest_obj = _gvn.transform(digest_obj); 8105 8106 Node* state = get_state_from_digest_object(digest_obj, elem_type); 8107 if (state == nullptr) return false; 8108 8109 Node* block_size = nullptr; 8110 if (strcmp("sha3_implCompressMB", stubName) == 0) { 8111 block_size = get_block_size_from_digest_object(digest_obj); 8112 if (block_size == nullptr) return false; 8113 } 8114 8115 // Call the stub. 8116 Node* call; 8117 if (block_size == nullptr) { 8118 call = make_runtime_call(RC_LEAF|RC_NO_FP, 8119 OptoRuntime::digestBase_implCompressMB_Type(false), 8120 stubAddr, stubName, TypePtr::BOTTOM, 8121 src_start, state, ofs, limit); 8122 } else { 8123 call = make_runtime_call(RC_LEAF|RC_NO_FP, 8124 OptoRuntime::digestBase_implCompressMB_Type(true), 8125 stubAddr, stubName, TypePtr::BOTTOM, 8126 src_start, state, block_size, ofs, limit); 8127 } 8128 8129 // return ofs (int) 8130 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 8131 set_result(result); 8132 8133 return true; 8134 } 8135 8136 //------------------------------inline_galoisCounterMode_AESCrypt----------------------- 8137 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() { 8138 assert(UseAES, "need AES instruction support"); 8139 address stubAddr = nullptr; 8140 const char *stubName = nullptr; 8141 stubAddr = StubRoutines::galoisCounterMode_AESCrypt(); 8142 stubName = "galoisCounterMode_AESCrypt"; 8143 8144 if (stubAddr == nullptr) return false; 8145 8146 Node* in = argument(0); 8147 Node* inOfs = argument(1); 8148 Node* len = argument(2); 8149 Node* ct = argument(3); 8150 Node* ctOfs = argument(4); 8151 Node* out = argument(5); 8152 Node* outOfs = argument(6); 8153 Node* gctr_object = argument(7); 8154 Node* ghash_object = argument(8); 8155 8156 // (1) in, ct and out are arrays. 8157 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr(); 8158 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr(); 8159 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr(); 8160 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM && 8161 ct_type != nullptr && ct_type->elem() != Type::BOTTOM && 8162 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange"); 8163 8164 // checks are the responsibility of the caller 8165 Node* in_start = in; 8166 Node* ct_start = ct; 8167 Node* out_start = out; 8168 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) { 8169 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, ""); 8170 in_start = array_element_address(in, inOfs, T_BYTE); 8171 ct_start = array_element_address(ct, ctOfs, T_BYTE); 8172 out_start = array_element_address(out, outOfs, T_BYTE); 8173 } 8174 8175 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 8176 // (because of the predicated logic executed earlier). 8177 // so we cast it here safely. 8178 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 8179 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 8180 Node* counter = load_field_from_object(gctr_object, "counter", "[B"); 8181 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J"); 8182 Node* state = load_field_from_object(ghash_object, "state", "[J"); 8183 8184 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) { 8185 return false; 8186 } 8187 // cast it to what we know it will be at runtime 8188 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr(); 8189 assert(tinst != nullptr, "GCTR obj is null"); 8190 assert(tinst->is_loaded(), "GCTR obj is not loaded"); 8191 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 8192 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 8193 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 8194 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 8195 const TypeOopPtr* xtype = aklass->as_instance_type(); 8196 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 8197 aescrypt_object = _gvn.transform(aescrypt_object); 8198 // we need to get the start of the aescrypt_object's expanded key array 8199 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 8200 if (k_start == nullptr) return false; 8201 // similarly, get the start address of the r vector 8202 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE); 8203 Node* state_start = array_element_address(state, intcon(0), T_LONG); 8204 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG); 8205 8206 8207 // Call the stub, passing params 8208 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 8209 OptoRuntime::galoisCounterMode_aescrypt_Type(), 8210 stubAddr, stubName, TypePtr::BOTTOM, 8211 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start); 8212 8213 // return cipher length (int) 8214 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms)); 8215 set_result(retvalue); 8216 8217 return true; 8218 } 8219 8220 //----------------------------inline_galoisCounterMode_AESCrypt_predicate---------------------------- 8221 // Return node representing slow path of predicate check. 8222 // the pseudo code we want to emulate with this predicate is: 8223 // for encryption: 8224 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 8225 // for decryption: 8226 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 8227 // note cipher==plain is more conservative than the original java code but that's OK 8228 // 8229 8230 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() { 8231 // The receiver was checked for null already. 8232 Node* objGCTR = argument(7); 8233 // Load embeddedCipher field of GCTR object. 8234 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 8235 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null"); 8236 8237 // get AESCrypt klass for instanceOf check 8238 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 8239 // will have same classloader as CipherBlockChaining object 8240 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr(); 8241 assert(tinst != nullptr, "GCTR obj is null"); 8242 assert(tinst->is_loaded(), "GCTR obj is not loaded"); 8243 8244 // we want to do an instanceof comparison against the AESCrypt class 8245 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 8246 if (!klass_AESCrypt->is_loaded()) { 8247 // if AESCrypt is not even loaded, we never take the intrinsic fast path 8248 Node* ctrl = control(); 8249 set_control(top()); // no regular fast path 8250 return ctrl; 8251 } 8252 8253 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 8254 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 8255 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 8256 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 8257 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 8258 8259 return instof_false; // even if it is null 8260 } 8261 8262 //------------------------------get_state_from_digest_object----------------------- 8263 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) { 8264 const char* state_type; 8265 switch (elem_type) { 8266 case T_BYTE: state_type = "[B"; break; 8267 case T_INT: state_type = "[I"; break; 8268 case T_LONG: state_type = "[J"; break; 8269 default: ShouldNotReachHere(); 8270 } 8271 Node* digest_state = load_field_from_object(digest_object, "state", state_type); 8272 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3"); 8273 if (digest_state == nullptr) return (Node *) nullptr; 8274 8275 // now have the array, need to get the start address of the state array 8276 Node* state = array_element_address(digest_state, intcon(0), elem_type); 8277 return state; 8278 } 8279 8280 //------------------------------get_block_size_from_sha3_object---------------------------------- 8281 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) { 8282 Node* block_size = load_field_from_object(digest_object, "blockSize", "I"); 8283 assert (block_size != nullptr, "sanity"); 8284 return block_size; 8285 } 8286 8287 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 8288 // Return node representing slow path of predicate check. 8289 // the pseudo code we want to emulate with this predicate is: 8290 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath 8291 // 8292 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 8293 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics, 8294 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support"); 8295 assert((uint)predicate < 5, "sanity"); 8296 8297 // The receiver was checked for null already. 8298 Node* digestBaseObj = argument(0); 8299 8300 // get DigestBase klass for instanceOf check 8301 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 8302 assert(tinst != nullptr, "digestBaseObj is null"); 8303 assert(tinst->is_loaded(), "DigestBase is not loaded"); 8304 8305 const char* klass_name = nullptr; 8306 switch (predicate) { 8307 case 0: 8308 if (UseMD5Intrinsics) { 8309 // we want to do an instanceof comparison against the MD5 class 8310 klass_name = "sun/security/provider/MD5"; 8311 } 8312 break; 8313 case 1: 8314 if (UseSHA1Intrinsics) { 8315 // we want to do an instanceof comparison against the SHA class 8316 klass_name = "sun/security/provider/SHA"; 8317 } 8318 break; 8319 case 2: 8320 if (UseSHA256Intrinsics) { 8321 // we want to do an instanceof comparison against the SHA2 class 8322 klass_name = "sun/security/provider/SHA2"; 8323 } 8324 break; 8325 case 3: 8326 if (UseSHA512Intrinsics) { 8327 // we want to do an instanceof comparison against the SHA5 class 8328 klass_name = "sun/security/provider/SHA5"; 8329 } 8330 break; 8331 case 4: 8332 if (UseSHA3Intrinsics) { 8333 // we want to do an instanceof comparison against the SHA3 class 8334 klass_name = "sun/security/provider/SHA3"; 8335 } 8336 break; 8337 default: 8338 fatal("unknown SHA intrinsic predicate: %d", predicate); 8339 } 8340 8341 ciKlass* klass = nullptr; 8342 if (klass_name != nullptr) { 8343 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name)); 8344 } 8345 if ((klass == nullptr) || !klass->is_loaded()) { 8346 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 8347 Node* ctrl = control(); 8348 set_control(top()); // no intrinsic path 8349 return ctrl; 8350 } 8351 ciInstanceKlass* instklass = klass->as_instance_klass(); 8352 8353 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass))); 8354 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 8355 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 8356 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 8357 8358 return instof_false; // even if it is null 8359 } 8360 8361 //-------------inline_fma----------------------------------- 8362 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) { 8363 Node *a = nullptr; 8364 Node *b = nullptr; 8365 Node *c = nullptr; 8366 Node* result = nullptr; 8367 switch (id) { 8368 case vmIntrinsics::_fmaD: 8369 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each."); 8370 // no receiver since it is static method 8371 a = round_double_node(argument(0)); 8372 b = round_double_node(argument(2)); 8373 c = round_double_node(argument(4)); 8374 result = _gvn.transform(new FmaDNode(control(), a, b, c)); 8375 break; 8376 case vmIntrinsics::_fmaF: 8377 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each."); 8378 a = argument(0); 8379 b = argument(1); 8380 c = argument(2); 8381 result = _gvn.transform(new FmaFNode(control(), a, b, c)); 8382 break; 8383 default: 8384 fatal_unexpected_iid(id); break; 8385 } 8386 set_result(result); 8387 return true; 8388 } 8389 8390 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) { 8391 // argument(0) is receiver 8392 Node* codePoint = argument(1); 8393 Node* n = nullptr; 8394 8395 switch (id) { 8396 case vmIntrinsics::_isDigit : 8397 n = new DigitNode(control(), codePoint); 8398 break; 8399 case vmIntrinsics::_isLowerCase : 8400 n = new LowerCaseNode(control(), codePoint); 8401 break; 8402 case vmIntrinsics::_isUpperCase : 8403 n = new UpperCaseNode(control(), codePoint); 8404 break; 8405 case vmIntrinsics::_isWhitespace : 8406 n = new WhitespaceNode(control(), codePoint); 8407 break; 8408 default: 8409 fatal_unexpected_iid(id); 8410 } 8411 8412 set_result(_gvn.transform(n)); 8413 return true; 8414 } 8415 8416 //------------------------------inline_fp_min_max------------------------------ 8417 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) { 8418 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416. 8419 8420 // The intrinsic should be used only when the API branches aren't predictable, 8421 // the last one performing the most important comparison. The following heuristic 8422 // uses the branch statistics to eventually bail out if necessary. 8423 8424 ciMethodData *md = callee()->method_data(); 8425 8426 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) { 8427 ciCallProfile cp = caller()->call_profile_at_bci(bci()); 8428 8429 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) { 8430 // Bail out if the call-site didn't contribute enough to the statistics. 8431 return false; 8432 } 8433 8434 uint taken = 0, not_taken = 0; 8435 8436 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) { 8437 if (p->is_BranchData()) { 8438 taken = ((ciBranchData*)p)->taken(); 8439 not_taken = ((ciBranchData*)p)->not_taken(); 8440 } 8441 } 8442 8443 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count()); 8444 balance = balance < 0 ? -balance : balance; 8445 if ( balance > 0.2 ) { 8446 // Bail out if the most important branch is predictable enough. 8447 return false; 8448 } 8449 } 8450 */ 8451 8452 Node *a = nullptr; 8453 Node *b = nullptr; 8454 Node *n = nullptr; 8455 switch (id) { 8456 case vmIntrinsics::_maxF: 8457 case vmIntrinsics::_minF: 8458 case vmIntrinsics::_maxF_strict: 8459 case vmIntrinsics::_minF_strict: 8460 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each."); 8461 a = argument(0); 8462 b = argument(1); 8463 break; 8464 case vmIntrinsics::_maxD: 8465 case vmIntrinsics::_minD: 8466 case vmIntrinsics::_maxD_strict: 8467 case vmIntrinsics::_minD_strict: 8468 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each."); 8469 a = round_double_node(argument(0)); 8470 b = round_double_node(argument(2)); 8471 break; 8472 default: 8473 fatal_unexpected_iid(id); 8474 break; 8475 } 8476 switch (id) { 8477 case vmIntrinsics::_maxF: 8478 case vmIntrinsics::_maxF_strict: 8479 n = new MaxFNode(a, b); 8480 break; 8481 case vmIntrinsics::_minF: 8482 case vmIntrinsics::_minF_strict: 8483 n = new MinFNode(a, b); 8484 break; 8485 case vmIntrinsics::_maxD: 8486 case vmIntrinsics::_maxD_strict: 8487 n = new MaxDNode(a, b); 8488 break; 8489 case vmIntrinsics::_minD: 8490 case vmIntrinsics::_minD_strict: 8491 n = new MinDNode(a, b); 8492 break; 8493 default: 8494 fatal_unexpected_iid(id); 8495 break; 8496 } 8497 set_result(_gvn.transform(n)); 8498 return true; 8499 } 8500 8501 bool LibraryCallKit::inline_profileBoolean() { 8502 Node* counts = argument(1); 8503 const TypeAryPtr* ary = nullptr; 8504 ciArray* aobj = nullptr; 8505 if (counts->is_Con() 8506 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr 8507 && (aobj = ary->const_oop()->as_array()) != nullptr 8508 && (aobj->length() == 2)) { 8509 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 8510 jint false_cnt = aobj->element_value(0).as_int(); 8511 jint true_cnt = aobj->element_value(1).as_int(); 8512 8513 if (C->log() != nullptr) { 8514 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 8515 false_cnt, true_cnt); 8516 } 8517 8518 if (false_cnt + true_cnt == 0) { 8519 // According to profile, never executed. 8520 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 8521 Deoptimization::Action_reinterpret); 8522 return true; 8523 } 8524 8525 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 8526 // is a number of each value occurrences. 8527 Node* result = argument(0); 8528 if (false_cnt == 0 || true_cnt == 0) { 8529 // According to profile, one value has been never seen. 8530 int expected_val = (false_cnt == 0) ? 1 : 0; 8531 8532 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 8533 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 8534 8535 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 8536 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 8537 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 8538 8539 { // Slow path: uncommon trap for never seen value and then reexecute 8540 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 8541 // the value has been seen at least once. 8542 PreserveJVMState pjvms(this); 8543 PreserveReexecuteState preexecs(this); 8544 jvms()->set_should_reexecute(true); 8545 8546 set_control(slow_path); 8547 set_i_o(i_o()); 8548 8549 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 8550 Deoptimization::Action_reinterpret); 8551 } 8552 // The guard for never seen value enables sharpening of the result and 8553 // returning a constant. It allows to eliminate branches on the same value 8554 // later on. 8555 set_control(fast_path); 8556 result = intcon(expected_val); 8557 } 8558 // Stop profiling. 8559 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 8560 // By replacing method body with profile data (represented as ProfileBooleanNode 8561 // on IR level) we effectively disable profiling. 8562 // It enables full speed execution once optimized code is generated. 8563 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 8564 C->record_for_igvn(profile); 8565 set_result(profile); 8566 return true; 8567 } else { 8568 // Continue profiling. 8569 // Profile data isn't available at the moment. So, execute method's bytecode version. 8570 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 8571 // is compiled and counters aren't available since corresponding MethodHandle 8572 // isn't a compile-time constant. 8573 return false; 8574 } 8575 } 8576 8577 bool LibraryCallKit::inline_isCompileConstant() { 8578 Node* n = argument(0); 8579 set_result(n->is_Con() ? intcon(1) : intcon(0)); 8580 return true; 8581 } 8582 8583 //------------------------------- inline_getObjectSize -------------------------------------- 8584 // 8585 // Calculate the runtime size of the object/array. 8586 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize); 8587 // 8588 bool LibraryCallKit::inline_getObjectSize() { 8589 Node* obj = argument(3); 8590 Node* klass_node = load_object_klass(obj); 8591 8592 jint layout_con = Klass::_lh_neutral_value; 8593 Node* layout_val = get_layout_helper(klass_node, layout_con); 8594 int layout_is_con = (layout_val == nullptr); 8595 8596 if (layout_is_con) { 8597 // Layout helper is constant, can figure out things at compile time. 8598 8599 if (Klass::layout_helper_is_instance(layout_con)) { 8600 // Instance case: layout_con contains the size itself. 8601 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con)); 8602 set_result(size); 8603 } else { 8604 // Array case: size is round(header + element_size*arraylength). 8605 // Since arraylength is different for every array instance, we have to 8606 // compute the whole thing at runtime. 8607 8608 Node* arr_length = load_array_length(obj); 8609 8610 int round_mask = MinObjAlignmentInBytes - 1; 8611 int hsize = Klass::layout_helper_header_size(layout_con); 8612 int eshift = Klass::layout_helper_log2_element_size(layout_con); 8613 8614 if ((round_mask & ~right_n_bits(eshift)) == 0) { 8615 round_mask = 0; // strength-reduce it if it goes away completely 8616 } 8617 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded"); 8618 Node* header_size = intcon(hsize + round_mask); 8619 8620 Node* lengthx = ConvI2X(arr_length); 8621 Node* headerx = ConvI2X(header_size); 8622 8623 Node* abody = lengthx; 8624 if (eshift != 0) { 8625 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift))); 8626 } 8627 Node* size = _gvn.transform( new AddXNode(headerx, abody) ); 8628 if (round_mask != 0) { 8629 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) ); 8630 } 8631 size = ConvX2L(size); 8632 set_result(size); 8633 } 8634 } else { 8635 // Layout helper is not constant, need to test for array-ness at runtime. 8636 8637 enum { _instance_path = 1, _array_path, PATH_LIMIT }; 8638 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 8639 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG); 8640 record_for_igvn(result_reg); 8641 8642 Node* array_ctl = generate_array_guard(klass_node, nullptr); 8643 if (array_ctl != nullptr) { 8644 // Array case: size is round(header + element_size*arraylength). 8645 // Since arraylength is different for every array instance, we have to 8646 // compute the whole thing at runtime. 8647 8648 PreserveJVMState pjvms(this); 8649 set_control(array_ctl); 8650 Node* arr_length = load_array_length(obj); 8651 8652 int round_mask = MinObjAlignmentInBytes - 1; 8653 Node* mask = intcon(round_mask); 8654 8655 Node* hss = intcon(Klass::_lh_header_size_shift); 8656 Node* hsm = intcon(Klass::_lh_header_size_mask); 8657 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss)); 8658 header_size = _gvn.transform(new AndINode(header_size, hsm)); 8659 header_size = _gvn.transform(new AddINode(header_size, mask)); 8660 8661 // There is no need to mask or shift this value. 8662 // The semantics of LShiftINode include an implicit mask to 0x1F. 8663 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place"); 8664 Node* elem_shift = layout_val; 8665 8666 Node* lengthx = ConvI2X(arr_length); 8667 Node* headerx = ConvI2X(header_size); 8668 8669 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift)); 8670 Node* size = _gvn.transform(new AddXNode(headerx, abody)); 8671 if (round_mask != 0) { 8672 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask))); 8673 } 8674 size = ConvX2L(size); 8675 8676 result_reg->init_req(_array_path, control()); 8677 result_val->init_req(_array_path, size); 8678 } 8679 8680 if (!stopped()) { 8681 // Instance case: the layout helper gives us instance size almost directly, 8682 // but we need to mask out the _lh_instance_slow_path_bit. 8683 Node* size = ConvI2X(layout_val); 8684 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit"); 8685 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong)); 8686 size = _gvn.transform(new AndXNode(size, mask)); 8687 size = ConvX2L(size); 8688 8689 result_reg->init_req(_instance_path, control()); 8690 result_val->init_req(_instance_path, size); 8691 } 8692 8693 set_result(result_reg, result_val); 8694 } 8695 8696 return true; 8697 } 8698 8699 //------------------------------- inline_blackhole -------------------------------------- 8700 // 8701 // Make sure all arguments to this node are alive. 8702 // This matches methods that were requested to be blackholed through compile commands. 8703 // 8704 bool LibraryCallKit::inline_blackhole() { 8705 assert(callee()->is_static(), "Should have been checked before: only static methods here"); 8706 assert(callee()->is_empty(), "Should have been checked before: only empty methods here"); 8707 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here"); 8708 8709 // Blackhole node pinches only the control, not memory. This allows 8710 // the blackhole to be pinned in the loop that computes blackholed 8711 // values, but have no other side effects, like breaking the optimizations 8712 // across the blackhole. 8713 8714 Node* bh = _gvn.transform(new BlackholeNode(control())); 8715 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control))); 8716 8717 // Bind call arguments as blackhole arguments to keep them alive 8718 uint nargs = callee()->arg_size(); 8719 for (uint i = 0; i < nargs; i++) { 8720 bh->add_req(argument(i)); 8721 } 8722 8723 return true; 8724 }