1 /* 2 * Copyright (c) 1999, 2025, 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 "asm/macroAssembler.hpp" 26 #include "ci/ciSymbols.hpp" 27 #include "ci/ciUtilities.inline.hpp" 28 #include "classfile/vmIntrinsics.hpp" 29 #include "compiler/compileBroker.hpp" 30 #include "compiler/compileLog.hpp" 31 #include "gc/shared/barrierSet.hpp" 32 #include "jfr/support/jfrIntrinsics.hpp" 33 #include "memory/resourceArea.hpp" 34 #include "oops/klass.inline.hpp" 35 #include "oops/objArrayKlass.hpp" 36 #include "opto/addnode.hpp" 37 #include "opto/arraycopynode.hpp" 38 #include "opto/c2compiler.hpp" 39 #include "opto/castnode.hpp" 40 #include "opto/cfgnode.hpp" 41 #include "opto/convertnode.hpp" 42 #include "opto/countbitsnode.hpp" 43 #include "opto/idealKit.hpp" 44 #include "opto/library_call.hpp" 45 #include "opto/mathexactnode.hpp" 46 #include "opto/mulnode.hpp" 47 #include "opto/narrowptrnode.hpp" 48 #include "opto/opaquenode.hpp" 49 #include "opto/parse.hpp" 50 #include "opto/rootnode.hpp" 51 #include "opto/runtime.hpp" 52 #include "opto/subnode.hpp" 53 #include "opto/vectornode.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 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg); 124 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 125 if (C->log()) { 126 C->log()->elem("intrinsic id='%s'%s nodes='%d'", 127 vmIntrinsics::name_at(intrinsic_id()), 128 (is_virtual() ? " virtual='1'" : ""), 129 C->unique() - nodes); 130 } 131 // Push the result from the inlined method onto the stack. 132 kit.push_result(); 133 return kit.transfer_exceptions_into_jvms(); 134 } 135 136 // The intrinsic bailed out 137 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out"); 138 assert(jvms->map() == kit.map(), "Out of sync JVM state"); 139 if (jvms->has_method()) { 140 // Not a root compile. 141 const char* msg; 142 if (callee->intrinsic_candidate()) { 143 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)"; 144 } else { 145 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated" 146 : "failed to inline (intrinsic), method not annotated"; 147 } 148 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg); 149 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg); 150 } else { 151 // Root compile 152 ResourceMark rm; 153 stringStream msg_stream; 154 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 155 vmIntrinsics::name_at(intrinsic_id()), 156 is_virtual() ? " (virtual)" : "", bci); 157 const char *msg = msg_stream.freeze(); 158 log_debug(jit, inlining)("%s", msg); 159 if (C->print_intrinsics() || C->print_inlining()) { 160 tty->print("%s", msg); 161 } 162 } 163 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 164 165 return nullptr; 166 } 167 168 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) { 169 LibraryCallKit kit(jvms, this); 170 Compile* C = kit.C; 171 int nodes = C->unique(); 172 _last_predicate = predicate; 173 #ifndef PRODUCT 174 assert(is_predicated() && predicate < predicates_count(), "sanity"); 175 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 176 char buf[1000]; 177 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 178 tty->print_cr("Predicate for intrinsic %s", str); 179 } 180 #endif 181 ciMethod* callee = kit.callee(); 182 const int bci = kit.bci(); 183 184 Node* slow_ctl = kit.try_to_predicate(predicate); 185 if (!kit.failing()) { 186 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)" 187 : "(intrinsic, predicate)"; 188 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg); 189 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg); 190 191 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 192 if (C->log()) { 193 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'", 194 vmIntrinsics::name_at(intrinsic_id()), 195 (is_virtual() ? " virtual='1'" : ""), 196 C->unique() - nodes); 197 } 198 return slow_ctl; // Could be null if the check folds. 199 } 200 201 // The intrinsic bailed out 202 if (jvms->has_method()) { 203 // Not a root compile. 204 const char* msg = "failed to generate predicate for intrinsic"; 205 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg); 206 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg); 207 } else { 208 // Root compile 209 ResourceMark rm; 210 stringStream msg_stream; 211 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 212 vmIntrinsics::name_at(intrinsic_id()), 213 is_virtual() ? " (virtual)" : "", bci); 214 const char *msg = msg_stream.freeze(); 215 log_debug(jit, inlining)("%s", msg); 216 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg); 217 } 218 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 219 return nullptr; 220 } 221 222 bool LibraryCallKit::try_to_inline(int predicate) { 223 // Handle symbolic names for otherwise undistinguished boolean switches: 224 const bool is_store = true; 225 const bool is_compress = true; 226 const bool is_static = true; 227 const bool is_volatile = true; 228 229 if (!jvms()->has_method()) { 230 // Root JVMState has a null method. 231 assert(map()->memory()->Opcode() == Op_Parm, ""); 232 // Insert the memory aliasing node 233 set_all_memory(reset_memory()); 234 } 235 assert(merged_memory(), ""); 236 237 switch (intrinsic_id()) { 238 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static); 239 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static); 240 case vmIntrinsics::_getClass: return inline_native_getClass(); 241 242 case vmIntrinsics::_ceil: 243 case vmIntrinsics::_floor: 244 case vmIntrinsics::_rint: 245 case vmIntrinsics::_dsin: 246 case vmIntrinsics::_dcos: 247 case vmIntrinsics::_dtan: 248 case vmIntrinsics::_dtanh: 249 case vmIntrinsics::_dcbrt: 250 case vmIntrinsics::_dabs: 251 case vmIntrinsics::_fabs: 252 case vmIntrinsics::_iabs: 253 case vmIntrinsics::_labs: 254 case vmIntrinsics::_datan2: 255 case vmIntrinsics::_dsqrt: 256 case vmIntrinsics::_dsqrt_strict: 257 case vmIntrinsics::_dexp: 258 case vmIntrinsics::_dlog: 259 case vmIntrinsics::_dlog10: 260 case vmIntrinsics::_dpow: 261 case vmIntrinsics::_dcopySign: 262 case vmIntrinsics::_fcopySign: 263 case vmIntrinsics::_dsignum: 264 case vmIntrinsics::_roundF: 265 case vmIntrinsics::_roundD: 266 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id()); 267 268 case vmIntrinsics::_notify: 269 case vmIntrinsics::_notifyAll: 270 return inline_notify(intrinsic_id()); 271 272 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 273 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 274 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 275 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 276 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 277 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 278 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 279 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 280 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh(); 281 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh(); 282 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 283 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 284 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 285 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 286 287 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 288 289 case vmIntrinsics::_arraySort: return inline_array_sort(); 290 case vmIntrinsics::_arrayPartition: return inline_array_partition(); 291 292 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 293 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 294 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 295 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 296 297 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 298 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 299 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 300 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 301 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 302 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 303 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U); 304 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L); 305 306 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 307 308 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode(); 309 310 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 311 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 312 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 313 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 314 315 case vmIntrinsics::_compressStringC: 316 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 317 case vmIntrinsics::_inflateStringC: 318 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 319 320 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false); 321 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false); 322 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false); 323 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false); 324 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false); 325 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false); 326 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false); 327 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false); 328 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false); 329 330 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false); 331 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false); 332 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false); 333 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false); 334 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false); 335 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false); 336 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false); 337 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false); 338 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false); 339 340 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false); 341 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false); 342 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false); 343 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false); 344 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false); 345 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false); 346 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false); 347 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false); 348 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false); 349 350 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false); 351 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false); 352 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false); 353 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false); 354 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false); 355 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false); 356 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false); 357 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false); 358 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false); 359 360 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true); 361 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true); 362 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true); 363 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true); 364 365 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true); 366 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true); 367 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true); 368 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true); 369 370 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false); 371 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false); 372 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false); 373 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false); 374 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false); 375 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false); 376 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false); 377 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false); 378 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false); 379 380 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false); 381 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false); 382 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false); 383 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false); 384 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false); 385 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false); 386 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false); 387 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false); 388 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false); 389 390 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false); 391 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false); 392 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false); 393 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false); 394 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false); 395 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false); 396 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false); 397 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false); 398 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false); 399 400 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false); 401 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false); 402 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false); 403 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false); 404 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false); 405 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false); 406 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false); 407 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false); 408 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false); 409 410 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 411 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile); 412 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile); 413 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 414 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 415 416 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 417 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 418 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 419 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile); 420 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed); 421 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire); 422 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release); 423 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile); 424 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed); 425 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire); 426 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release); 427 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile); 428 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 429 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 430 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 431 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile); 432 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 433 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 434 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 435 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile); 436 437 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 438 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 439 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 440 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile); 441 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire); 442 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release); 443 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile); 444 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire); 445 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release); 446 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 447 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 448 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 449 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 450 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 451 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 452 453 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile); 454 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile); 455 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 456 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 457 458 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile); 459 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile); 460 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 461 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 462 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 463 464 case vmIntrinsics::_loadFence: 465 case vmIntrinsics::_storeFence: 466 case vmIntrinsics::_storeStoreFence: 467 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 468 469 case vmIntrinsics::_onSpinWait: return inline_onspinwait(); 470 471 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread(); 472 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 473 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread(); 474 475 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache(); 476 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache(); 477 478 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false); 479 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true); 480 481 #if INCLUDE_JVMTI 482 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()), 483 "notifyJvmtiStart", true, false); 484 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()), 485 "notifyJvmtiEnd", false, true); 486 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()), 487 "notifyJvmtiMount", false, false); 488 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()), 489 "notifyJvmtiUnmount", false, false); 490 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync(); 491 #endif 492 493 #ifdef JFR_HAVE_INTRINSICS 494 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime"); 495 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter(); 496 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit(); 497 #endif 498 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 499 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 500 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0(); 501 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true); 502 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false); 503 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 504 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 505 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory(); 506 case vmIntrinsics::_getLength: return inline_native_getLength(); 507 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 508 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 509 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 510 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 511 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT); 512 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG); 513 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 514 515 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true); 516 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false); 517 518 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 519 520 case vmIntrinsics::_isInstance: 521 case vmIntrinsics::_isHidden: 522 case vmIntrinsics::_getSuperclass: 523 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id()); 524 525 case vmIntrinsics::_floatToRawIntBits: 526 case vmIntrinsics::_floatToIntBits: 527 case vmIntrinsics::_intBitsToFloat: 528 case vmIntrinsics::_doubleToRawLongBits: 529 case vmIntrinsics::_doubleToLongBits: 530 case vmIntrinsics::_longBitsToDouble: 531 case vmIntrinsics::_floatToFloat16: 532 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id()); 533 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1); 534 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3); 535 case vmIntrinsics::_floatIsFinite: 536 case vmIntrinsics::_floatIsInfinite: 537 case vmIntrinsics::_doubleIsFinite: 538 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id()); 539 540 case vmIntrinsics::_numberOfLeadingZeros_i: 541 case vmIntrinsics::_numberOfLeadingZeros_l: 542 case vmIntrinsics::_numberOfTrailingZeros_i: 543 case vmIntrinsics::_numberOfTrailingZeros_l: 544 case vmIntrinsics::_bitCount_i: 545 case vmIntrinsics::_bitCount_l: 546 case vmIntrinsics::_reverse_i: 547 case vmIntrinsics::_reverse_l: 548 case vmIntrinsics::_reverseBytes_i: 549 case vmIntrinsics::_reverseBytes_l: 550 case vmIntrinsics::_reverseBytes_s: 551 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id()); 552 553 case vmIntrinsics::_compress_i: 554 case vmIntrinsics::_compress_l: 555 case vmIntrinsics::_expand_i: 556 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id()); 557 558 case vmIntrinsics::_compareUnsigned_i: 559 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id()); 560 561 case vmIntrinsics::_divideUnsigned_i: 562 case vmIntrinsics::_divideUnsigned_l: 563 case vmIntrinsics::_remainderUnsigned_i: 564 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id()); 565 566 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass(); 567 568 case vmIntrinsics::_Reference_get0: return inline_reference_get0(); 569 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false); 570 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true); 571 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false); 572 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true); 573 574 case vmIntrinsics::_Class_cast: return inline_Class_cast(); 575 576 case vmIntrinsics::_aescrypt_encryptBlock: 577 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id()); 578 579 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 580 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 581 return inline_cipherBlockChaining_AESCrypt(intrinsic_id()); 582 583 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 584 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 585 return inline_electronicCodeBook_AESCrypt(intrinsic_id()); 586 587 case vmIntrinsics::_counterMode_AESCrypt: 588 return inline_counterMode_AESCrypt(intrinsic_id()); 589 590 case vmIntrinsics::_galoisCounterMode_AESCrypt: 591 return inline_galoisCounterMode_AESCrypt(); 592 593 case vmIntrinsics::_md5_implCompress: 594 case vmIntrinsics::_sha_implCompress: 595 case vmIntrinsics::_sha2_implCompress: 596 case vmIntrinsics::_sha5_implCompress: 597 case vmIntrinsics::_sha3_implCompress: 598 return inline_digestBase_implCompress(intrinsic_id()); 599 case vmIntrinsics::_double_keccak: 600 return inline_double_keccak(); 601 602 case vmIntrinsics::_digestBase_implCompressMB: 603 return inline_digestBase_implCompressMB(predicate); 604 605 case vmIntrinsics::_multiplyToLen: 606 return inline_multiplyToLen(); 607 608 case vmIntrinsics::_squareToLen: 609 return inline_squareToLen(); 610 611 case vmIntrinsics::_mulAdd: 612 return inline_mulAdd(); 613 614 case vmIntrinsics::_montgomeryMultiply: 615 return inline_montgomeryMultiply(); 616 case vmIntrinsics::_montgomerySquare: 617 return inline_montgomerySquare(); 618 619 case vmIntrinsics::_bigIntegerRightShiftWorker: 620 return inline_bigIntegerShift(true); 621 case vmIntrinsics::_bigIntegerLeftShiftWorker: 622 return inline_bigIntegerShift(false); 623 624 case vmIntrinsics::_vectorizedMismatch: 625 return inline_vectorizedMismatch(); 626 627 case vmIntrinsics::_ghash_processBlocks: 628 return inline_ghash_processBlocks(); 629 case vmIntrinsics::_chacha20Block: 630 return inline_chacha20Block(); 631 case vmIntrinsics::_kyberNtt: 632 return inline_kyberNtt(); 633 case vmIntrinsics::_kyberInverseNtt: 634 return inline_kyberInverseNtt(); 635 case vmIntrinsics::_kyberNttMult: 636 return inline_kyberNttMult(); 637 case vmIntrinsics::_kyberAddPoly_2: 638 return inline_kyberAddPoly_2(); 639 case vmIntrinsics::_kyberAddPoly_3: 640 return inline_kyberAddPoly_3(); 641 case vmIntrinsics::_kyber12To16: 642 return inline_kyber12To16(); 643 case vmIntrinsics::_kyberBarrettReduce: 644 return inline_kyberBarrettReduce(); 645 case vmIntrinsics::_dilithiumAlmostNtt: 646 return inline_dilithiumAlmostNtt(); 647 case vmIntrinsics::_dilithiumAlmostInverseNtt: 648 return inline_dilithiumAlmostInverseNtt(); 649 case vmIntrinsics::_dilithiumNttMult: 650 return inline_dilithiumNttMult(); 651 case vmIntrinsics::_dilithiumMontMulByConstant: 652 return inline_dilithiumMontMulByConstant(); 653 case vmIntrinsics::_dilithiumDecomposePoly: 654 return inline_dilithiumDecomposePoly(); 655 case vmIntrinsics::_base64_encodeBlock: 656 return inline_base64_encodeBlock(); 657 case vmIntrinsics::_base64_decodeBlock: 658 return inline_base64_decodeBlock(); 659 case vmIntrinsics::_poly1305_processBlocks: 660 return inline_poly1305_processBlocks(); 661 case vmIntrinsics::_intpoly_montgomeryMult_P256: 662 return inline_intpoly_montgomeryMult_P256(); 663 case vmIntrinsics::_intpoly_assign: 664 return inline_intpoly_assign(); 665 case vmIntrinsics::_encodeISOArray: 666 case vmIntrinsics::_encodeByteISOArray: 667 return inline_encodeISOArray(false); 668 case vmIntrinsics::_encodeAsciiArray: 669 return inline_encodeISOArray(true); 670 671 case vmIntrinsics::_updateCRC32: 672 return inline_updateCRC32(); 673 case vmIntrinsics::_updateBytesCRC32: 674 return inline_updateBytesCRC32(); 675 case vmIntrinsics::_updateByteBufferCRC32: 676 return inline_updateByteBufferCRC32(); 677 678 case vmIntrinsics::_updateBytesCRC32C: 679 return inline_updateBytesCRC32C(); 680 case vmIntrinsics::_updateDirectByteBufferCRC32C: 681 return inline_updateDirectByteBufferCRC32C(); 682 683 case vmIntrinsics::_updateBytesAdler32: 684 return inline_updateBytesAdler32(); 685 case vmIntrinsics::_updateByteBufferAdler32: 686 return inline_updateByteBufferAdler32(); 687 688 case vmIntrinsics::_profileBoolean: 689 return inline_profileBoolean(); 690 case vmIntrinsics::_isCompileConstant: 691 return inline_isCompileConstant(); 692 693 case vmIntrinsics::_countPositives: 694 return inline_countPositives(); 695 696 case vmIntrinsics::_fmaD: 697 case vmIntrinsics::_fmaF: 698 return inline_fma(intrinsic_id()); 699 700 case vmIntrinsics::_isDigit: 701 case vmIntrinsics::_isLowerCase: 702 case vmIntrinsics::_isUpperCase: 703 case vmIntrinsics::_isWhitespace: 704 return inline_character_compare(intrinsic_id()); 705 706 case vmIntrinsics::_min: 707 case vmIntrinsics::_max: 708 case vmIntrinsics::_min_strict: 709 case vmIntrinsics::_max_strict: 710 case vmIntrinsics::_minL: 711 case vmIntrinsics::_maxL: 712 case vmIntrinsics::_minF: 713 case vmIntrinsics::_maxF: 714 case vmIntrinsics::_minD: 715 case vmIntrinsics::_maxD: 716 case vmIntrinsics::_minF_strict: 717 case vmIntrinsics::_maxF_strict: 718 case vmIntrinsics::_minD_strict: 719 case vmIntrinsics::_maxD_strict: 720 return inline_min_max(intrinsic_id()); 721 722 case vmIntrinsics::_VectorUnaryOp: 723 return inline_vector_nary_operation(1); 724 case vmIntrinsics::_VectorBinaryOp: 725 return inline_vector_nary_operation(2); 726 case vmIntrinsics::_VectorUnaryLibOp: 727 return inline_vector_call(1); 728 case vmIntrinsics::_VectorBinaryLibOp: 729 return inline_vector_call(2); 730 case vmIntrinsics::_VectorTernaryOp: 731 return inline_vector_nary_operation(3); 732 case vmIntrinsics::_VectorFromBitsCoerced: 733 return inline_vector_frombits_coerced(); 734 case vmIntrinsics::_VectorMaskOp: 735 return inline_vector_mask_operation(); 736 case vmIntrinsics::_VectorLoadOp: 737 return inline_vector_mem_operation(/*is_store=*/false); 738 case vmIntrinsics::_VectorLoadMaskedOp: 739 return inline_vector_mem_masked_operation(/*is_store*/false); 740 case vmIntrinsics::_VectorStoreOp: 741 return inline_vector_mem_operation(/*is_store=*/true); 742 case vmIntrinsics::_VectorStoreMaskedOp: 743 return inline_vector_mem_masked_operation(/*is_store=*/true); 744 case vmIntrinsics::_VectorGatherOp: 745 return inline_vector_gather_scatter(/*is_scatter*/ false); 746 case vmIntrinsics::_VectorScatterOp: 747 return inline_vector_gather_scatter(/*is_scatter*/ true); 748 case vmIntrinsics::_VectorReductionCoerced: 749 return inline_vector_reduction(); 750 case vmIntrinsics::_VectorTest: 751 return inline_vector_test(); 752 case vmIntrinsics::_VectorBlend: 753 return inline_vector_blend(); 754 case vmIntrinsics::_VectorRearrange: 755 return inline_vector_rearrange(); 756 case vmIntrinsics::_VectorSelectFrom: 757 return inline_vector_select_from(); 758 case vmIntrinsics::_VectorCompare: 759 return inline_vector_compare(); 760 case vmIntrinsics::_VectorBroadcastInt: 761 return inline_vector_broadcast_int(); 762 case vmIntrinsics::_VectorConvert: 763 return inline_vector_convert(); 764 case vmIntrinsics::_VectorInsert: 765 return inline_vector_insert(); 766 case vmIntrinsics::_VectorExtract: 767 return inline_vector_extract(); 768 case vmIntrinsics::_VectorCompressExpand: 769 return inline_vector_compress_expand(); 770 case vmIntrinsics::_VectorSelectFromTwoVectorOp: 771 return inline_vector_select_from_two_vectors(); 772 case vmIntrinsics::_IndexVector: 773 return inline_index_vector(); 774 case vmIntrinsics::_IndexPartiallyInUpperRange: 775 return inline_index_partially_in_upper_range(); 776 777 case vmIntrinsics::_getObjectSize: 778 return inline_getObjectSize(); 779 780 case vmIntrinsics::_blackhole: 781 return inline_blackhole(); 782 783 default: 784 // If you get here, it may be that someone has added a new intrinsic 785 // to the list in vmIntrinsics.hpp without implementing it here. 786 #ifndef PRODUCT 787 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 788 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 789 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id())); 790 } 791 #endif 792 return false; 793 } 794 } 795 796 Node* LibraryCallKit::try_to_predicate(int predicate) { 797 if (!jvms()->has_method()) { 798 // Root JVMState has a null method. 799 assert(map()->memory()->Opcode() == Op_Parm, ""); 800 // Insert the memory aliasing node 801 set_all_memory(reset_memory()); 802 } 803 assert(merged_memory(), ""); 804 805 switch (intrinsic_id()) { 806 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 807 return inline_cipherBlockChaining_AESCrypt_predicate(false); 808 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 809 return inline_cipherBlockChaining_AESCrypt_predicate(true); 810 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 811 return inline_electronicCodeBook_AESCrypt_predicate(false); 812 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 813 return inline_electronicCodeBook_AESCrypt_predicate(true); 814 case vmIntrinsics::_counterMode_AESCrypt: 815 return inline_counterMode_AESCrypt_predicate(); 816 case vmIntrinsics::_digestBase_implCompressMB: 817 return inline_digestBase_implCompressMB_predicate(predicate); 818 case vmIntrinsics::_galoisCounterMode_AESCrypt: 819 return inline_galoisCounterMode_AESCrypt_predicate(); 820 821 default: 822 // If you get here, it may be that someone has added a new intrinsic 823 // to the list in vmIntrinsics.hpp without implementing it here. 824 #ifndef PRODUCT 825 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 826 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 827 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id())); 828 } 829 #endif 830 Node* slow_ctl = control(); 831 set_control(top()); // No fast path intrinsic 832 return slow_ctl; 833 } 834 } 835 836 //------------------------------set_result------------------------------- 837 // Helper function for finishing intrinsics. 838 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 839 record_for_igvn(region); 840 set_control(_gvn.transform(region)); 841 set_result( _gvn.transform(value)); 842 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 843 } 844 845 //------------------------------generate_guard--------------------------- 846 // Helper function for generating guarded fast-slow graph structures. 847 // The given 'test', if true, guards a slow path. If the test fails 848 // then a fast path can be taken. (We generally hope it fails.) 849 // In all cases, GraphKit::control() is updated to the fast path. 850 // The returned value represents the control for the slow path. 851 // The return value is never 'top'; it is either a valid control 852 // or null if it is obvious that the slow path can never be taken. 853 // Also, if region and the slow control are not null, the slow edge 854 // is appended to the region. 855 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 856 if (stopped()) { 857 // Already short circuited. 858 return nullptr; 859 } 860 861 // Build an if node and its projections. 862 // If test is true we take the slow path, which we assume is uncommon. 863 if (_gvn.type(test) == TypeInt::ZERO) { 864 // The slow branch is never taken. No need to build this guard. 865 return nullptr; 866 } 867 868 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 869 870 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 871 if (if_slow == top()) { 872 // The slow branch is never taken. No need to build this guard. 873 return nullptr; 874 } 875 876 if (region != nullptr) 877 region->add_req(if_slow); 878 879 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 880 set_control(if_fast); 881 882 return if_slow; 883 } 884 885 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 886 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 887 } 888 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 889 return generate_guard(test, region, PROB_FAIR); 890 } 891 892 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 893 Node* *pos_index) { 894 if (stopped()) 895 return nullptr; // already stopped 896 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 897 return nullptr; // index is already adequately typed 898 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 899 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 900 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 901 if (is_neg != nullptr && pos_index != nullptr) { 902 // Emulate effect of Parse::adjust_map_after_if. 903 Node* ccast = new CastIINode(control(), index, TypeInt::POS); 904 (*pos_index) = _gvn.transform(ccast); 905 } 906 return is_neg; 907 } 908 909 // Make sure that 'position' is a valid limit index, in [0..length]. 910 // There are two equivalent plans for checking this: 911 // A. (offset + copyLength) unsigned<= arrayLength 912 // B. offset <= (arrayLength - copyLength) 913 // We require that all of the values above, except for the sum and 914 // difference, are already known to be non-negative. 915 // Plan A is robust in the face of overflow, if offset and copyLength 916 // are both hugely positive. 917 // 918 // Plan B is less direct and intuitive, but it does not overflow at 919 // all, since the difference of two non-negatives is always 920 // representable. Whenever Java methods must perform the equivalent 921 // check they generally use Plan B instead of Plan A. 922 // For the moment we use Plan A. 923 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 924 Node* subseq_length, 925 Node* array_length, 926 RegionNode* region) { 927 if (stopped()) 928 return nullptr; // already stopped 929 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 930 if (zero_offset && subseq_length->eqv_uncast(array_length)) 931 return nullptr; // common case of whole-array copy 932 Node* last = subseq_length; 933 if (!zero_offset) // last += offset 934 last = _gvn.transform(new AddINode(last, offset)); 935 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 936 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 937 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 938 return is_over; 939 } 940 941 // Emit range checks for the given String.value byte array 942 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) { 943 if (stopped()) { 944 return; // already stopped 945 } 946 RegionNode* bailout = new RegionNode(1); 947 record_for_igvn(bailout); 948 if (char_count) { 949 // Convert char count to byte count 950 count = _gvn.transform(new LShiftINode(count, intcon(1))); 951 } 952 953 // Offset and count must not be negative 954 generate_negative_guard(offset, bailout); 955 generate_negative_guard(count, bailout); 956 // Offset + count must not exceed length of array 957 generate_limit_guard(offset, count, load_array_length(array), bailout); 958 959 if (bailout->req() > 1) { 960 PreserveJVMState pjvms(this); 961 set_control(_gvn.transform(bailout)); 962 uncommon_trap(Deoptimization::Reason_intrinsic, 963 Deoptimization::Action_maybe_recompile); 964 } 965 } 966 967 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset, 968 bool is_immutable) { 969 ciKlass* thread_klass = env()->Thread_klass(); 970 const Type* thread_type 971 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 972 973 Node* thread = _gvn.transform(new ThreadLocalNode()); 974 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset)); 975 tls_output = thread; 976 977 Node* thread_obj_handle 978 = (is_immutable 979 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(), 980 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered) 981 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered)); 982 thread_obj_handle = _gvn.transform(thread_obj_handle); 983 984 DecoratorSet decorators = IN_NATIVE; 985 if (is_immutable) { 986 decorators |= C2_IMMUTABLE_MEMORY; 987 } 988 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators); 989 } 990 991 //--------------------------generate_current_thread-------------------- 992 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 993 return current_thread_helper(tls_output, JavaThread::threadObj_offset(), 994 /*is_immutable*/false); 995 } 996 997 //--------------------------generate_virtual_thread-------------------- 998 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) { 999 return current_thread_helper(tls_output, JavaThread::vthread_offset(), 1000 !C->method()->changes_current_thread()); 1001 } 1002 1003 //------------------------------make_string_method_node------------------------ 1004 // Helper method for String intrinsic functions. This version is called with 1005 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 1006 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 1007 // containing the lengths of str1 and str2. 1008 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 1009 Node* result = nullptr; 1010 switch (opcode) { 1011 case Op_StrIndexOf: 1012 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 1013 str1_start, cnt1, str2_start, cnt2, ae); 1014 break; 1015 case Op_StrComp: 1016 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 1017 str1_start, cnt1, str2_start, cnt2, ae); 1018 break; 1019 case Op_StrEquals: 1020 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1021 // Use the constant length if there is one because optimized match rule may exist. 1022 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1023 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1024 break; 1025 default: 1026 ShouldNotReachHere(); 1027 return nullptr; 1028 } 1029 1030 // All these intrinsics have checks. 1031 C->set_has_split_ifs(true); // Has chance for split-if optimization 1032 clear_upper_avx(); 1033 1034 return _gvn.transform(result); 1035 } 1036 1037 //------------------------------inline_string_compareTo------------------------ 1038 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1039 Node* arg1 = argument(0); 1040 Node* arg2 = argument(1); 1041 1042 arg1 = must_be_not_null(arg1, true); 1043 arg2 = must_be_not_null(arg2, true); 1044 1045 // Get start addr and length of first argument 1046 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1047 Node* arg1_cnt = load_array_length(arg1); 1048 1049 // Get start addr and length of second argument 1050 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1051 Node* arg2_cnt = load_array_length(arg2); 1052 1053 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1054 set_result(result); 1055 return true; 1056 } 1057 1058 //------------------------------inline_string_equals------------------------ 1059 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1060 Node* arg1 = argument(0); 1061 Node* arg2 = argument(1); 1062 1063 // paths (plus control) merge 1064 RegionNode* region = new RegionNode(3); 1065 Node* phi = new PhiNode(region, TypeInt::BOOL); 1066 1067 if (!stopped()) { 1068 1069 arg1 = must_be_not_null(arg1, true); 1070 arg2 = must_be_not_null(arg2, true); 1071 1072 // Get start addr and length of first argument 1073 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1074 Node* arg1_cnt = load_array_length(arg1); 1075 1076 // Get start addr and length of second argument 1077 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1078 Node* arg2_cnt = load_array_length(arg2); 1079 1080 // Check for arg1_cnt != arg2_cnt 1081 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1082 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1083 Node* if_ne = generate_slow_guard(bol, nullptr); 1084 if (if_ne != nullptr) { 1085 phi->init_req(2, intcon(0)); 1086 region->init_req(2, if_ne); 1087 } 1088 1089 // Check for count == 0 is done by assembler code for StrEquals. 1090 1091 if (!stopped()) { 1092 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1093 phi->init_req(1, equals); 1094 region->init_req(1, control()); 1095 } 1096 } 1097 1098 // post merge 1099 set_control(_gvn.transform(region)); 1100 record_for_igvn(region); 1101 1102 set_result(_gvn.transform(phi)); 1103 return true; 1104 } 1105 1106 //------------------------------inline_array_equals---------------------------- 1107 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1108 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1109 Node* arg1 = argument(0); 1110 Node* arg2 = argument(1); 1111 1112 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1113 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1114 clear_upper_avx(); 1115 1116 return true; 1117 } 1118 1119 1120 //------------------------------inline_countPositives------------------------------ 1121 bool LibraryCallKit::inline_countPositives() { 1122 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1123 return false; 1124 } 1125 1126 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters"); 1127 // no receiver since it is static method 1128 Node* ba = argument(0); 1129 Node* offset = argument(1); 1130 Node* len = argument(2); 1131 1132 ba = must_be_not_null(ba, true); 1133 1134 // Range checks 1135 generate_string_range_check(ba, offset, len, false); 1136 if (stopped()) { 1137 return true; 1138 } 1139 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1140 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1141 set_result(_gvn.transform(result)); 1142 clear_upper_avx(); 1143 return true; 1144 } 1145 1146 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) { 1147 Node* index = argument(0); 1148 Node* length = bt == T_INT ? argument(1) : argument(2); 1149 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1150 return false; 1151 } 1152 1153 // check that length is positive 1154 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt)); 1155 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1156 1157 { 1158 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1159 uncommon_trap(Deoptimization::Reason_intrinsic, 1160 Deoptimization::Action_make_not_entrant); 1161 } 1162 1163 if (stopped()) { 1164 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success 1165 return true; 1166 } 1167 1168 // length is now known positive, add a cast node to make this explicit 1169 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long(); 1170 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type( 1171 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), 1172 ConstraintCastNode::RegularDependency, bt); 1173 casted_length = _gvn.transform(casted_length); 1174 replace_in_map(length, casted_length); 1175 length = casted_length; 1176 1177 // Use an unsigned comparison for the range check itself 1178 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true)); 1179 BoolTest::mask btest = BoolTest::lt; 1180 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1181 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1182 _gvn.set_type(rc, rc->Value(&_gvn)); 1183 if (!rc_bool->is_Con()) { 1184 record_for_igvn(rc); 1185 } 1186 set_control(_gvn.transform(new IfTrueNode(rc))); 1187 { 1188 PreserveJVMState pjvms(this); 1189 set_control(_gvn.transform(new IfFalseNode(rc))); 1190 uncommon_trap(Deoptimization::Reason_range_check, 1191 Deoptimization::Action_make_not_entrant); 1192 } 1193 1194 if (stopped()) { 1195 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success 1196 return true; 1197 } 1198 1199 // index is now known to be >= 0 and < length, cast it 1200 Node* result = ConstraintCastNode::make_cast_for_basic_type( 1201 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), 1202 ConstraintCastNode::RegularDependency, bt); 1203 result = _gvn.transform(result); 1204 set_result(result); 1205 replace_in_map(index, result); 1206 return true; 1207 } 1208 1209 //------------------------------inline_string_indexOf------------------------ 1210 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1211 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1212 return false; 1213 } 1214 Node* src = argument(0); 1215 Node* tgt = argument(1); 1216 1217 // Make the merge point 1218 RegionNode* result_rgn = new RegionNode(4); 1219 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 1220 1221 src = must_be_not_null(src, true); 1222 tgt = must_be_not_null(tgt, true); 1223 1224 // Get start addr and length of source string 1225 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1226 Node* src_count = load_array_length(src); 1227 1228 // Get start addr and length of substring 1229 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1230 Node* tgt_count = load_array_length(tgt); 1231 1232 Node* result = nullptr; 1233 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr); 1234 1235 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1236 // Divide src size by 2 if String is UTF16 encoded 1237 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1238 } 1239 if (ae == StrIntrinsicNode::UU) { 1240 // Divide substring size by 2 if String is UTF16 encoded 1241 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1242 } 1243 1244 if (call_opt_stub) { 1245 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(), 1246 StubRoutines::_string_indexof_array[ae], 1247 "stringIndexOf", TypePtr::BOTTOM, src_start, 1248 src_count, tgt_start, tgt_count); 1249 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 1250 } else { 1251 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, 1252 result_rgn, result_phi, ae); 1253 } 1254 if (result != nullptr) { 1255 result_phi->init_req(3, result); 1256 result_rgn->init_req(3, control()); 1257 } 1258 set_control(_gvn.transform(result_rgn)); 1259 record_for_igvn(result_rgn); 1260 set_result(_gvn.transform(result_phi)); 1261 1262 return true; 1263 } 1264 1265 //-----------------------------inline_string_indexOfI----------------------- 1266 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1267 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1268 return false; 1269 } 1270 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1271 return false; 1272 } 1273 1274 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1275 Node* src = argument(0); // byte[] 1276 Node* src_count = argument(1); // char count 1277 Node* tgt = argument(2); // byte[] 1278 Node* tgt_count = argument(3); // char count 1279 Node* from_index = argument(4); // char index 1280 1281 src = must_be_not_null(src, true); 1282 tgt = must_be_not_null(tgt, true); 1283 1284 // Multiply byte array index by 2 if String is UTF16 encoded 1285 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1286 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1287 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1288 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1289 1290 // Range checks 1291 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1292 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1293 if (stopped()) { 1294 return true; 1295 } 1296 1297 RegionNode* region = new RegionNode(5); 1298 Node* phi = new PhiNode(region, TypeInt::INT); 1299 Node* result = nullptr; 1300 1301 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr); 1302 1303 if (call_opt_stub) { 1304 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(), 1305 StubRoutines::_string_indexof_array[ae], 1306 "stringIndexOf", TypePtr::BOTTOM, src_start, 1307 src_count, tgt_start, tgt_count); 1308 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 1309 } else { 1310 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, 1311 region, phi, ae); 1312 } 1313 if (result != nullptr) { 1314 // The result is index relative to from_index if substring was found, -1 otherwise. 1315 // Generate code which will fold into cmove. 1316 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1317 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1318 1319 Node* if_lt = generate_slow_guard(bol, nullptr); 1320 if (if_lt != nullptr) { 1321 // result == -1 1322 phi->init_req(3, result); 1323 region->init_req(3, if_lt); 1324 } 1325 if (!stopped()) { 1326 result = _gvn.transform(new AddINode(result, from_index)); 1327 phi->init_req(4, result); 1328 region->init_req(4, control()); 1329 } 1330 } 1331 1332 set_control(_gvn.transform(region)); 1333 record_for_igvn(region); 1334 set_result(_gvn.transform(phi)); 1335 clear_upper_avx(); 1336 1337 return true; 1338 } 1339 1340 // Create StrIndexOfNode with fast path checks 1341 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1342 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1343 // Check for substr count > string count 1344 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1345 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1346 Node* if_gt = generate_slow_guard(bol, nullptr); 1347 if (if_gt != nullptr) { 1348 phi->init_req(1, intcon(-1)); 1349 region->init_req(1, if_gt); 1350 } 1351 if (!stopped()) { 1352 // Check for substr count == 0 1353 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1354 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1355 Node* if_zero = generate_slow_guard(bol, nullptr); 1356 if (if_zero != nullptr) { 1357 phi->init_req(2, intcon(0)); 1358 region->init_req(2, if_zero); 1359 } 1360 } 1361 if (!stopped()) { 1362 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1363 } 1364 return nullptr; 1365 } 1366 1367 //-----------------------------inline_string_indexOfChar----------------------- 1368 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) { 1369 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1370 return false; 1371 } 1372 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1373 return false; 1374 } 1375 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1376 Node* src = argument(0); // byte[] 1377 Node* int_ch = argument(1); 1378 Node* from_index = argument(2); 1379 Node* max = argument(3); 1380 1381 src = must_be_not_null(src, true); 1382 1383 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1384 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1385 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1386 1387 // Range checks 1388 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U); 1389 1390 // Check for int_ch >= 0 1391 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0))); 1392 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge)); 1393 { 1394 BuildCutout unless(this, int_ch_bol, PROB_MAX); 1395 uncommon_trap(Deoptimization::Reason_intrinsic, 1396 Deoptimization::Action_maybe_recompile); 1397 } 1398 if (stopped()) { 1399 return true; 1400 } 1401 1402 RegionNode* region = new RegionNode(3); 1403 Node* phi = new PhiNode(region, TypeInt::INT); 1404 1405 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae); 1406 C->set_has_split_ifs(true); // Has chance for split-if optimization 1407 _gvn.transform(result); 1408 1409 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1410 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1411 1412 Node* if_lt = generate_slow_guard(bol, nullptr); 1413 if (if_lt != nullptr) { 1414 // result == -1 1415 phi->init_req(2, result); 1416 region->init_req(2, if_lt); 1417 } 1418 if (!stopped()) { 1419 result = _gvn.transform(new AddINode(result, from_index)); 1420 phi->init_req(1, result); 1421 region->init_req(1, control()); 1422 } 1423 set_control(_gvn.transform(region)); 1424 record_for_igvn(region); 1425 set_result(_gvn.transform(phi)); 1426 clear_upper_avx(); 1427 1428 return true; 1429 } 1430 //---------------------------inline_string_copy--------------------- 1431 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1432 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1433 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1434 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1435 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1436 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1437 bool LibraryCallKit::inline_string_copy(bool compress) { 1438 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1439 return false; 1440 } 1441 int nargs = 5; // 2 oops, 3 ints 1442 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1443 1444 Node* src = argument(0); 1445 Node* src_offset = argument(1); 1446 Node* dst = argument(2); 1447 Node* dst_offset = argument(3); 1448 Node* length = argument(4); 1449 1450 // Check for allocation before we add nodes that would confuse 1451 // tightly_coupled_allocation() 1452 AllocateArrayNode* alloc = tightly_coupled_allocation(dst); 1453 1454 // Figure out the size and type of the elements we will be copying. 1455 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 1456 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr(); 1457 if (src_type == nullptr || dst_type == nullptr) { 1458 return false; 1459 } 1460 BasicType src_elem = src_type->elem()->array_element_basic_type(); 1461 BasicType dst_elem = dst_type->elem()->array_element_basic_type(); 1462 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1463 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1464 "Unsupported array types for inline_string_copy"); 1465 1466 src = must_be_not_null(src, true); 1467 dst = must_be_not_null(dst, true); 1468 1469 // Convert char[] offsets to byte[] offsets 1470 bool convert_src = (compress && src_elem == T_BYTE); 1471 bool convert_dst = (!compress && dst_elem == T_BYTE); 1472 if (convert_src) { 1473 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1474 } else if (convert_dst) { 1475 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1476 } 1477 1478 // Range checks 1479 generate_string_range_check(src, src_offset, length, convert_src); 1480 generate_string_range_check(dst, dst_offset, length, convert_dst); 1481 if (stopped()) { 1482 return true; 1483 } 1484 1485 Node* src_start = array_element_address(src, src_offset, src_elem); 1486 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1487 // 'src_start' points to src array + scaled offset 1488 // 'dst_start' points to dst array + scaled offset 1489 Node* count = nullptr; 1490 if (compress) { 1491 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1492 } else { 1493 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1494 } 1495 1496 if (alloc != nullptr) { 1497 if (alloc->maybe_set_complete(&_gvn)) { 1498 // "You break it, you buy it." 1499 InitializeNode* init = alloc->initialization(); 1500 assert(init->is_complete(), "we just did this"); 1501 init->set_complete_with_arraycopy(); 1502 assert(dst->is_CheckCastPP(), "sanity"); 1503 assert(dst->in(0)->in(0) == init, "dest pinned"); 1504 } 1505 // Do not let stores that initialize this object be reordered with 1506 // a subsequent store that would make this object accessible by 1507 // other threads. 1508 // Record what AllocateNode this StoreStore protects so that 1509 // escape analysis can go from the MemBarStoreStoreNode to the 1510 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1511 // based on the escape status of the AllocateNode. 1512 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1513 } 1514 if (compress) { 1515 set_result(_gvn.transform(count)); 1516 } 1517 clear_upper_avx(); 1518 1519 return true; 1520 } 1521 1522 #ifdef _LP64 1523 #define XTOP ,top() /*additional argument*/ 1524 #else //_LP64 1525 #define XTOP /*no additional argument*/ 1526 #endif //_LP64 1527 1528 //------------------------inline_string_toBytesU-------------------------- 1529 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1530 bool LibraryCallKit::inline_string_toBytesU() { 1531 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1532 return false; 1533 } 1534 // Get the arguments. 1535 Node* value = argument(0); 1536 Node* offset = argument(1); 1537 Node* length = argument(2); 1538 1539 Node* newcopy = nullptr; 1540 1541 // Set the original stack and the reexecute bit for the interpreter to reexecute 1542 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1543 { PreserveReexecuteState preexecs(this); 1544 jvms()->set_should_reexecute(true); 1545 1546 // Check if a null path was taken unconditionally. 1547 value = null_check(value); 1548 1549 RegionNode* bailout = new RegionNode(1); 1550 record_for_igvn(bailout); 1551 1552 // Range checks 1553 generate_negative_guard(offset, bailout); 1554 generate_negative_guard(length, bailout); 1555 generate_limit_guard(offset, length, load_array_length(value), bailout); 1556 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1557 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1558 1559 if (bailout->req() > 1) { 1560 PreserveJVMState pjvms(this); 1561 set_control(_gvn.transform(bailout)); 1562 uncommon_trap(Deoptimization::Reason_intrinsic, 1563 Deoptimization::Action_maybe_recompile); 1564 } 1565 if (stopped()) { 1566 return true; 1567 } 1568 1569 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1570 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1571 newcopy = new_array(klass_node, size, 0); // no arguments to push 1572 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy); 1573 guarantee(alloc != nullptr, "created above"); 1574 1575 // Calculate starting addresses. 1576 Node* src_start = array_element_address(value, offset, T_CHAR); 1577 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1578 1579 // Check if dst array address is aligned to HeapWordSize 1580 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0); 1581 // If true, then check if src array address is aligned to HeapWordSize 1582 if (aligned) { 1583 const TypeInt* toffset = gvn().type(offset)->is_int(); 1584 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + 1585 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1586 } 1587 1588 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1589 const char* copyfunc_name = "arraycopy"; 1590 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1591 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1592 OptoRuntime::fast_arraycopy_Type(), 1593 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1594 src_start, dst_start, ConvI2X(length) XTOP); 1595 // Do not let reads from the cloned object float above the arraycopy. 1596 if (alloc->maybe_set_complete(&_gvn)) { 1597 // "You break it, you buy it." 1598 InitializeNode* init = alloc->initialization(); 1599 assert(init->is_complete(), "we just did this"); 1600 init->set_complete_with_arraycopy(); 1601 assert(newcopy->is_CheckCastPP(), "sanity"); 1602 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1603 } 1604 // Do not let stores that initialize this object be reordered with 1605 // a subsequent store that would make this object accessible by 1606 // other threads. 1607 // Record what AllocateNode this StoreStore protects so that 1608 // escape analysis can go from the MemBarStoreStoreNode to the 1609 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1610 // based on the escape status of the AllocateNode. 1611 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1612 } // original reexecute is set back here 1613 1614 C->set_has_split_ifs(true); // Has chance for split-if optimization 1615 if (!stopped()) { 1616 set_result(newcopy); 1617 } 1618 clear_upper_avx(); 1619 1620 return true; 1621 } 1622 1623 //------------------------inline_string_getCharsU-------------------------- 1624 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1625 bool LibraryCallKit::inline_string_getCharsU() { 1626 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1627 return false; 1628 } 1629 1630 // Get the arguments. 1631 Node* src = argument(0); 1632 Node* src_begin = argument(1); 1633 Node* src_end = argument(2); // exclusive offset (i < src_end) 1634 Node* dst = argument(3); 1635 Node* dst_begin = argument(4); 1636 1637 // Check for allocation before we add nodes that would confuse 1638 // tightly_coupled_allocation() 1639 AllocateArrayNode* alloc = tightly_coupled_allocation(dst); 1640 1641 // Check if a null path was taken unconditionally. 1642 src = null_check(src); 1643 dst = null_check(dst); 1644 if (stopped()) { 1645 return true; 1646 } 1647 1648 // Get length and convert char[] offset to byte[] offset 1649 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1650 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1651 1652 // Range checks 1653 generate_string_range_check(src, src_begin, length, true); 1654 generate_string_range_check(dst, dst_begin, length, false); 1655 if (stopped()) { 1656 return true; 1657 } 1658 1659 if (!stopped()) { 1660 // Calculate starting addresses. 1661 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1662 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1663 1664 // Check if array addresses are aligned to HeapWordSize 1665 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1666 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1667 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1668 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1669 1670 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1671 const char* copyfunc_name = "arraycopy"; 1672 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1673 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1674 OptoRuntime::fast_arraycopy_Type(), 1675 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1676 src_start, dst_start, ConvI2X(length) XTOP); 1677 // Do not let reads from the cloned object float above the arraycopy. 1678 if (alloc != nullptr) { 1679 if (alloc->maybe_set_complete(&_gvn)) { 1680 // "You break it, you buy it." 1681 InitializeNode* init = alloc->initialization(); 1682 assert(init->is_complete(), "we just did this"); 1683 init->set_complete_with_arraycopy(); 1684 assert(dst->is_CheckCastPP(), "sanity"); 1685 assert(dst->in(0)->in(0) == init, "dest pinned"); 1686 } 1687 // Do not let stores that initialize this object be reordered with 1688 // a subsequent store that would make this object accessible by 1689 // other threads. 1690 // Record what AllocateNode this StoreStore protects so that 1691 // escape analysis can go from the MemBarStoreStoreNode to the 1692 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1693 // based on the escape status of the AllocateNode. 1694 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1695 } else { 1696 insert_mem_bar(Op_MemBarCPUOrder); 1697 } 1698 } 1699 1700 C->set_has_split_ifs(true); // Has chance for split-if optimization 1701 return true; 1702 } 1703 1704 //----------------------inline_string_char_access---------------------------- 1705 // Store/Load char to/from byte[] array. 1706 // static void StringUTF16.putChar(byte[] val, int index, int c) 1707 // static char StringUTF16.getChar(byte[] val, int index) 1708 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1709 Node* value = argument(0); 1710 Node* index = argument(1); 1711 Node* ch = is_store ? argument(2) : nullptr; 1712 1713 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1714 // correctly requires matched array shapes. 1715 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1716 "sanity: byte[] and char[] bases agree"); 1717 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1718 "sanity: byte[] and char[] scales agree"); 1719 1720 // Bail when getChar over constants is requested: constant folding would 1721 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1722 // Java method would constant fold nicely instead. 1723 if (!is_store && value->is_Con() && index->is_Con()) { 1724 return false; 1725 } 1726 1727 // Save state and restore on bailout 1728 SavedState old_state(this); 1729 1730 value = must_be_not_null(value, true); 1731 1732 Node* adr = array_element_address(value, index, T_CHAR); 1733 if (adr->is_top()) { 1734 return false; 1735 } 1736 old_state.discard(); 1737 if (is_store) { 1738 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED); 1739 } else { 1740 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); 1741 set_result(ch); 1742 } 1743 return true; 1744 } 1745 1746 1747 //------------------------------inline_math----------------------------------- 1748 // public static double Math.abs(double) 1749 // public static double Math.sqrt(double) 1750 // public static double Math.log(double) 1751 // public static double Math.log10(double) 1752 // public static double Math.round(double) 1753 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) { 1754 Node* arg = argument(0); 1755 Node* n = nullptr; 1756 switch (id) { 1757 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1758 case vmIntrinsics::_dsqrt: 1759 case vmIntrinsics::_dsqrt_strict: 1760 n = new SqrtDNode(C, control(), arg); break; 1761 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break; 1762 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break; 1763 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break; 1764 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break; 1765 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break; 1766 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break; 1767 default: fatal_unexpected_iid(id); break; 1768 } 1769 set_result(_gvn.transform(n)); 1770 return true; 1771 } 1772 1773 //------------------------------inline_math----------------------------------- 1774 // public static float Math.abs(float) 1775 // public static int Math.abs(int) 1776 // public static long Math.abs(long) 1777 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1778 Node* arg = argument(0); 1779 Node* n = nullptr; 1780 switch (id) { 1781 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break; 1782 case vmIntrinsics::_iabs: n = new AbsINode( arg); break; 1783 case vmIntrinsics::_labs: n = new AbsLNode( arg); break; 1784 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break; 1785 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break; 1786 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break; 1787 default: fatal_unexpected_iid(id); break; 1788 } 1789 set_result(_gvn.transform(n)); 1790 return true; 1791 } 1792 1793 //------------------------------runtime_math----------------------------- 1794 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1795 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1796 "must be (DD)D or (D)D type"); 1797 1798 // Inputs 1799 Node* a = argument(0); 1800 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr; 1801 1802 const TypePtr* no_memory_effects = nullptr; 1803 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName, 1804 no_memory_effects, 1805 a, top(), b, b ? top() : nullptr); 1806 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1807 #ifdef ASSERT 1808 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1809 assert(value_top == top(), "second value must be top"); 1810 #endif 1811 1812 set_result(value); 1813 return true; 1814 } 1815 1816 //------------------------------inline_math_pow----------------------------- 1817 bool LibraryCallKit::inline_math_pow() { 1818 Node* exp = argument(2); 1819 const TypeD* d = _gvn.type(exp)->isa_double_constant(); 1820 if (d != nullptr) { 1821 if (d->getd() == 2.0) { 1822 // Special case: pow(x, 2.0) => x * x 1823 Node* base = argument(0); 1824 set_result(_gvn.transform(new MulDNode(base, base))); 1825 return true; 1826 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) { 1827 // Special case: pow(x, 0.5) => sqrt(x) 1828 Node* base = argument(0); 1829 Node* zero = _gvn.zerocon(T_DOUBLE); 1830 1831 RegionNode* region = new RegionNode(3); 1832 Node* phi = new PhiNode(region, Type::DOUBLE); 1833 1834 Node* cmp = _gvn.transform(new CmpDNode(base, zero)); 1835 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0. 1836 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0). 1837 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0. 1838 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le)); 1839 1840 Node* if_pow = generate_slow_guard(test, nullptr); 1841 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base)); 1842 phi->init_req(1, value_sqrt); 1843 region->init_req(1, control()); 1844 1845 if (if_pow != nullptr) { 1846 set_control(if_pow); 1847 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() : 1848 CAST_FROM_FN_PTR(address, SharedRuntime::dpow); 1849 const TypePtr* no_memory_effects = nullptr; 1850 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW", 1851 no_memory_effects, base, top(), exp, top()); 1852 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1853 #ifdef ASSERT 1854 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1855 assert(value_top == top(), "second value must be top"); 1856 #endif 1857 phi->init_req(2, value_pow); 1858 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control))); 1859 } 1860 1861 C->set_has_split_ifs(true); // Has chance for split-if optimization 1862 set_control(_gvn.transform(region)); 1863 record_for_igvn(region); 1864 set_result(_gvn.transform(phi)); 1865 1866 return true; 1867 } 1868 } 1869 1870 return StubRoutines::dpow() != nullptr ? 1871 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1872 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW"); 1873 } 1874 1875 //------------------------------inline_math_native----------------------------- 1876 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1877 switch (id) { 1878 case vmIntrinsics::_dsin: 1879 return StubRoutines::dsin() != nullptr ? 1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1881 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN"); 1882 case vmIntrinsics::_dcos: 1883 return StubRoutines::dcos() != nullptr ? 1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1885 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS"); 1886 case vmIntrinsics::_dtan: 1887 return StubRoutines::dtan() != nullptr ? 1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") : 1889 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN"); 1890 case vmIntrinsics::_dtanh: 1891 return StubRoutines::dtanh() != nullptr ? 1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false; 1893 case vmIntrinsics::_dcbrt: 1894 return StubRoutines::dcbrt() != nullptr ? 1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false; 1896 case vmIntrinsics::_dexp: 1897 return StubRoutines::dexp() != nullptr ? 1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP"); 1900 case vmIntrinsics::_dlog: 1901 return StubRoutines::dlog() != nullptr ? 1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG"); 1904 case vmIntrinsics::_dlog10: 1905 return StubRoutines::dlog10() != nullptr ? 1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") : 1907 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10"); 1908 1909 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false; 1910 case vmIntrinsics::_ceil: 1911 case vmIntrinsics::_floor: 1912 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false; 1913 1914 case vmIntrinsics::_dsqrt: 1915 case vmIntrinsics::_dsqrt_strict: 1916 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false; 1917 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false; 1918 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false; 1919 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false; 1920 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false; 1921 1922 case vmIntrinsics::_dpow: return inline_math_pow(); 1923 case vmIntrinsics::_dcopySign: return inline_double_math(id); 1924 case vmIntrinsics::_fcopySign: return inline_math(id); 1925 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false; 1926 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false; 1927 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false; 1928 1929 // These intrinsics are not yet correctly implemented 1930 case vmIntrinsics::_datan2: 1931 return false; 1932 1933 default: 1934 fatal_unexpected_iid(id); 1935 return false; 1936 } 1937 } 1938 1939 //----------------------------inline_notify-----------------------------------* 1940 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1941 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1942 address func; 1943 if (id == vmIntrinsics::_notify) { 1944 func = OptoRuntime::monitor_notify_Java(); 1945 } else { 1946 func = OptoRuntime::monitor_notifyAll_Java(); 1947 } 1948 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0)); 1949 make_slow_call_ex(call, env()->Throwable_klass(), false); 1950 return true; 1951 } 1952 1953 1954 //----------------------------inline_min_max----------------------------------- 1955 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1956 Node* a = nullptr; 1957 Node* b = nullptr; 1958 Node* n = nullptr; 1959 switch (id) { 1960 case vmIntrinsics::_min: 1961 case vmIntrinsics::_max: 1962 case vmIntrinsics::_minF: 1963 case vmIntrinsics::_maxF: 1964 case vmIntrinsics::_minF_strict: 1965 case vmIntrinsics::_maxF_strict: 1966 case vmIntrinsics::_min_strict: 1967 case vmIntrinsics::_max_strict: 1968 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each."); 1969 a = argument(0); 1970 b = argument(1); 1971 break; 1972 case vmIntrinsics::_minD: 1973 case vmIntrinsics::_maxD: 1974 case vmIntrinsics::_minD_strict: 1975 case vmIntrinsics::_maxD_strict: 1976 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each."); 1977 a = argument(0); 1978 b = argument(2); 1979 break; 1980 case vmIntrinsics::_minL: 1981 case vmIntrinsics::_maxL: 1982 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each."); 1983 a = argument(0); 1984 b = argument(2); 1985 break; 1986 default: 1987 fatal_unexpected_iid(id); 1988 break; 1989 } 1990 1991 switch (id) { 1992 case vmIntrinsics::_min: 1993 case vmIntrinsics::_min_strict: 1994 n = new MinINode(a, b); 1995 break; 1996 case vmIntrinsics::_max: 1997 case vmIntrinsics::_max_strict: 1998 n = new MaxINode(a, b); 1999 break; 2000 case vmIntrinsics::_minF: 2001 case vmIntrinsics::_minF_strict: 2002 n = new MinFNode(a, b); 2003 break; 2004 case vmIntrinsics::_maxF: 2005 case vmIntrinsics::_maxF_strict: 2006 n = new MaxFNode(a, b); 2007 break; 2008 case vmIntrinsics::_minD: 2009 case vmIntrinsics::_minD_strict: 2010 n = new MinDNode(a, b); 2011 break; 2012 case vmIntrinsics::_maxD: 2013 case vmIntrinsics::_maxD_strict: 2014 n = new MaxDNode(a, b); 2015 break; 2016 case vmIntrinsics::_minL: 2017 n = new MinLNode(_gvn.C, a, b); 2018 break; 2019 case vmIntrinsics::_maxL: 2020 n = new MaxLNode(_gvn.C, a, b); 2021 break; 2022 default: 2023 fatal_unexpected_iid(id); 2024 break; 2025 } 2026 2027 set_result(_gvn.transform(n)); 2028 return true; 2029 } 2030 2031 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) { 2032 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic, 2033 env()->ArithmeticException_instance())) { 2034 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces), 2035 // so let's bail out intrinsic rather than risking deopting again. 2036 return false; 2037 } 2038 2039 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 2040 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 2041 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 2042 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 2043 2044 { 2045 PreserveJVMState pjvms(this); 2046 PreserveReexecuteState preexecs(this); 2047 jvms()->set_should_reexecute(true); 2048 2049 set_control(slow_path); 2050 set_i_o(i_o()); 2051 2052 builtin_throw(Deoptimization::Reason_intrinsic, 2053 env()->ArithmeticException_instance(), 2054 /*allow_too_many_traps*/ false); 2055 } 2056 2057 set_control(fast_path); 2058 set_result(math); 2059 return true; 2060 } 2061 2062 template <typename OverflowOp> 2063 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 2064 typedef typename OverflowOp::MathOp MathOp; 2065 2066 MathOp* mathOp = new MathOp(arg1, arg2); 2067 Node* operation = _gvn.transform( mathOp ); 2068 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 2069 return inline_math_mathExact(operation, ofcheck); 2070 } 2071 2072 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 2073 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 2074 } 2075 2076 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 2077 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 2078 } 2079 2080 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 2081 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 2082 } 2083 2084 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 2085 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 2086 } 2087 2088 bool LibraryCallKit::inline_math_negateExactI() { 2089 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 2090 } 2091 2092 bool LibraryCallKit::inline_math_negateExactL() { 2093 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 2094 } 2095 2096 bool LibraryCallKit::inline_math_multiplyExactI() { 2097 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 2098 } 2099 2100 bool LibraryCallKit::inline_math_multiplyExactL() { 2101 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 2102 } 2103 2104 bool LibraryCallKit::inline_math_multiplyHigh() { 2105 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2)))); 2106 return true; 2107 } 2108 2109 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() { 2110 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2)))); 2111 return true; 2112 } 2113 2114 inline int 2115 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) { 2116 const TypePtr* base_type = TypePtr::NULL_PTR; 2117 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr(); 2118 if (base_type == nullptr) { 2119 // Unknown type. 2120 return Type::AnyPtr; 2121 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) { 2122 // Since this is a null+long form, we have to switch to a rawptr. 2123 base = _gvn.transform(new CastX2PNode(offset)); 2124 offset = MakeConX(0); 2125 return Type::RawPtr; 2126 } else if (base_type->base() == Type::RawPtr) { 2127 return Type::RawPtr; 2128 } else if (base_type->isa_oopptr()) { 2129 // Base is never null => always a heap address. 2130 if (!TypePtr::NULL_PTR->higher_equal(base_type)) { 2131 return Type::OopPtr; 2132 } 2133 // Offset is small => always a heap address. 2134 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2135 if (offset_type != nullptr && 2136 base_type->offset() == 0 && // (should always be?) 2137 offset_type->_lo >= 0 && 2138 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2139 return Type::OopPtr; 2140 } else if (type == T_OBJECT) { 2141 // off heap access to an oop doesn't make any sense. Has to be on 2142 // heap. 2143 return Type::OopPtr; 2144 } 2145 // Otherwise, it might either be oop+off or null+addr. 2146 return Type::AnyPtr; 2147 } else { 2148 // No information: 2149 return Type::AnyPtr; 2150 } 2151 } 2152 2153 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) { 2154 Node* uncasted_base = base; 2155 int kind = classify_unsafe_addr(uncasted_base, offset, type); 2156 if (kind == Type::RawPtr) { 2157 return basic_plus_adr(top(), uncasted_base, offset); 2158 } else if (kind == Type::AnyPtr) { 2159 assert(base == uncasted_base, "unexpected base change"); 2160 if (can_cast) { 2161 if (!_gvn.type(base)->speculative_maybe_null() && 2162 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 2163 // According to profiling, this access is always on 2164 // heap. Casting the base to not null and thus avoiding membars 2165 // around the access should allow better optimizations 2166 Node* null_ctl = top(); 2167 base = null_check_oop(base, &null_ctl, true, true, true); 2168 assert(null_ctl->is_top(), "no null control here"); 2169 return basic_plus_adr(base, offset); 2170 } else if (_gvn.type(base)->speculative_always_null() && 2171 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 2172 // According to profiling, this access is always off 2173 // heap. 2174 base = null_assert(base); 2175 Node* raw_base = _gvn.transform(new CastX2PNode(offset)); 2176 offset = MakeConX(0); 2177 return basic_plus_adr(top(), raw_base, offset); 2178 } 2179 } 2180 // We don't know if it's an on heap or off heap access. Fall back 2181 // to raw memory access. 2182 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM)); 2183 return basic_plus_adr(top(), raw, offset); 2184 } else { 2185 assert(base == uncasted_base, "unexpected base change"); 2186 // We know it's an on heap access so base can't be null 2187 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) { 2188 base = must_be_not_null(base, true); 2189 } 2190 return basic_plus_adr(base, offset); 2191 } 2192 } 2193 2194 //--------------------------inline_number_methods----------------------------- 2195 // inline int Integer.numberOfLeadingZeros(int) 2196 // inline int Long.numberOfLeadingZeros(long) 2197 // 2198 // inline int Integer.numberOfTrailingZeros(int) 2199 // inline int Long.numberOfTrailingZeros(long) 2200 // 2201 // inline int Integer.bitCount(int) 2202 // inline int Long.bitCount(long) 2203 // 2204 // inline char Character.reverseBytes(char) 2205 // inline short Short.reverseBytes(short) 2206 // inline int Integer.reverseBytes(int) 2207 // inline long Long.reverseBytes(long) 2208 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2209 Node* arg = argument(0); 2210 Node* n = nullptr; 2211 switch (id) { 2212 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2213 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2214 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2215 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2216 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2217 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2218 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break; 2219 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break; 2220 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break; 2221 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break; 2222 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break; 2223 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break; 2224 default: fatal_unexpected_iid(id); break; 2225 } 2226 set_result(_gvn.transform(n)); 2227 return true; 2228 } 2229 2230 //--------------------------inline_bitshuffle_methods----------------------------- 2231 // inline int Integer.compress(int, int) 2232 // inline int Integer.expand(int, int) 2233 // inline long Long.compress(long, long) 2234 // inline long Long.expand(long, long) 2235 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) { 2236 Node* n = nullptr; 2237 switch (id) { 2238 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break; 2239 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break; 2240 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break; 2241 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break; 2242 default: fatal_unexpected_iid(id); break; 2243 } 2244 set_result(_gvn.transform(n)); 2245 return true; 2246 } 2247 2248 //--------------------------inline_number_methods----------------------------- 2249 // inline int Integer.compareUnsigned(int, int) 2250 // inline int Long.compareUnsigned(long, long) 2251 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) { 2252 Node* arg1 = argument(0); 2253 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1); 2254 Node* n = nullptr; 2255 switch (id) { 2256 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break; 2257 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break; 2258 default: fatal_unexpected_iid(id); break; 2259 } 2260 set_result(_gvn.transform(n)); 2261 return true; 2262 } 2263 2264 //--------------------------inline_unsigned_divmod_methods----------------------------- 2265 // inline int Integer.divideUnsigned(int, int) 2266 // inline int Integer.remainderUnsigned(int, int) 2267 // inline long Long.divideUnsigned(long, long) 2268 // inline long Long.remainderUnsigned(long, long) 2269 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) { 2270 Node* n = nullptr; 2271 switch (id) { 2272 case vmIntrinsics::_divideUnsigned_i: { 2273 zero_check_int(argument(1)); 2274 // Compile-time detect of null-exception 2275 if (stopped()) { 2276 return true; // keep the graph constructed so far 2277 } 2278 n = new UDivINode(control(), argument(0), argument(1)); 2279 break; 2280 } 2281 case vmIntrinsics::_divideUnsigned_l: { 2282 zero_check_long(argument(2)); 2283 // Compile-time detect of null-exception 2284 if (stopped()) { 2285 return true; // keep the graph constructed so far 2286 } 2287 n = new UDivLNode(control(), argument(0), argument(2)); 2288 break; 2289 } 2290 case vmIntrinsics::_remainderUnsigned_i: { 2291 zero_check_int(argument(1)); 2292 // Compile-time detect of null-exception 2293 if (stopped()) { 2294 return true; // keep the graph constructed so far 2295 } 2296 n = new UModINode(control(), argument(0), argument(1)); 2297 break; 2298 } 2299 case vmIntrinsics::_remainderUnsigned_l: { 2300 zero_check_long(argument(2)); 2301 // Compile-time detect of null-exception 2302 if (stopped()) { 2303 return true; // keep the graph constructed so far 2304 } 2305 n = new UModLNode(control(), argument(0), argument(2)); 2306 break; 2307 } 2308 default: fatal_unexpected_iid(id); break; 2309 } 2310 set_result(_gvn.transform(n)); 2311 return true; 2312 } 2313 2314 //----------------------------inline_unsafe_access---------------------------- 2315 2316 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) { 2317 // Attempt to infer a sharper value type from the offset and base type. 2318 ciKlass* sharpened_klass = nullptr; 2319 2320 // See if it is an instance field, with an object type. 2321 if (alias_type->field() != nullptr) { 2322 if (alias_type->field()->type()->is_klass()) { 2323 sharpened_klass = alias_type->field()->type()->as_klass(); 2324 } 2325 } 2326 2327 const TypeOopPtr* result = nullptr; 2328 // See if it is a narrow oop array. 2329 if (adr_type->isa_aryptr()) { 2330 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2331 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr(); 2332 if (elem_type != nullptr && elem_type->is_loaded()) { 2333 // Sharpen the value type. 2334 result = elem_type; 2335 } 2336 } 2337 } 2338 2339 // The sharpened class might be unloaded if there is no class loader 2340 // contraint in place. 2341 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) { 2342 // Sharpen the value type. 2343 result = TypeOopPtr::make_from_klass(sharpened_klass); 2344 } 2345 if (result != nullptr) { 2346 #ifndef PRODUCT 2347 if (C->print_intrinsics() || C->print_inlining()) { 2348 tty->print(" from base type: "); adr_type->dump(); tty->cr(); 2349 tty->print(" sharpened value: "); result->dump(); tty->cr(); 2350 } 2351 #endif 2352 } 2353 return result; 2354 } 2355 2356 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) { 2357 switch (kind) { 2358 case Relaxed: 2359 return MO_UNORDERED; 2360 case Opaque: 2361 return MO_RELAXED; 2362 case Acquire: 2363 return MO_ACQUIRE; 2364 case Release: 2365 return MO_RELEASE; 2366 case Volatile: 2367 return MO_SEQ_CST; 2368 default: 2369 ShouldNotReachHere(); 2370 return 0; 2371 } 2372 } 2373 2374 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) : 2375 _kit(kit), 2376 _sp(kit->sp()), 2377 _jvms(kit->jvms()), 2378 _map(kit->clone_map()), 2379 _discarded(false) 2380 { 2381 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) { 2382 Node* out = kit->control()->fast_out(i); 2383 if (out->is_CFG()) { 2384 _ctrl_succ.push(out); 2385 } 2386 } 2387 } 2388 2389 LibraryCallKit::SavedState::~SavedState() { 2390 if (_discarded) { 2391 _kit->destruct_map_clone(_map); 2392 return; 2393 } 2394 _kit->jvms()->set_map(_map); 2395 _kit->jvms()->set_sp(_sp); 2396 _map->set_jvms(_kit->jvms()); 2397 _kit->set_map(_map); 2398 _kit->set_sp(_sp); 2399 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) { 2400 Node* out = _kit->control()->fast_out(i); 2401 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) { 2402 _kit->_gvn.hash_delete(out); 2403 out->set_req(0, _kit->C->top()); 2404 _kit->C->record_for_igvn(out); 2405 --i; --imax; 2406 _kit->_gvn.hash_find_insert(out); 2407 } 2408 } 2409 } 2410 2411 void LibraryCallKit::SavedState::discard() { 2412 _discarded = true; 2413 } 2414 2415 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) { 2416 if (callee()->is_static()) return false; // caller must have the capability! 2417 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2418 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2419 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2420 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2421 2422 if (is_reference_type(type)) { 2423 decorators |= ON_UNKNOWN_OOP_REF; 2424 } 2425 2426 if (unaligned) { 2427 decorators |= C2_UNALIGNED; 2428 } 2429 2430 #ifndef PRODUCT 2431 { 2432 ResourceMark rm; 2433 // Check the signatures. 2434 ciSignature* sig = callee()->signature(); 2435 #ifdef ASSERT 2436 if (!is_store) { 2437 // Object getReference(Object base, int/long offset), etc. 2438 BasicType rtype = sig->return_type()->basic_type(); 2439 assert(rtype == type, "getter must return the expected value"); 2440 assert(sig->count() == 2, "oop getter has 2 arguments"); 2441 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2442 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2443 } else { 2444 // void putReference(Object base, int/long offset, Object x), etc. 2445 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2446 assert(sig->count() == 3, "oop putter has 3 arguments"); 2447 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2448 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2449 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2450 assert(vtype == type, "putter must accept the expected value"); 2451 } 2452 #endif // ASSERT 2453 } 2454 #endif //PRODUCT 2455 2456 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2457 2458 Node* receiver = argument(0); // type: oop 2459 2460 // Build address expression. 2461 Node* heap_base_oop = top(); 2462 2463 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2464 Node* base = argument(1); // type: oop 2465 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2466 Node* offset = argument(2); // type: long 2467 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2468 // to be plain byte offsets, which are also the same as those accepted 2469 // by oopDesc::field_addr. 2470 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2471 "fieldOffset must be byte-scaled"); 2472 // 32-bit machines ignore the high half! 2473 offset = ConvL2X(offset); 2474 2475 // Save state and restore on bailout 2476 SavedState old_state(this); 2477 2478 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed); 2479 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly"); 2480 2481 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) { 2482 if (type != T_OBJECT) { 2483 decorators |= IN_NATIVE; // off-heap primitive access 2484 } else { 2485 return false; // off-heap oop accesses are not supported 2486 } 2487 } else { 2488 heap_base_oop = base; // on-heap or mixed access 2489 } 2490 2491 // Can base be null? Otherwise, always on-heap access. 2492 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base)); 2493 2494 if (!can_access_non_heap) { 2495 decorators |= IN_HEAP; 2496 } 2497 2498 Node* val = is_store ? argument(4) : nullptr; 2499 2500 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr(); 2501 if (adr_type == TypePtr::NULL_PTR) { 2502 return false; // off-heap access with zero address 2503 } 2504 2505 // Try to categorize the address. 2506 Compile::AliasType* alias_type = C->alias_type(adr_type); 2507 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2508 2509 if (alias_type->adr_type() == TypeInstPtr::KLASS || 2510 alias_type->adr_type() == TypeAryPtr::RANGE) { 2511 return false; // not supported 2512 } 2513 2514 bool mismatched = false; 2515 BasicType bt = alias_type->basic_type(); 2516 if (bt != T_ILLEGAL) { 2517 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access"); 2518 if (bt == T_BYTE && adr_type->isa_aryptr()) { 2519 // Alias type doesn't differentiate between byte[] and boolean[]). 2520 // Use address type to get the element type. 2521 bt = adr_type->is_aryptr()->elem()->array_element_basic_type(); 2522 } 2523 if (is_reference_type(bt, true)) { 2524 // accessing an array field with getReference is not a mismatch 2525 bt = T_OBJECT; 2526 } 2527 if ((bt == T_OBJECT) != (type == T_OBJECT)) { 2528 // Don't intrinsify mismatched object accesses 2529 return false; 2530 } 2531 mismatched = (bt != type); 2532 } else if (alias_type->adr_type()->isa_oopptr()) { 2533 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched 2534 } 2535 2536 old_state.discard(); 2537 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched"); 2538 2539 if (mismatched) { 2540 decorators |= C2_MISMATCHED; 2541 } 2542 2543 // First guess at the value type. 2544 const Type *value_type = Type::get_const_basic_type(type); 2545 2546 // Figure out the memory ordering. 2547 decorators |= mo_decorator_for_access_kind(kind); 2548 2549 if (!is_store && type == T_OBJECT) { 2550 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2551 if (tjp != nullptr) { 2552 value_type = tjp; 2553 } 2554 } 2555 2556 receiver = null_check(receiver); 2557 if (stopped()) { 2558 return true; 2559 } 2560 // Heap pointers get a null-check from the interpreter, 2561 // as a courtesy. However, this is not guaranteed by Unsafe, 2562 // and it is not possible to fully distinguish unintended nulls 2563 // from intended ones in this API. 2564 2565 if (!is_store) { 2566 Node* p = nullptr; 2567 // Try to constant fold a load from a constant field 2568 ciField* field = alias_type->field(); 2569 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) { 2570 // final or stable field 2571 p = make_constant_from_field(field, heap_base_oop); 2572 } 2573 2574 if (p == nullptr) { // Could not constant fold the load 2575 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators); 2576 // Normalize the value returned by getBoolean in the following cases 2577 if (type == T_BOOLEAN && 2578 (mismatched || 2579 heap_base_oop == top() || // - heap_base_oop is null or 2580 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null 2581 // and the unsafe access is made to large offset 2582 // (i.e., larger than the maximum offset necessary for any 2583 // field access) 2584 ) { 2585 IdealKit ideal = IdealKit(this); 2586 #define __ ideal. 2587 IdealVariable normalized_result(ideal); 2588 __ declarations_done(); 2589 __ set(normalized_result, p); 2590 __ if_then(p, BoolTest::ne, ideal.ConI(0)); 2591 __ set(normalized_result, ideal.ConI(1)); 2592 ideal.end_if(); 2593 final_sync(ideal); 2594 p = __ value(normalized_result); 2595 #undef __ 2596 } 2597 } 2598 if (type == T_ADDRESS) { 2599 p = gvn().transform(new CastP2XNode(nullptr, p)); 2600 p = ConvX2UL(p); 2601 } 2602 // The load node has the control of the preceding MemBarCPUOrder. All 2603 // following nodes will have the control of the MemBarCPUOrder inserted at 2604 // the end of this method. So, pushing the load onto the stack at a later 2605 // point is fine. 2606 set_result(p); 2607 } else { 2608 if (bt == T_ADDRESS) { 2609 // Repackage the long as a pointer. 2610 val = ConvL2X(val); 2611 val = gvn().transform(new CastX2PNode(val)); 2612 } 2613 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators); 2614 } 2615 2616 return true; 2617 } 2618 2619 //----------------------------inline_unsafe_load_store---------------------------- 2620 // This method serves a couple of different customers (depending on LoadStoreKind): 2621 // 2622 // LS_cmp_swap: 2623 // 2624 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x); 2625 // boolean compareAndSetInt( Object o, long offset, int expected, int x); 2626 // boolean compareAndSetLong( Object o, long offset, long expected, long x); 2627 // 2628 // LS_cmp_swap_weak: 2629 // 2630 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x); 2631 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x); 2632 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x); 2633 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x); 2634 // 2635 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x); 2636 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x); 2637 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x); 2638 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x); 2639 // 2640 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x); 2641 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x); 2642 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x); 2643 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x); 2644 // 2645 // LS_cmp_exchange: 2646 // 2647 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x); 2648 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x); 2649 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x); 2650 // 2651 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2652 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2653 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2654 // 2655 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2656 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2657 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2658 // 2659 // LS_get_add: 2660 // 2661 // int getAndAddInt( Object o, long offset, int delta) 2662 // long getAndAddLong(Object o, long offset, long delta) 2663 // 2664 // LS_get_set: 2665 // 2666 // int getAndSet(Object o, long offset, int newValue) 2667 // long getAndSet(Object o, long offset, long newValue) 2668 // Object getAndSet(Object o, long offset, Object newValue) 2669 // 2670 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2671 // This basic scheme here is the same as inline_unsafe_access, but 2672 // differs in enough details that combining them would make the code 2673 // overly confusing. (This is a true fact! I originally combined 2674 // them, but even I was confused by it!) As much code/comments as 2675 // possible are retained from inline_unsafe_access though to make 2676 // the correspondences clearer. - dl 2677 2678 if (callee()->is_static()) return false; // caller must have the capability! 2679 2680 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2681 decorators |= mo_decorator_for_access_kind(access_kind); 2682 2683 #ifndef PRODUCT 2684 BasicType rtype; 2685 { 2686 ResourceMark rm; 2687 // Check the signatures. 2688 ciSignature* sig = callee()->signature(); 2689 rtype = sig->return_type()->basic_type(); 2690 switch(kind) { 2691 case LS_get_add: 2692 case LS_get_set: { 2693 // Check the signatures. 2694 #ifdef ASSERT 2695 assert(rtype == type, "get and set must return the expected type"); 2696 assert(sig->count() == 3, "get and set has 3 arguments"); 2697 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2698 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2699 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2700 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation"); 2701 #endif // ASSERT 2702 break; 2703 } 2704 case LS_cmp_swap: 2705 case LS_cmp_swap_weak: { 2706 // Check the signatures. 2707 #ifdef ASSERT 2708 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2709 assert(sig->count() == 4, "CAS has 4 arguments"); 2710 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2711 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2712 #endif // ASSERT 2713 break; 2714 } 2715 case LS_cmp_exchange: { 2716 // Check the signatures. 2717 #ifdef ASSERT 2718 assert(rtype == type, "CAS must return the expected type"); 2719 assert(sig->count() == 4, "CAS has 4 arguments"); 2720 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2721 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2722 #endif // ASSERT 2723 break; 2724 } 2725 default: 2726 ShouldNotReachHere(); 2727 } 2728 } 2729 #endif //PRODUCT 2730 2731 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2732 2733 // Get arguments: 2734 Node* receiver = nullptr; 2735 Node* base = nullptr; 2736 Node* offset = nullptr; 2737 Node* oldval = nullptr; 2738 Node* newval = nullptr; 2739 switch(kind) { 2740 case LS_cmp_swap: 2741 case LS_cmp_swap_weak: 2742 case LS_cmp_exchange: { 2743 const bool two_slot_type = type2size[type] == 2; 2744 receiver = argument(0); // type: oop 2745 base = argument(1); // type: oop 2746 offset = argument(2); // type: long 2747 oldval = argument(4); // type: oop, int, or long 2748 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2749 break; 2750 } 2751 case LS_get_add: 2752 case LS_get_set: { 2753 receiver = argument(0); // type: oop 2754 base = argument(1); // type: oop 2755 offset = argument(2); // type: long 2756 oldval = nullptr; 2757 newval = argument(4); // type: oop, int, or long 2758 break; 2759 } 2760 default: 2761 ShouldNotReachHere(); 2762 } 2763 2764 // Build field offset expression. 2765 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2766 // to be plain byte offsets, which are also the same as those accepted 2767 // by oopDesc::field_addr. 2768 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2769 // 32-bit machines ignore the high half of long offsets 2770 offset = ConvL2X(offset); 2771 // Save state and restore on bailout 2772 SavedState old_state(this); 2773 Node* adr = make_unsafe_address(base, offset,type, false); 2774 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2775 2776 Compile::AliasType* alias_type = C->alias_type(adr_type); 2777 BasicType bt = alias_type->basic_type(); 2778 if (bt != T_ILLEGAL && 2779 (is_reference_type(bt) != (type == T_OBJECT))) { 2780 // Don't intrinsify mismatched object accesses. 2781 return false; 2782 } 2783 2784 old_state.discard(); 2785 2786 // For CAS, unlike inline_unsafe_access, there seems no point in 2787 // trying to refine types. Just use the coarse types here. 2788 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2789 const Type *value_type = Type::get_const_basic_type(type); 2790 2791 switch (kind) { 2792 case LS_get_set: 2793 case LS_cmp_exchange: { 2794 if (type == T_OBJECT) { 2795 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2796 if (tjp != nullptr) { 2797 value_type = tjp; 2798 } 2799 } 2800 break; 2801 } 2802 case LS_cmp_swap: 2803 case LS_cmp_swap_weak: 2804 case LS_get_add: 2805 break; 2806 default: 2807 ShouldNotReachHere(); 2808 } 2809 2810 // Null check receiver. 2811 receiver = null_check(receiver); 2812 if (stopped()) { 2813 return true; 2814 } 2815 2816 int alias_idx = C->get_alias_index(adr_type); 2817 2818 if (is_reference_type(type)) { 2819 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF; 2820 2821 // Transformation of a value which could be null pointer (CastPP #null) 2822 // could be delayed during Parse (for example, in adjust_map_after_if()). 2823 // Execute transformation here to avoid barrier generation in such case. 2824 if (_gvn.type(newval) == TypePtr::NULL_PTR) 2825 newval = _gvn.makecon(TypePtr::NULL_PTR); 2826 2827 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) { 2828 // Refine the value to a null constant, when it is known to be null 2829 oldval = _gvn.makecon(TypePtr::NULL_PTR); 2830 } 2831 } 2832 2833 Node* result = nullptr; 2834 switch (kind) { 2835 case LS_cmp_exchange: { 2836 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx, 2837 oldval, newval, value_type, type, decorators); 2838 break; 2839 } 2840 case LS_cmp_swap_weak: 2841 decorators |= C2_WEAK_CMPXCHG; 2842 case LS_cmp_swap: { 2843 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx, 2844 oldval, newval, value_type, type, decorators); 2845 break; 2846 } 2847 case LS_get_set: { 2848 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx, 2849 newval, value_type, type, decorators); 2850 break; 2851 } 2852 case LS_get_add: { 2853 result = access_atomic_add_at(base, adr, adr_type, alias_idx, 2854 newval, value_type, type, decorators); 2855 break; 2856 } 2857 default: 2858 ShouldNotReachHere(); 2859 } 2860 2861 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 2862 set_result(result); 2863 return true; 2864 } 2865 2866 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 2867 // Regardless of form, don't allow previous ld/st to move down, 2868 // then issue acquire, release, or volatile mem_bar. 2869 insert_mem_bar(Op_MemBarCPUOrder); 2870 switch(id) { 2871 case vmIntrinsics::_loadFence: 2872 insert_mem_bar(Op_LoadFence); 2873 return true; 2874 case vmIntrinsics::_storeFence: 2875 insert_mem_bar(Op_StoreFence); 2876 return true; 2877 case vmIntrinsics::_storeStoreFence: 2878 insert_mem_bar(Op_StoreStoreFence); 2879 return true; 2880 case vmIntrinsics::_fullFence: 2881 insert_mem_bar(Op_MemBarVolatile); 2882 return true; 2883 default: 2884 fatal_unexpected_iid(id); 2885 return false; 2886 } 2887 } 2888 2889 bool LibraryCallKit::inline_onspinwait() { 2890 insert_mem_bar(Op_OnSpinWait); 2891 return true; 2892 } 2893 2894 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 2895 if (!kls->is_Con()) { 2896 return true; 2897 } 2898 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr(); 2899 if (klsptr == nullptr) { 2900 return true; 2901 } 2902 ciInstanceKlass* ik = klsptr->instance_klass(); 2903 // don't need a guard for a klass that is already initialized 2904 return !ik->is_initialized(); 2905 } 2906 2907 //----------------------------inline_unsafe_writeback0------------------------- 2908 // public native void Unsafe.writeback0(long address) 2909 bool LibraryCallKit::inline_unsafe_writeback0() { 2910 if (!Matcher::has_match_rule(Op_CacheWB)) { 2911 return false; 2912 } 2913 #ifndef PRODUCT 2914 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync"); 2915 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync"); 2916 ciSignature* sig = callee()->signature(); 2917 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!"); 2918 #endif 2919 null_check_receiver(); // null-check, then ignore 2920 Node *addr = argument(1); 2921 addr = new CastX2PNode(addr); 2922 addr = _gvn.transform(addr); 2923 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr); 2924 flush = _gvn.transform(flush); 2925 set_memory(flush, TypeRawPtr::BOTTOM); 2926 return true; 2927 } 2928 2929 //----------------------------inline_unsafe_writeback0------------------------- 2930 // public native void Unsafe.writeback0(long address) 2931 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) { 2932 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) { 2933 return false; 2934 } 2935 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) { 2936 return false; 2937 } 2938 #ifndef PRODUCT 2939 assert(Matcher::has_match_rule(Op_CacheWB), 2940 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB" 2941 : "found match rule for CacheWBPostSync but not CacheWB")); 2942 2943 #endif 2944 null_check_receiver(); // null-check, then ignore 2945 Node *sync; 2946 if (is_pre) { 2947 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2948 } else { 2949 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2950 } 2951 sync = _gvn.transform(sync); 2952 set_memory(sync, TypeRawPtr::BOTTOM); 2953 return true; 2954 } 2955 2956 //----------------------------inline_unsafe_allocate--------------------------- 2957 // public native Object Unsafe.allocateInstance(Class<?> cls); 2958 bool LibraryCallKit::inline_unsafe_allocate() { 2959 2960 #if INCLUDE_JVMTI 2961 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 2962 return false; 2963 } 2964 #endif //INCLUDE_JVMTI 2965 2966 if (callee()->is_static()) return false; // caller must have the capability! 2967 2968 null_check_receiver(); // null-check, then ignore 2969 Node* cls = null_check(argument(1)); 2970 if (stopped()) return true; 2971 2972 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0); 2973 kls = null_check(kls); 2974 if (stopped()) return true; // argument was like int.class 2975 2976 #if INCLUDE_JVMTI 2977 // Don't try to access new allocated obj in the intrinsic. 2978 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled. 2979 // Deoptimize and allocate in interpreter instead. 2980 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc)); 2981 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered); 2982 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0))); 2983 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq)); 2984 { 2985 BuildCutout unless(this, tst, PROB_MAX); 2986 uncommon_trap(Deoptimization::Reason_intrinsic, 2987 Deoptimization::Action_make_not_entrant); 2988 } 2989 if (stopped()) { 2990 return true; 2991 } 2992 #endif //INCLUDE_JVMTI 2993 2994 Node* test = nullptr; 2995 if (LibraryCallKit::klass_needs_init_guard(kls)) { 2996 // Note: The argument might still be an illegal value like 2997 // Serializable.class or Object[].class. The runtime will handle it. 2998 // But we must make an explicit check for initialization. 2999 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 3000 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 3001 // can generate code to load it as unsigned byte. 3002 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire); 3003 Node* bits = intcon(InstanceKlass::fully_initialized); 3004 test = _gvn.transform(new SubINode(inst, bits)); 3005 // The 'test' is non-zero if we need to take a slow path. 3006 } 3007 3008 Node* obj = new_instance(kls, test); 3009 set_result(obj); 3010 return true; 3011 } 3012 3013 //------------------------inline_native_time_funcs-------------- 3014 // inline code for System.currentTimeMillis() and System.nanoTime() 3015 // these have the same type and signature 3016 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 3017 const TypeFunc* tf = OptoRuntime::void_long_Type(); 3018 const TypePtr* no_memory_effects = nullptr; 3019 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 3020 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 3021 #ifdef ASSERT 3022 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 3023 assert(value_top == top(), "second value must be top"); 3024 #endif 3025 set_result(value); 3026 return true; 3027 } 3028 3029 3030 #if INCLUDE_JVMTI 3031 3032 // When notifications are disabled then just update the VTMS transition bit and return. 3033 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol. 3034 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) { 3035 if (!DoJVMTIVirtualThreadTransitions) { 3036 return true; 3037 } 3038 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument 3039 IdealKit ideal(this); 3040 3041 Node* ONE = ideal.ConI(1); 3042 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1))); 3043 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events)); 3044 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw); 3045 3046 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); { 3047 sync_kit(ideal); 3048 // if notifyJvmti enabled then make a call to the given SharedRuntime function 3049 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type(); 3050 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide); 3051 ideal.sync_kit(this); 3052 } ideal.else_(); { 3053 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object 3054 Node* thread = ideal.thread(); 3055 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset())); 3056 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset()); 3057 3058 sync_kit(ideal); 3059 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3060 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3061 3062 ideal.sync_kit(this); 3063 } ideal.end_if(); 3064 final_sync(ideal); 3065 3066 return true; 3067 } 3068 3069 // Always update the is_disable_suspend bit. 3070 bool LibraryCallKit::inline_native_notify_jvmti_sync() { 3071 if (!DoJVMTIVirtualThreadTransitions) { 3072 return true; 3073 } 3074 IdealKit ideal(this); 3075 3076 { 3077 // unconditionally update the is_disable_suspend bit in current JavaThread 3078 Node* thread = ideal.thread(); 3079 Node* arg = _gvn.transform(argument(0)); // argument for notification 3080 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset())); 3081 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr(); 3082 3083 sync_kit(ideal); 3084 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3085 ideal.sync_kit(this); 3086 } 3087 final_sync(ideal); 3088 3089 return true; 3090 } 3091 3092 #endif // INCLUDE_JVMTI 3093 3094 #ifdef JFR_HAVE_INTRINSICS 3095 3096 /** 3097 * if oop->klass != null 3098 * // normal class 3099 * epoch = _epoch_state ? 2 : 1 3100 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch { 3101 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts 3102 * } 3103 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path 3104 * else 3105 * // primitive class 3106 * if oop->array_klass != null 3107 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path 3108 * else 3109 * id = LAST_TYPE_ID + 1 // void class path 3110 * if (!signaled) 3111 * signaled = true 3112 */ 3113 bool LibraryCallKit::inline_native_classID() { 3114 Node* cls = argument(0); 3115 3116 IdealKit ideal(this); 3117 #define __ ideal. 3118 IdealVariable result(ideal); __ declarations_done(); 3119 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), 3120 basic_plus_adr(cls, java_lang_Class::klass_offset()), 3121 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 3122 3123 3124 __ if_then(kls, BoolTest::ne, null()); { 3125 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET)); 3126 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw); 3127 3128 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address())); 3129 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw); 3130 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch)); 3131 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT))); 3132 mask = _gvn.transform(new OrLNode(mask, epoch)); 3133 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask)); 3134 3135 float unlikely = PROB_UNLIKELY(0.999); 3136 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); { 3137 sync_kit(ideal); 3138 make_runtime_call(RC_LEAF, 3139 OptoRuntime::class_id_load_barrier_Type(), 3140 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier), 3141 "class id load barrier", 3142 TypePtr::BOTTOM, 3143 kls); 3144 ideal.sync_kit(this); 3145 } __ end_if(); 3146 3147 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)))); 3148 } __ else_(); { 3149 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), 3150 basic_plus_adr(cls, java_lang_Class::array_klass_offset()), 3151 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 3152 __ if_then(array_kls, BoolTest::ne, null()); { 3153 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET)); 3154 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw); 3155 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))); 3156 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1)))); 3157 } __ else_(); { 3158 // void class case 3159 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1))); 3160 } __ end_if(); 3161 3162 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address())); 3163 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire); 3164 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); { 3165 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true); 3166 } __ end_if(); 3167 } __ end_if(); 3168 3169 final_sync(ideal); 3170 set_result(ideal.value(result)); 3171 #undef __ 3172 return true; 3173 } 3174 3175 //------------------------inline_native_jvm_commit------------------ 3176 bool LibraryCallKit::inline_native_jvm_commit() { 3177 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3178 3179 // Save input memory and i_o state. 3180 Node* input_memory_state = reset_memory(); 3181 set_all_memory(input_memory_state); 3182 Node* input_io_state = i_o(); 3183 3184 // TLS. 3185 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3186 // Jfr java buffer. 3187 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))))); 3188 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered)); 3189 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))))); 3190 3191 // Load the current value of the notified field in the JfrThreadLocal. 3192 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR)); 3193 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered); 3194 3195 // Test for notification. 3196 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1))); 3197 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq)); 3198 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN); 3199 3200 // True branch, is notified. 3201 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified)); 3202 set_control(is_notified); 3203 3204 // Reset notified state. 3205 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered); 3206 Node* notified_reset_memory = reset_memory(); 3207 3208 // 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. 3209 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered)); 3210 // Convert the machine-word to a long. 3211 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X)); 3212 3213 // False branch, not notified. 3214 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified)); 3215 set_control(not_notified); 3216 set_all_memory(input_memory_state); 3217 3218 // Arg is the next position as a long. 3219 Node* arg = argument(0); 3220 // Convert long to machine-word. 3221 Node* next_pos_X = _gvn.transform(ConvL2X(arg)); 3222 3223 // Store the next_position to the underlying jfr java buffer. 3224 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release); 3225 3226 Node* commit_memory = reset_memory(); 3227 set_all_memory(commit_memory); 3228 3229 // 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. 3230 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))))); 3231 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered); 3232 Node* lease_constant = _gvn.transform(_gvn.intcon(4)); 3233 3234 // And flags with lease constant. 3235 Node* lease = _gvn.transform(new AndINode(flags, lease_constant)); 3236 3237 // Branch on lease to conditionalize returning the leased java buffer. 3238 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant)); 3239 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq)); 3240 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN); 3241 3242 // False branch, not a lease. 3243 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease)); 3244 3245 // True branch, is lease. 3246 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease)); 3247 set_control(is_lease); 3248 3249 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop. 3250 Node* call_return_lease = make_runtime_call(RC_NO_LEAF, 3251 OptoRuntime::void_void_Type(), 3252 SharedRuntime::jfr_return_lease(), 3253 "return_lease", TypePtr::BOTTOM); 3254 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control)); 3255 3256 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT); 3257 record_for_igvn(lease_compare_rgn); 3258 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3259 record_for_igvn(lease_compare_mem); 3260 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO); 3261 record_for_igvn(lease_compare_io); 3262 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG); 3263 record_for_igvn(lease_result_value); 3264 3265 // Update control and phi nodes. 3266 lease_compare_rgn->init_req(_true_path, call_return_lease_control); 3267 lease_compare_rgn->init_req(_false_path, not_lease); 3268 3269 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3270 lease_compare_mem->init_req(_false_path, commit_memory); 3271 3272 lease_compare_io->init_req(_true_path, i_o()); 3273 lease_compare_io->init_req(_false_path, input_io_state); 3274 3275 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0. 3276 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position. 3277 3278 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3279 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3280 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO); 3281 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG); 3282 3283 // Update control and phi nodes. 3284 result_rgn->init_req(_true_path, is_notified); 3285 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn)); 3286 3287 result_mem->init_req(_true_path, notified_reset_memory); 3288 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem)); 3289 3290 result_io->init_req(_true_path, input_io_state); 3291 result_io->init_req(_false_path, _gvn.transform(lease_compare_io)); 3292 3293 result_value->init_req(_true_path, current_pos); 3294 result_value->init_req(_false_path, _gvn.transform(lease_result_value)); 3295 3296 // Set output state. 3297 set_control(_gvn.transform(result_rgn)); 3298 set_all_memory(_gvn.transform(result_mem)); 3299 set_i_o(_gvn.transform(result_io)); 3300 set_result(result_rgn, result_value); 3301 return true; 3302 } 3303 3304 /* 3305 * The intrinsic is a model of this pseudo-code: 3306 * 3307 * JfrThreadLocal* const tl = Thread::jfr_thread_local() 3308 * jobject h_event_writer = tl->java_event_writer(); 3309 * if (h_event_writer == nullptr) { 3310 * return nullptr; 3311 * } 3312 * oop threadObj = Thread::threadObj(); 3313 * oop vthread = java_lang_Thread::vthread(threadObj); 3314 * traceid tid; 3315 * bool pinVirtualThread; 3316 * bool excluded; 3317 * if (vthread != threadObj) { // i.e. current thread is virtual 3318 * tid = java_lang_Thread::tid(vthread); 3319 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread); 3320 * pinVirtualThread = VMContinuations; 3321 * excluded = vthread_epoch_raw & excluded_mask; 3322 * if (!excluded) { 3323 * traceid current_epoch = JfrTraceIdEpoch::current_generation(); 3324 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask; 3325 * if (vthread_epoch != current_epoch) { 3326 * write_checkpoint(); 3327 * } 3328 * } 3329 * } else { 3330 * tid = java_lang_Thread::tid(threadObj); 3331 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj); 3332 * pinVirtualThread = false; 3333 * excluded = thread_epoch_raw & excluded_mask; 3334 * } 3335 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer); 3336 * traceid tid_in_event_writer = getField(event_writer, "threadID"); 3337 * if (tid_in_event_writer != tid) { 3338 * setField(event_writer, "pinVirtualThread", pinVirtualThread); 3339 * setField(event_writer, "excluded", excluded); 3340 * setField(event_writer, "threadID", tid); 3341 * } 3342 * return event_writer 3343 */ 3344 bool LibraryCallKit::inline_native_getEventWriter() { 3345 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3346 3347 // Save input memory and i_o state. 3348 Node* input_memory_state = reset_memory(); 3349 set_all_memory(input_memory_state); 3350 Node* input_io_state = i_o(); 3351 3352 // The most significant bit of the u2 is used to denote thread exclusion 3353 Node* excluded_shift = _gvn.intcon(15); 3354 Node* excluded_mask = _gvn.intcon(1 << 15); 3355 // The epoch generation is the range [1-32767] 3356 Node* epoch_mask = _gvn.intcon(32767); 3357 3358 // TLS 3359 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3360 3361 // Load the address of java event writer jobject handle from the jfr_thread_local structure. 3362 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)); 3363 3364 // Load the eventwriter jobject handle. 3365 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered); 3366 3367 // Null check the jobject handle. 3368 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null())); 3369 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne)); 3370 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN); 3371 3372 // False path, jobj is null. 3373 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null)); 3374 3375 // True path, jobj is not null. 3376 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null)); 3377 3378 set_control(jobj_is_not_null); 3379 3380 // Load the threadObj for the CarrierThread. 3381 Node* threadObj = generate_current_thread(tls_ptr); 3382 3383 // Load the vthread. 3384 Node* vthread = generate_virtual_thread(tls_ptr); 3385 3386 // If vthread != threadObj, this is a virtual thread. 3387 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj)); 3388 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne)); 3389 IfNode* iff_vthread_not_equal_threadObj = 3390 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN); 3391 3392 // False branch, fallback to threadObj. 3393 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj)); 3394 set_control(vthread_equal_threadObj); 3395 3396 // Load the tid field from the vthread object. 3397 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J"); 3398 3399 // Load the raw epoch value from the threadObj. 3400 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset()); 3401 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, 3402 _gvn.type(threadObj_epoch_offset)->isa_ptr(), 3403 TypeInt::CHAR, T_CHAR, 3404 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3405 3406 // Mask off the excluded information from the epoch. 3407 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask)); 3408 3409 // True branch, this is a virtual thread. 3410 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj)); 3411 set_control(vthread_not_equal_threadObj); 3412 3413 // Load the tid field from the vthread object. 3414 Node* vthread_tid = load_field_from_object(vthread, "tid", "J"); 3415 3416 // Continuation support determines if a virtual thread should be pinned. 3417 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations)); 3418 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered); 3419 3420 // Load the raw epoch value from the vthread. 3421 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset()); 3422 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(), 3423 TypeInt::CHAR, T_CHAR, 3424 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3425 3426 // Mask off the excluded information from the epoch. 3427 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask))); 3428 3429 // Branch on excluded to conditionalize updating the epoch for the virtual thread. 3430 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask))); 3431 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne)); 3432 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN); 3433 3434 // False branch, vthread is excluded, no need to write epoch info. 3435 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded)); 3436 3437 // True branch, vthread is included, update epoch info. 3438 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded)); 3439 set_control(included); 3440 3441 // Get epoch value. 3442 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask))); 3443 3444 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR. 3445 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address())); 3446 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered); 3447 3448 // Compare the epoch in the vthread to the current epoch generation. 3449 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch)); 3450 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne)); 3451 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN); 3452 3453 // False path, epoch is equal, checkpoint information is valid. 3454 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal)); 3455 3456 // True path, epoch is not equal, write a checkpoint for the vthread. 3457 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal)); 3458 3459 set_control(epoch_is_not_equal); 3460 3461 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch. 3462 // The call also updates the native thread local thread id and the vthread with the current epoch. 3463 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF, 3464 OptoRuntime::jfr_write_checkpoint_Type(), 3465 SharedRuntime::jfr_write_checkpoint(), 3466 "write_checkpoint", TypePtr::BOTTOM); 3467 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control)); 3468 3469 // vthread epoch != current epoch 3470 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT); 3471 record_for_igvn(epoch_compare_rgn); 3472 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3473 record_for_igvn(epoch_compare_mem); 3474 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO); 3475 record_for_igvn(epoch_compare_io); 3476 3477 // Update control and phi nodes. 3478 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control); 3479 epoch_compare_rgn->init_req(_false_path, epoch_is_equal); 3480 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3481 epoch_compare_mem->init_req(_false_path, input_memory_state); 3482 epoch_compare_io->init_req(_true_path, i_o()); 3483 epoch_compare_io->init_req(_false_path, input_io_state); 3484 3485 // excluded != true 3486 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT); 3487 record_for_igvn(exclude_compare_rgn); 3488 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3489 record_for_igvn(exclude_compare_mem); 3490 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO); 3491 record_for_igvn(exclude_compare_io); 3492 3493 // Update control and phi nodes. 3494 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn)); 3495 exclude_compare_rgn->init_req(_false_path, excluded); 3496 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem)); 3497 exclude_compare_mem->init_req(_false_path, input_memory_state); 3498 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io)); 3499 exclude_compare_io->init_req(_false_path, input_io_state); 3500 3501 // vthread != threadObj 3502 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT); 3503 record_for_igvn(vthread_compare_rgn); 3504 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3505 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO); 3506 record_for_igvn(vthread_compare_io); 3507 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG); 3508 record_for_igvn(tid); 3509 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR); 3510 record_for_igvn(exclusion); 3511 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL); 3512 record_for_igvn(pinVirtualThread); 3513 3514 // Update control and phi nodes. 3515 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn)); 3516 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj); 3517 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem)); 3518 vthread_compare_mem->init_req(_false_path, input_memory_state); 3519 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io)); 3520 vthread_compare_io->init_req(_false_path, input_io_state); 3521 tid->init_req(_true_path, _gvn.transform(vthread_tid)); 3522 tid->init_req(_false_path, _gvn.transform(thread_obj_tid)); 3523 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded)); 3524 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded)); 3525 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support)); 3526 pinVirtualThread->init_req(_false_path, _gvn.intcon(0)); 3527 3528 // Update branch state. 3529 set_control(_gvn.transform(vthread_compare_rgn)); 3530 set_all_memory(_gvn.transform(vthread_compare_mem)); 3531 set_i_o(_gvn.transform(vthread_compare_io)); 3532 3533 // Load the event writer oop by dereferencing the jobject handle. 3534 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter")); 3535 assert(klass_EventWriter->is_loaded(), "invariant"); 3536 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass(); 3537 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter); 3538 const TypeOopPtr* const xtype = aklass->as_instance_type(); 3539 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global))); 3540 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD); 3541 3542 // Load the current thread id from the event writer object. 3543 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J"); 3544 // Get the field offset to, conditionally, store an updated tid value later. 3545 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false); 3546 // Get the field offset to, conditionally, store an updated exclusion value later. 3547 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false); 3548 // Get the field offset to, conditionally, store an updated pinVirtualThread value later. 3549 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false); 3550 3551 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT); 3552 record_for_igvn(event_writer_tid_compare_rgn); 3553 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3554 record_for_igvn(event_writer_tid_compare_mem); 3555 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO); 3556 record_for_igvn(event_writer_tid_compare_io); 3557 3558 // Compare the current tid from the thread object to what is currently stored in the event writer object. 3559 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid))); 3560 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne)); 3561 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN); 3562 3563 // False path, tids are the same. 3564 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal)); 3565 3566 // True path, tid is not equal, need to update the tid in the event writer. 3567 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal)); 3568 record_for_igvn(tid_is_not_equal); 3569 3570 // Store the pin state to the event writer. 3571 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered); 3572 3573 // Store the exclusion state to the event writer. 3574 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift)); 3575 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered); 3576 3577 // Store the tid to the event writer. 3578 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered); 3579 3580 // Update control and phi nodes. 3581 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal); 3582 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal); 3583 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3584 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem)); 3585 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o())); 3586 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io)); 3587 3588 // Result of top level CFG, Memory, IO and Value. 3589 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3590 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3591 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO); 3592 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM); 3593 3594 // Result control. 3595 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn)); 3596 result_rgn->init_req(_false_path, jobj_is_null); 3597 3598 // Result memory. 3599 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem)); 3600 result_mem->init_req(_false_path, _gvn.transform(input_memory_state)); 3601 3602 // Result IO. 3603 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io)); 3604 result_io->init_req(_false_path, _gvn.transform(input_io_state)); 3605 3606 // Result value. 3607 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop 3608 result_value->init_req(_false_path, null()); // return null 3609 3610 // Set output state. 3611 set_control(_gvn.transform(result_rgn)); 3612 set_all_memory(_gvn.transform(result_mem)); 3613 set_i_o(_gvn.transform(result_io)); 3614 set_result(result_rgn, result_value); 3615 return true; 3616 } 3617 3618 /* 3619 * The intrinsic is a model of this pseudo-code: 3620 * 3621 * JfrThreadLocal* const tl = thread->jfr_thread_local(); 3622 * if (carrierThread != thread) { // is virtual thread 3623 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread); 3624 * bool excluded = vthread_epoch_raw & excluded_mask; 3625 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread)); 3626 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded); 3627 * if (!excluded) { 3628 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask; 3629 * Atomic::store(&tl->_vthread_epoch, vthread_epoch); 3630 * } 3631 * Atomic::release_store(&tl->_vthread, true); 3632 * return; 3633 * } 3634 * Atomic::release_store(&tl->_vthread, false); 3635 */ 3636 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) { 3637 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3638 3639 Node* input_memory_state = reset_memory(); 3640 set_all_memory(input_memory_state); 3641 3642 // The most significant bit of the u2 is used to denote thread exclusion 3643 Node* excluded_mask = _gvn.intcon(1 << 15); 3644 // The epoch generation is the range [1-32767] 3645 Node* epoch_mask = _gvn.intcon(32767); 3646 3647 Node* const carrierThread = generate_current_thread(jt); 3648 // If thread != carrierThread, this is a virtual thread. 3649 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread)); 3650 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne)); 3651 IfNode* iff_thread_not_equal_carrierThread = 3652 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN); 3653 3654 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR)); 3655 3656 // False branch, is carrierThread. 3657 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread)); 3658 // Store release 3659 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true); 3660 3661 set_all_memory(input_memory_state); 3662 3663 // True branch, is virtual thread. 3664 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread)); 3665 set_control(thread_not_equal_carrierThread); 3666 3667 // Load the raw epoch value from the vthread. 3668 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset()); 3669 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR, 3670 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3671 3672 // Mask off the excluded information from the epoch. 3673 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask))); 3674 3675 // Load the tid field from the thread. 3676 Node* tid = load_field_from_object(thread, "tid", "J"); 3677 3678 // Store the vthread tid to the jfr thread local. 3679 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR)); 3680 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true); 3681 3682 // Branch is_excluded to conditionalize updating the epoch . 3683 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask))); 3684 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq)); 3685 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN); 3686 3687 // True branch, vthread is excluded, no need to write epoch info. 3688 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded)); 3689 set_control(excluded); 3690 Node* vthread_is_excluded = _gvn.intcon(1); 3691 3692 // False branch, vthread is included, update epoch info. 3693 Node* included = _gvn.transform(new IfFalseNode(iff_excluded)); 3694 set_control(included); 3695 Node* vthread_is_included = _gvn.intcon(0); 3696 3697 // Get epoch value. 3698 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask))); 3699 3700 // Store the vthread epoch to the jfr thread local. 3701 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR)); 3702 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true); 3703 3704 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT); 3705 record_for_igvn(excluded_rgn); 3706 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM); 3707 record_for_igvn(excluded_mem); 3708 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL); 3709 record_for_igvn(exclusion); 3710 3711 // Merge the excluded control and memory. 3712 excluded_rgn->init_req(_true_path, excluded); 3713 excluded_rgn->init_req(_false_path, included); 3714 excluded_mem->init_req(_true_path, tid_memory); 3715 excluded_mem->init_req(_false_path, included_memory); 3716 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded)); 3717 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included)); 3718 3719 // Set intermediate state. 3720 set_control(_gvn.transform(excluded_rgn)); 3721 set_all_memory(excluded_mem); 3722 3723 // Store the vthread exclusion state to the jfr thread local. 3724 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR)); 3725 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true); 3726 3727 // Store release 3728 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true); 3729 3730 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT); 3731 record_for_igvn(thread_compare_rgn); 3732 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3733 record_for_igvn(thread_compare_mem); 3734 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL); 3735 record_for_igvn(vthread); 3736 3737 // Merge the thread_compare control and memory. 3738 thread_compare_rgn->init_req(_true_path, control()); 3739 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread); 3740 thread_compare_mem->init_req(_true_path, vthread_true_memory); 3741 thread_compare_mem->init_req(_false_path, vthread_false_memory); 3742 3743 // Set output state. 3744 set_control(_gvn.transform(thread_compare_rgn)); 3745 set_all_memory(_gvn.transform(thread_compare_mem)); 3746 } 3747 3748 #endif // JFR_HAVE_INTRINSICS 3749 3750 //------------------------inline_native_currentCarrierThread------------------ 3751 bool LibraryCallKit::inline_native_currentCarrierThread() { 3752 Node* junk = nullptr; 3753 set_result(generate_current_thread(junk)); 3754 return true; 3755 } 3756 3757 //------------------------inline_native_currentThread------------------ 3758 bool LibraryCallKit::inline_native_currentThread() { 3759 Node* junk = nullptr; 3760 set_result(generate_virtual_thread(junk)); 3761 return true; 3762 } 3763 3764 //------------------------inline_native_setVthread------------------ 3765 bool LibraryCallKit::inline_native_setCurrentThread() { 3766 assert(C->method()->changes_current_thread(), 3767 "method changes current Thread but is not annotated ChangesCurrentThread"); 3768 Node* arr = argument(1); 3769 Node* thread = _gvn.transform(new ThreadLocalNode()); 3770 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset())); 3771 Node* thread_obj_handle 3772 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered); 3773 thread_obj_handle = _gvn.transform(thread_obj_handle); 3774 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr(); 3775 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED); 3776 3777 // Change the _monitor_owner_id of the JavaThread 3778 Node* tid = load_field_from_object(arr, "tid", "J"); 3779 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset())); 3780 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true); 3781 3782 JFR_ONLY(extend_setCurrentThread(thread, arr);) 3783 return true; 3784 } 3785 3786 const Type* LibraryCallKit::scopedValueCache_type() { 3787 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass()); 3788 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass()); 3789 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS); 3790 3791 // Because we create the scopedValue cache lazily we have to make the 3792 // type of the result BotPTR. 3793 bool xk = etype->klass_is_exact(); 3794 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0); 3795 return objects_type; 3796 } 3797 3798 Node* LibraryCallKit::scopedValueCache_helper() { 3799 Node* thread = _gvn.transform(new ThreadLocalNode()); 3800 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset())); 3801 // We cannot use immutable_memory() because we might flip onto a 3802 // different carrier thread, at which point we'll need to use that 3803 // carrier thread's cache. 3804 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(), 3805 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)); 3806 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered); 3807 } 3808 3809 //------------------------inline_native_scopedValueCache------------------ 3810 bool LibraryCallKit::inline_native_scopedValueCache() { 3811 Node* cache_obj_handle = scopedValueCache_helper(); 3812 const Type* objects_type = scopedValueCache_type(); 3813 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE)); 3814 3815 return true; 3816 } 3817 3818 //------------------------inline_native_setScopedValueCache------------------ 3819 bool LibraryCallKit::inline_native_setScopedValueCache() { 3820 Node* arr = argument(0); 3821 Node* cache_obj_handle = scopedValueCache_helper(); 3822 const Type* objects_type = scopedValueCache_type(); 3823 3824 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr(); 3825 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED); 3826 3827 return true; 3828 } 3829 3830 //------------------------inline_native_Continuation_pin and unpin----------- 3831 3832 // Shared implementation routine for both pin and unpin. 3833 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) { 3834 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3835 3836 // Save input memory. 3837 Node* input_memory_state = reset_memory(); 3838 set_all_memory(input_memory_state); 3839 3840 // TLS 3841 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3842 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset())); 3843 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered); 3844 3845 // Null check the last continuation object. 3846 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null())); 3847 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne)); 3848 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN); 3849 3850 // False path, last continuation is null. 3851 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null)); 3852 3853 // True path, last continuation is not null. 3854 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null)); 3855 3856 set_control(continuation_is_not_null); 3857 3858 // Load the pin count from the last continuation. 3859 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset())); 3860 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered); 3861 3862 // The loaded pin count is compared against a context specific rhs for over/underflow detection. 3863 Node* pin_count_rhs; 3864 if (unpin) { 3865 pin_count_rhs = _gvn.intcon(0); 3866 } else { 3867 pin_count_rhs = _gvn.intcon(UINT32_MAX); 3868 } 3869 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs)); 3870 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq)); 3871 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN); 3872 3873 // True branch, pin count over/underflow. 3874 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow)); 3875 { 3876 // Trap (but not deoptimize (Action_none)) and continue in the interpreter 3877 // which will throw IllegalStateException for pin count over/underflow. 3878 // No memory changed so far - we can use memory create by reset_memory() 3879 // at the beginning of this intrinsic. No need to call reset_memory() again. 3880 PreserveJVMState pjvms(this); 3881 set_control(pin_count_over_underflow); 3882 uncommon_trap(Deoptimization::Reason_intrinsic, 3883 Deoptimization::Action_none); 3884 assert(stopped(), "invariant"); 3885 } 3886 3887 // False branch, no pin count over/underflow. Increment or decrement pin count and store back. 3888 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow)); 3889 set_control(valid_pin_count); 3890 3891 Node* next_pin_count; 3892 if (unpin) { 3893 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1))); 3894 } else { 3895 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1))); 3896 } 3897 3898 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered); 3899 3900 // Result of top level CFG and Memory. 3901 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3902 record_for_igvn(result_rgn); 3903 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3904 record_for_igvn(result_mem); 3905 3906 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count)); 3907 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null)); 3908 result_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3909 result_mem->init_req(_false_path, _gvn.transform(input_memory_state)); 3910 3911 // Set output state. 3912 set_control(_gvn.transform(result_rgn)); 3913 set_all_memory(_gvn.transform(result_mem)); 3914 3915 return true; 3916 } 3917 3918 //---------------------------load_mirror_from_klass---------------------------- 3919 // Given a klass oop, load its java mirror (a java.lang.Class oop). 3920 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) { 3921 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset())); 3922 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3923 // mirror = ((OopHandle)mirror)->resolve(); 3924 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE); 3925 } 3926 3927 //-----------------------load_klass_from_mirror_common------------------------- 3928 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3929 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3930 // and branch to the given path on the region. 3931 // If never_see_null, take an uncommon trap on null, so we can optimistically 3932 // compile for the non-null case. 3933 // If the region is null, force never_see_null = true. 3934 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3935 bool never_see_null, 3936 RegionNode* region, 3937 int null_path, 3938 int offset) { 3939 if (region == nullptr) never_see_null = true; 3940 Node* p = basic_plus_adr(mirror, offset); 3941 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL; 3942 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3943 Node* null_ctl = top(); 3944 kls = null_check_oop(kls, &null_ctl, never_see_null); 3945 if (region != nullptr) { 3946 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3947 region->init_req(null_path, null_ctl); 3948 } else { 3949 assert(null_ctl == top(), "no loose ends"); 3950 } 3951 return kls; 3952 } 3953 3954 //--------------------(inline_native_Class_query helpers)--------------------- 3955 // Use this for JVM_ACC_INTERFACE. 3956 // Fall through if (mods & mask) == bits, take the guard otherwise. 3957 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region, 3958 ByteSize offset, const Type* type, BasicType bt) { 3959 // Branch around if the given klass has the given modifier bit set. 3960 // Like generate_guard, adds a new path onto the region. 3961 Node* modp = basic_plus_adr(kls, in_bytes(offset)); 3962 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered); 3963 Node* mask = intcon(modifier_mask); 3964 Node* bits = intcon(modifier_bits); 3965 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3966 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3967 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3968 return generate_fair_guard(bol, region); 3969 } 3970 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3971 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region, 3972 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR); 3973 } 3974 3975 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast. 3976 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 3977 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region, 3978 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN); 3979 } 3980 3981 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) { 3982 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region); 3983 } 3984 3985 //-------------------------inline_native_Class_query------------------- 3986 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 3987 const Type* return_type = TypeInt::BOOL; 3988 Node* prim_return_value = top(); // what happens if it's a primitive class? 3989 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3990 bool expect_prim = false; // most of these guys expect to work on refs 3991 3992 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 3993 3994 Node* mirror = argument(0); 3995 Node* obj = top(); 3996 3997 switch (id) { 3998 case vmIntrinsics::_isInstance: 3999 // nothing is an instance of a primitive type 4000 prim_return_value = intcon(0); 4001 obj = argument(1); 4002 break; 4003 case vmIntrinsics::_isHidden: 4004 prim_return_value = intcon(0); 4005 break; 4006 case vmIntrinsics::_getSuperclass: 4007 prim_return_value = null(); 4008 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 4009 break; 4010 case vmIntrinsics::_getClassAccessFlags: 4011 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 4012 return_type = TypeInt::CHAR; 4013 break; 4014 default: 4015 fatal_unexpected_iid(id); 4016 break; 4017 } 4018 4019 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 4020 if (mirror_con == nullptr) return false; // cannot happen? 4021 4022 #ifndef PRODUCT 4023 if (C->print_intrinsics() || C->print_inlining()) { 4024 ciType* k = mirror_con->java_mirror_type(); 4025 if (k) { 4026 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 4027 k->print_name(); 4028 tty->cr(); 4029 } 4030 } 4031 #endif 4032 4033 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 4034 RegionNode* region = new RegionNode(PATH_LIMIT); 4035 record_for_igvn(region); 4036 PhiNode* phi = new PhiNode(region, return_type); 4037 4038 // The mirror will never be null of Reflection.getClassAccessFlags, however 4039 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 4040 // if it is. See bug 4774291. 4041 4042 // For Reflection.getClassAccessFlags(), the null check occurs in 4043 // the wrong place; see inline_unsafe_access(), above, for a similar 4044 // situation. 4045 mirror = null_check(mirror); 4046 // If mirror or obj is dead, only null-path is taken. 4047 if (stopped()) return true; 4048 4049 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 4050 4051 // Now load the mirror's klass metaobject, and null-check it. 4052 // Side-effects region with the control path if the klass is null. 4053 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 4054 // If kls is null, we have a primitive mirror. 4055 phi->init_req(_prim_path, prim_return_value); 4056 if (stopped()) { set_result(region, phi); return true; } 4057 bool safe_for_replace = (region->in(_prim_path) == top()); 4058 4059 Node* p; // handy temp 4060 Node* null_ctl; 4061 4062 // Now that we have the non-null klass, we can perform the real query. 4063 // For constant classes, the query will constant-fold in LoadNode::Value. 4064 Node* query_value = top(); 4065 switch (id) { 4066 case vmIntrinsics::_isInstance: 4067 // nothing is an instance of a primitive type 4068 query_value = gen_instanceof(obj, kls, safe_for_replace); 4069 break; 4070 4071 case vmIntrinsics::_isHidden: 4072 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.) 4073 if (generate_hidden_class_guard(kls, region) != nullptr) 4074 // A guard was added. If the guard is taken, it was an hidden class. 4075 phi->add_req(intcon(1)); 4076 // If we fall through, it's a plain class. 4077 query_value = intcon(0); 4078 break; 4079 4080 4081 case vmIntrinsics::_getSuperclass: 4082 // The rules here are somewhat unfortunate, but we can still do better 4083 // with random logic than with a JNI call. 4084 // Interfaces store null or Object as _super, but must report null. 4085 // Arrays store an intermediate super as _super, but must report Object. 4086 // Other types can report the actual _super. 4087 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 4088 if (generate_interface_guard(kls, region) != nullptr) 4089 // A guard was added. If the guard is taken, it was an interface. 4090 phi->add_req(null()); 4091 if (generate_array_guard(kls, region) != nullptr) 4092 // A guard was added. If the guard is taken, it was an array. 4093 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 4094 // If we fall through, it's a plain class. Get its _super. 4095 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 4096 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 4097 null_ctl = top(); 4098 kls = null_check_oop(kls, &null_ctl); 4099 if (null_ctl != top()) { 4100 // If the guard is taken, Object.superClass is null (both klass and mirror). 4101 region->add_req(null_ctl); 4102 phi ->add_req(null()); 4103 } 4104 if (!stopped()) { 4105 query_value = load_mirror_from_klass(kls); 4106 } 4107 break; 4108 4109 case vmIntrinsics::_getClassAccessFlags: 4110 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 4111 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered); 4112 break; 4113 4114 default: 4115 fatal_unexpected_iid(id); 4116 break; 4117 } 4118 4119 // Fall-through is the normal case of a query to a real class. 4120 phi->init_req(1, query_value); 4121 region->init_req(1, control()); 4122 4123 C->set_has_split_ifs(true); // Has chance for split-if optimization 4124 set_result(region, phi); 4125 return true; 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 ciType* tm = mirror_con->java_mirror_type(); 4144 if (tm != nullptr && tm->is_klass() && 4145 tp != nullptr) { 4146 if (!tp->is_loaded()) { 4147 // Don't use intrinsic when class is not loaded. 4148 return false; 4149 } else { 4150 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type()); 4151 if (static_res == Compile::SSC_always_true) { 4152 // isInstance() is true - fold the code. 4153 set_result(obj); 4154 return true; 4155 } else if (static_res == Compile::SSC_always_false) { 4156 // Don't use intrinsic, have to throw ClassCastException. 4157 // If the reference is null, the non-intrinsic bytecode will 4158 // be optimized appropriately. 4159 return false; 4160 } 4161 } 4162 } 4163 4164 // Bailout intrinsic and do normal inlining if exception path is frequent. 4165 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 4166 return false; 4167 } 4168 4169 // Generate dynamic checks. 4170 // Class.cast() is java implementation of _checkcast bytecode. 4171 // Do checkcast (Parse::do_checkcast()) optimizations here. 4172 4173 mirror = null_check(mirror); 4174 // If mirror is dead, only null-path is taken. 4175 if (stopped()) { 4176 return true; 4177 } 4178 4179 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive). 4180 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT }; 4181 RegionNode* region = new RegionNode(PATH_LIMIT); 4182 record_for_igvn(region); 4183 4184 // Now load the mirror's klass metaobject, and null-check it. 4185 // If kls is null, we have a primitive mirror and 4186 // nothing is an instance of a primitive type. 4187 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 4188 4189 Node* res = top(); 4190 if (!stopped()) { 4191 Node* bad_type_ctrl = top(); 4192 // Do checkcast optimizations. 4193 res = gen_checkcast(obj, kls, &bad_type_ctrl); 4194 region->init_req(_bad_type_path, bad_type_ctrl); 4195 } 4196 if (region->in(_prim_path) != top() || 4197 region->in(_bad_type_path) != top()) { 4198 // Let Interpreter throw ClassCastException. 4199 PreserveJVMState pjvms(this); 4200 set_control(_gvn.transform(region)); 4201 uncommon_trap(Deoptimization::Reason_intrinsic, 4202 Deoptimization::Action_maybe_recompile); 4203 } 4204 if (!stopped()) { 4205 set_result(res); 4206 } 4207 return true; 4208 } 4209 4210 4211 //--------------------------inline_native_subtype_check------------------------ 4212 // This intrinsic takes the JNI calls out of the heart of 4213 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 4214 bool LibraryCallKit::inline_native_subtype_check() { 4215 // Pull both arguments off the stack. 4216 Node* args[2]; // two java.lang.Class mirrors: superc, subc 4217 args[0] = argument(0); 4218 args[1] = argument(1); 4219 Node* klasses[2]; // corresponding Klasses: superk, subk 4220 klasses[0] = klasses[1] = top(); 4221 4222 enum { 4223 // A full decision tree on {superc is prim, subc is prim}: 4224 _prim_0_path = 1, // {P,N} => false 4225 // {P,P} & superc!=subc => false 4226 _prim_same_path, // {P,P} & superc==subc => true 4227 _prim_1_path, // {N,P} => false 4228 _ref_subtype_path, // {N,N} & subtype check wins => true 4229 _both_ref_path, // {N,N} & subtype check loses => false 4230 PATH_LIMIT 4231 }; 4232 4233 RegionNode* region = new RegionNode(PATH_LIMIT); 4234 Node* phi = new PhiNode(region, TypeInt::BOOL); 4235 record_for_igvn(region); 4236 4237 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 4238 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL; 4239 int class_klass_offset = java_lang_Class::klass_offset(); 4240 4241 // First null-check both mirrors and load each mirror's klass metaobject. 4242 int which_arg; 4243 for (which_arg = 0; which_arg <= 1; which_arg++) { 4244 Node* arg = args[which_arg]; 4245 arg = null_check(arg); 4246 if (stopped()) break; 4247 args[which_arg] = arg; 4248 4249 Node* p = basic_plus_adr(arg, class_klass_offset); 4250 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type); 4251 klasses[which_arg] = _gvn.transform(kls); 4252 } 4253 4254 // Having loaded both klasses, test each for null. 4255 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 4256 for (which_arg = 0; which_arg <= 1; which_arg++) { 4257 Node* kls = klasses[which_arg]; 4258 Node* null_ctl = top(); 4259 kls = null_check_oop(kls, &null_ctl, never_see_null); 4260 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path); 4261 region->init_req(prim_path, null_ctl); 4262 if (stopped()) break; 4263 klasses[which_arg] = kls; 4264 } 4265 4266 if (!stopped()) { 4267 // now we have two reference types, in klasses[0..1] 4268 Node* subk = klasses[1]; // the argument to isAssignableFrom 4269 Node* superk = klasses[0]; // the receiver 4270 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 4271 // now we have a successful reference subtype check 4272 region->set_req(_ref_subtype_path, control()); 4273 } 4274 4275 // If both operands are primitive (both klasses null), then 4276 // we must return true when they are identical primitives. 4277 // It is convenient to test this after the first null klass check. 4278 set_control(region->in(_prim_0_path)); // go back to first null check 4279 if (!stopped()) { 4280 // Since superc is primitive, make a guard for the superc==subc case. 4281 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 4282 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 4283 generate_guard(bol_eq, region, PROB_FAIR); 4284 if (region->req() == PATH_LIMIT+1) { 4285 // A guard was added. If the added guard is taken, superc==subc. 4286 region->swap_edges(PATH_LIMIT, _prim_same_path); 4287 region->del_req(PATH_LIMIT); 4288 } 4289 region->set_req(_prim_0_path, control()); // Not equal after all. 4290 } 4291 4292 // these are the only paths that produce 'true': 4293 phi->set_req(_prim_same_path, intcon(1)); 4294 phi->set_req(_ref_subtype_path, intcon(1)); 4295 4296 // pull together the cases: 4297 assert(region->req() == PATH_LIMIT, "sane region"); 4298 for (uint i = 1; i < region->req(); i++) { 4299 Node* ctl = region->in(i); 4300 if (ctl == nullptr || ctl == top()) { 4301 region->set_req(i, top()); 4302 phi ->set_req(i, top()); 4303 } else if (phi->in(i) == nullptr) { 4304 phi->set_req(i, intcon(0)); // all other paths produce 'false' 4305 } 4306 } 4307 4308 set_control(_gvn.transform(region)); 4309 set_result(_gvn.transform(phi)); 4310 return true; 4311 } 4312 4313 //---------------------generate_array_guard_common------------------------ 4314 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, 4315 bool obj_array, bool not_array, Node** obj) { 4316 4317 if (stopped()) { 4318 return nullptr; 4319 } 4320 4321 // If obj_array/non_array==false/false: 4322 // Branch around if the given klass is in fact an array (either obj or prim). 4323 // If obj_array/non_array==false/true: 4324 // Branch around if the given klass is not an array klass of any kind. 4325 // If obj_array/non_array==true/true: 4326 // Branch around if the kls is not an oop array (kls is int[], String, etc.) 4327 // If obj_array/non_array==true/false: 4328 // Branch around if the kls is an oop array (Object[] or subtype) 4329 // 4330 // Like generate_guard, adds a new path onto the region. 4331 jint layout_con = 0; 4332 Node* layout_val = get_layout_helper(kls, layout_con); 4333 if (layout_val == nullptr) { 4334 bool query = (obj_array 4335 ? Klass::layout_helper_is_objArray(layout_con) 4336 : Klass::layout_helper_is_array(layout_con)); 4337 if (query == not_array) { 4338 return nullptr; // never a branch 4339 } else { // always a branch 4340 Node* always_branch = control(); 4341 if (region != nullptr) 4342 region->add_req(always_branch); 4343 set_control(top()); 4344 return always_branch; 4345 } 4346 } 4347 // Now test the correct condition. 4348 jint nval = (obj_array 4349 ? (jint)(Klass::_lh_array_tag_type_value 4350 << Klass::_lh_array_tag_shift) 4351 : Klass::_lh_neutral_value); 4352 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 4353 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array 4354 // invert the test if we are looking for a non-array 4355 if (not_array) btest = BoolTest(btest).negate(); 4356 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 4357 Node* ctrl = generate_fair_guard(bol, region); 4358 Node* is_array_ctrl = not_array ? control() : ctrl; 4359 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) { 4360 // Keep track of the fact that 'obj' is an array to prevent 4361 // array specific accesses from floating above the guard. 4362 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM)); 4363 } 4364 return ctrl; 4365 } 4366 4367 4368 //-----------------------inline_native_newArray-------------------------- 4369 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length); 4370 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size); 4371 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) { 4372 Node* mirror; 4373 Node* count_val; 4374 if (uninitialized) { 4375 null_check_receiver(); 4376 mirror = argument(1); 4377 count_val = argument(2); 4378 } else { 4379 mirror = argument(0); 4380 count_val = argument(1); 4381 } 4382 4383 mirror = null_check(mirror); 4384 // If mirror or obj is dead, only null-path is taken. 4385 if (stopped()) return true; 4386 4387 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 4388 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4389 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4390 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4391 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4392 4393 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 4394 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 4395 result_reg, _slow_path); 4396 Node* normal_ctl = control(); 4397 Node* no_array_ctl = result_reg->in(_slow_path); 4398 4399 // Generate code for the slow case. We make a call to newArray(). 4400 set_control(no_array_ctl); 4401 if (!stopped()) { 4402 // Either the input type is void.class, or else the 4403 // array klass has not yet been cached. Either the 4404 // ensuing call will throw an exception, or else it 4405 // will cache the array klass for next time. 4406 PreserveJVMState pjvms(this); 4407 CallJavaNode* slow_call = nullptr; 4408 if (uninitialized) { 4409 // Generate optimized virtual call (holder class 'Unsafe' is final) 4410 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true); 4411 } else { 4412 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true); 4413 } 4414 Node* slow_result = set_results_for_java_call(slow_call); 4415 // this->control() comes from set_results_for_java_call 4416 result_reg->set_req(_slow_path, control()); 4417 result_val->set_req(_slow_path, slow_result); 4418 result_io ->set_req(_slow_path, i_o()); 4419 result_mem->set_req(_slow_path, reset_memory()); 4420 } 4421 4422 set_control(normal_ctl); 4423 if (!stopped()) { 4424 // Normal case: The array type has been cached in the java.lang.Class. 4425 // The following call works fine even if the array type is polymorphic. 4426 // It could be a dynamic mix of int[], boolean[], Object[], etc. 4427 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push 4428 result_reg->init_req(_normal_path, control()); 4429 result_val->init_req(_normal_path, obj); 4430 result_io ->init_req(_normal_path, i_o()); 4431 result_mem->init_req(_normal_path, reset_memory()); 4432 4433 if (uninitialized) { 4434 // Mark the allocation so that zeroing is skipped 4435 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj); 4436 alloc->maybe_set_complete(&_gvn); 4437 } 4438 } 4439 4440 // Return the combined state. 4441 set_i_o( _gvn.transform(result_io) ); 4442 set_all_memory( _gvn.transform(result_mem)); 4443 4444 C->set_has_split_ifs(true); // Has chance for split-if optimization 4445 set_result(result_reg, result_val); 4446 return true; 4447 } 4448 4449 //----------------------inline_native_getLength-------------------------- 4450 // public static native int java.lang.reflect.Array.getLength(Object array); 4451 bool LibraryCallKit::inline_native_getLength() { 4452 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 4453 4454 Node* array = null_check(argument(0)); 4455 // If array is dead, only null-path is taken. 4456 if (stopped()) return true; 4457 4458 // Deoptimize if it is a non-array. 4459 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array); 4460 4461 if (non_array != nullptr) { 4462 PreserveJVMState pjvms(this); 4463 set_control(non_array); 4464 uncommon_trap(Deoptimization::Reason_intrinsic, 4465 Deoptimization::Action_maybe_recompile); 4466 } 4467 4468 // If control is dead, only non-array-path is taken. 4469 if (stopped()) return true; 4470 4471 // The works fine even if the array type is polymorphic. 4472 // It could be a dynamic mix of int[], boolean[], Object[], etc. 4473 Node* result = load_array_length(array); 4474 4475 C->set_has_split_ifs(true); // Has chance for split-if optimization 4476 set_result(result); 4477 return true; 4478 } 4479 4480 //------------------------inline_array_copyOf---------------------------- 4481 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 4482 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 4483 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 4484 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 4485 4486 // Get the arguments. 4487 Node* original = argument(0); 4488 Node* start = is_copyOfRange? argument(1): intcon(0); 4489 Node* end = is_copyOfRange? argument(2): argument(1); 4490 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 4491 4492 Node* newcopy = nullptr; 4493 4494 // Set the original stack and the reexecute bit for the interpreter to reexecute 4495 // the bytecode that invokes Arrays.copyOf if deoptimization happens. 4496 { PreserveReexecuteState preexecs(this); 4497 jvms()->set_should_reexecute(true); 4498 4499 array_type_mirror = null_check(array_type_mirror); 4500 original = null_check(original); 4501 4502 // Check if a null path was taken unconditionally. 4503 if (stopped()) return true; 4504 4505 Node* orig_length = load_array_length(original); 4506 4507 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0); 4508 klass_node = null_check(klass_node); 4509 4510 RegionNode* bailout = new RegionNode(1); 4511 record_for_igvn(bailout); 4512 4513 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc. 4514 // Bail out if that is so. 4515 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout); 4516 if (not_objArray != nullptr) { 4517 // Improve the klass node's type from the new optimistic assumption: 4518 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 4519 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/); 4520 Node* cast = new CastPPNode(control(), klass_node, akls); 4521 klass_node = _gvn.transform(cast); 4522 } 4523 4524 // Bail out if either start or end is negative. 4525 generate_negative_guard(start, bailout, &start); 4526 generate_negative_guard(end, bailout, &end); 4527 4528 Node* length = end; 4529 if (_gvn.type(start) != TypeInt::ZERO) { 4530 length = _gvn.transform(new SubINode(end, start)); 4531 } 4532 4533 // Bail out if length is negative (i.e., if start > end). 4534 // Without this the new_array would throw 4535 // NegativeArraySizeException but IllegalArgumentException is what 4536 // should be thrown 4537 generate_negative_guard(length, bailout, &length); 4538 4539 // Bail out if start is larger than the original length 4540 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 4541 generate_negative_guard(orig_tail, bailout, &orig_tail); 4542 4543 if (bailout->req() > 1) { 4544 PreserveJVMState pjvms(this); 4545 set_control(_gvn.transform(bailout)); 4546 uncommon_trap(Deoptimization::Reason_intrinsic, 4547 Deoptimization::Action_maybe_recompile); 4548 } 4549 4550 if (!stopped()) { 4551 // How many elements will we copy from the original? 4552 // The answer is MinI(orig_tail, length). 4553 Node* moved = _gvn.transform(new MinINode(orig_tail, length)); 4554 4555 // Generate a direct call to the right arraycopy function(s). 4556 // We know the copy is disjoint but we might not know if the 4557 // oop stores need checking. 4558 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 4559 // This will fail a store-check if x contains any non-nulls. 4560 4561 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 4562 // loads/stores but it is legal only if we're sure the 4563 // Arrays.copyOf would succeed. So we need all input arguments 4564 // to the copyOf to be validated, including that the copy to the 4565 // new array won't trigger an ArrayStoreException. That subtype 4566 // check can be optimized if we know something on the type of 4567 // the input array from type speculation. 4568 if (_gvn.type(klass_node)->singleton()) { 4569 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr(); 4570 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr(); 4571 4572 int test = C->static_subtype_check(superk, subk); 4573 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 4574 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 4575 if (t_original->speculative_type() != nullptr) { 4576 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 4577 } 4578 } 4579 } 4580 4581 bool validated = false; 4582 // Reason_class_check rather than Reason_intrinsic because we 4583 // want to intrinsify even if this traps. 4584 if (!too_many_traps(Deoptimization::Reason_class_check)) { 4585 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node); 4586 4587 if (not_subtype_ctrl != top()) { 4588 PreserveJVMState pjvms(this); 4589 set_control(not_subtype_ctrl); 4590 uncommon_trap(Deoptimization::Reason_class_check, 4591 Deoptimization::Action_make_not_entrant); 4592 assert(stopped(), "Should be stopped"); 4593 } 4594 validated = true; 4595 } 4596 4597 if (!stopped()) { 4598 newcopy = new_array(klass_node, length, 0); // no arguments to push 4599 4600 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true, 4601 load_object_klass(original), klass_node); 4602 if (!is_copyOfRange) { 4603 ac->set_copyof(validated); 4604 } else { 4605 ac->set_copyofrange(validated); 4606 } 4607 Node* n = _gvn.transform(ac); 4608 if (n == ac) { 4609 ac->connect_outputs(this); 4610 } else { 4611 assert(validated, "shouldn't transform if all arguments not validated"); 4612 set_all_memory(n); 4613 } 4614 } 4615 } 4616 } // original reexecute is set back here 4617 4618 C->set_has_split_ifs(true); // Has chance for split-if optimization 4619 if (!stopped()) { 4620 set_result(newcopy); 4621 } 4622 return true; 4623 } 4624 4625 4626 //----------------------generate_virtual_guard--------------------------- 4627 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 4628 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 4629 RegionNode* slow_region) { 4630 ciMethod* method = callee(); 4631 int vtable_index = method->vtable_index(); 4632 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4633 "bad index %d", vtable_index); 4634 // Get the Method* out of the appropriate vtable entry. 4635 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 4636 vtable_index*vtableEntry::size_in_bytes() + 4637 in_bytes(vtableEntry::method_offset()); 4638 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 4639 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 4640 4641 // Compare the target method with the expected method (e.g., Object.hashCode). 4642 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 4643 4644 Node* native_call = makecon(native_call_addr); 4645 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 4646 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 4647 4648 return generate_slow_guard(test_native, slow_region); 4649 } 4650 4651 //-----------------------generate_method_call---------------------------- 4652 // Use generate_method_call to make a slow-call to the real 4653 // method if the fast path fails. An alternative would be to 4654 // use a stub like OptoRuntime::slow_arraycopy_Java. 4655 // This only works for expanding the current library call, 4656 // not another intrinsic. (E.g., don't use this for making an 4657 // arraycopy call inside of the copyOf intrinsic.) 4658 CallJavaNode* 4659 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) { 4660 // When compiling the intrinsic method itself, do not use this technique. 4661 guarantee(callee() != C->method(), "cannot make slow-call to self"); 4662 4663 ciMethod* method = callee(); 4664 // ensure the JVMS we have will be correct for this call 4665 guarantee(method_id == method->intrinsic_id(), "must match"); 4666 4667 const TypeFunc* tf = TypeFunc::make(method); 4668 if (res_not_null) { 4669 assert(tf->return_type() == T_OBJECT, ""); 4670 const TypeTuple* range = tf->range(); 4671 const Type** fields = TypeTuple::fields(range->cnt()); 4672 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL); 4673 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields); 4674 tf = TypeFunc::make(tf->domain(), new_range); 4675 } 4676 CallJavaNode* slow_call; 4677 if (is_static) { 4678 assert(!is_virtual, ""); 4679 slow_call = new CallStaticJavaNode(C, tf, 4680 SharedRuntime::get_resolve_static_call_stub(), method); 4681 } else if (is_virtual) { 4682 assert(!gvn().type(argument(0))->maybe_null(), "should not be null"); 4683 int vtable_index = Method::invalid_vtable_index; 4684 if (UseInlineCaches) { 4685 // Suppress the vtable call 4686 } else { 4687 // hashCode and clone are not a miranda methods, 4688 // so the vtable index is fixed. 4689 // No need to use the linkResolver to get it. 4690 vtable_index = method->vtable_index(); 4691 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4692 "bad index %d", vtable_index); 4693 } 4694 slow_call = new CallDynamicJavaNode(tf, 4695 SharedRuntime::get_resolve_virtual_call_stub(), 4696 method, vtable_index); 4697 } else { // neither virtual nor static: opt_virtual 4698 assert(!gvn().type(argument(0))->maybe_null(), "should not be null"); 4699 slow_call = new CallStaticJavaNode(C, tf, 4700 SharedRuntime::get_resolve_opt_virtual_call_stub(), method); 4701 slow_call->set_optimized_virtual(true); 4702 } 4703 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) { 4704 // To be able to issue a direct call (optimized virtual or virtual) 4705 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information 4706 // about the method being invoked should be attached to the call site to 4707 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C). 4708 slow_call->set_override_symbolic_info(true); 4709 } 4710 set_arguments_for_java_call(slow_call); 4711 set_edges_for_java_call(slow_call); 4712 return slow_call; 4713 } 4714 4715 4716 /** 4717 * Build special case code for calls to hashCode on an object. This call may 4718 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 4719 * slightly different code. 4720 */ 4721 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 4722 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 4723 assert(!(is_virtual && is_static), "either virtual, special, or static"); 4724 4725 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 4726 4727 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4728 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 4729 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4730 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4731 Node* obj = nullptr; 4732 if (!is_static) { 4733 // Check for hashing null object 4734 obj = null_check_receiver(); 4735 if (stopped()) return true; // unconditionally null 4736 result_reg->init_req(_null_path, top()); 4737 result_val->init_req(_null_path, top()); 4738 } else { 4739 // Do a null check, and return zero if null. 4740 // System.identityHashCode(null) == 0 4741 obj = argument(0); 4742 Node* null_ctl = top(); 4743 obj = null_check_oop(obj, &null_ctl); 4744 result_reg->init_req(_null_path, null_ctl); 4745 result_val->init_req(_null_path, _gvn.intcon(0)); 4746 } 4747 4748 // Unconditionally null? Then return right away. 4749 if (stopped()) { 4750 set_control( result_reg->in(_null_path)); 4751 if (!stopped()) 4752 set_result(result_val->in(_null_path)); 4753 return true; 4754 } 4755 4756 // We only go to the fast case code if we pass a number of guards. The 4757 // paths which do not pass are accumulated in the slow_region. 4758 RegionNode* slow_region = new RegionNode(1); 4759 record_for_igvn(slow_region); 4760 4761 // If this is a virtual call, we generate a funny guard. We pull out 4762 // the vtable entry corresponding to hashCode() from the target object. 4763 // If the target method which we are calling happens to be the native 4764 // Object hashCode() method, we pass the guard. We do not need this 4765 // guard for non-virtual calls -- the caller is known to be the native 4766 // Object hashCode(). 4767 if (is_virtual) { 4768 // After null check, get the object's klass. 4769 Node* obj_klass = load_object_klass(obj); 4770 generate_virtual_guard(obj_klass, slow_region); 4771 } 4772 4773 // Get the header out of the object, use LoadMarkNode when available 4774 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 4775 // The control of the load must be null. Otherwise, the load can move before 4776 // the null check after castPP removal. 4777 Node* no_ctrl = nullptr; 4778 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 4779 4780 if (!UseObjectMonitorTable) { 4781 // Test the header to see if it is safe to read w.r.t. locking. 4782 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place); 4783 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 4784 if (LockingMode == LM_LIGHTWEIGHT) { 4785 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value); 4786 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val)); 4787 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq)); 4788 4789 generate_slow_guard(test_monitor, slow_region); 4790 } else { 4791 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value); 4792 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val)); 4793 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne)); 4794 4795 generate_slow_guard(test_not_unlocked, slow_region); 4796 } 4797 } 4798 4799 // Get the hash value and check to see that it has been properly assigned. 4800 // We depend on hash_mask being at most 32 bits and avoid the use of 4801 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 4802 // vm: see markWord.hpp. 4803 Node *hash_mask = _gvn.intcon(markWord::hash_mask); 4804 Node *hash_shift = _gvn.intcon(markWord::hash_shift); 4805 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 4806 // This hack lets the hash bits live anywhere in the mark object now, as long 4807 // as the shift drops the relevant bits into the low 32 bits. Note that 4808 // Java spec says that HashCode is an int so there's no point in capturing 4809 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 4810 hshifted_header = ConvX2I(hshifted_header); 4811 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 4812 4813 Node *no_hash_val = _gvn.intcon(markWord::no_hash); 4814 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 4815 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 4816 4817 generate_slow_guard(test_assigned, slow_region); 4818 4819 Node* init_mem = reset_memory(); 4820 // fill in the rest of the null path: 4821 result_io ->init_req(_null_path, i_o()); 4822 result_mem->init_req(_null_path, init_mem); 4823 4824 result_val->init_req(_fast_path, hash_val); 4825 result_reg->init_req(_fast_path, control()); 4826 result_io ->init_req(_fast_path, i_o()); 4827 result_mem->init_req(_fast_path, init_mem); 4828 4829 // Generate code for the slow case. We make a call to hashCode(). 4830 set_control(_gvn.transform(slow_region)); 4831 if (!stopped()) { 4832 // No need for PreserveJVMState, because we're using up the present state. 4833 set_all_memory(init_mem); 4834 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 4835 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false); 4836 Node* slow_result = set_results_for_java_call(slow_call); 4837 // this->control() comes from set_results_for_java_call 4838 result_reg->init_req(_slow_path, control()); 4839 result_val->init_req(_slow_path, slow_result); 4840 result_io ->set_req(_slow_path, i_o()); 4841 result_mem ->set_req(_slow_path, reset_memory()); 4842 } 4843 4844 // Return the combined state. 4845 set_i_o( _gvn.transform(result_io) ); 4846 set_all_memory( _gvn.transform(result_mem)); 4847 4848 set_result(result_reg, result_val); 4849 return true; 4850 } 4851 4852 //---------------------------inline_native_getClass---------------------------- 4853 // public final native Class<?> java.lang.Object.getClass(); 4854 // 4855 // Build special case code for calls to getClass on an object. 4856 bool LibraryCallKit::inline_native_getClass() { 4857 Node* obj = null_check_receiver(); 4858 if (stopped()) return true; 4859 set_result(load_mirror_from_klass(load_object_klass(obj))); 4860 return true; 4861 } 4862 4863 //-----------------inline_native_Reflection_getCallerClass--------------------- 4864 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 4865 // 4866 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 4867 // 4868 // NOTE: This code must perform the same logic as JVM_GetCallerClass 4869 // in that it must skip particular security frames and checks for 4870 // caller sensitive methods. 4871 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 4872 #ifndef PRODUCT 4873 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4874 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 4875 } 4876 #endif 4877 4878 if (!jvms()->has_method()) { 4879 #ifndef PRODUCT 4880 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4881 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 4882 } 4883 #endif 4884 return false; 4885 } 4886 4887 // Walk back up the JVM state to find the caller at the required 4888 // depth. 4889 JVMState* caller_jvms = jvms(); 4890 4891 // Cf. JVM_GetCallerClass 4892 // NOTE: Start the loop at depth 1 because the current JVM state does 4893 // not include the Reflection.getCallerClass() frame. 4894 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) { 4895 ciMethod* m = caller_jvms->method(); 4896 switch (n) { 4897 case 0: 4898 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 4899 break; 4900 case 1: 4901 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 4902 if (!m->caller_sensitive()) { 4903 #ifndef PRODUCT 4904 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4905 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 4906 } 4907 #endif 4908 return false; // bail-out; let JVM_GetCallerClass do the work 4909 } 4910 break; 4911 default: 4912 if (!m->is_ignored_by_security_stack_walk()) { 4913 // We have reached the desired frame; return the holder class. 4914 // Acquire method holder as java.lang.Class and push as constant. 4915 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 4916 ciInstance* caller_mirror = caller_klass->java_mirror(); 4917 set_result(makecon(TypeInstPtr::make(caller_mirror))); 4918 4919 #ifndef PRODUCT 4920 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4921 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()); 4922 tty->print_cr(" JVM state at this point:"); 4923 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4924 ciMethod* m = jvms()->of_depth(i)->method(); 4925 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4926 } 4927 } 4928 #endif 4929 return true; 4930 } 4931 break; 4932 } 4933 } 4934 4935 #ifndef PRODUCT 4936 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4937 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 4938 tty->print_cr(" JVM state at this point:"); 4939 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4940 ciMethod* m = jvms()->of_depth(i)->method(); 4941 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4942 } 4943 } 4944 #endif 4945 4946 return false; // bail-out; let JVM_GetCallerClass do the work 4947 } 4948 4949 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 4950 Node* arg = argument(0); 4951 Node* result = nullptr; 4952 4953 switch (id) { 4954 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 4955 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 4956 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 4957 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 4958 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break; 4959 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break; 4960 4961 case vmIntrinsics::_doubleToLongBits: { 4962 // two paths (plus control) merge in a wood 4963 RegionNode *r = new RegionNode(3); 4964 Node *phi = new PhiNode(r, TypeLong::LONG); 4965 4966 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 4967 // Build the boolean node 4968 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4969 4970 // Branch either way. 4971 // NaN case is less traveled, which makes all the difference. 4972 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4973 Node *opt_isnan = _gvn.transform(ifisnan); 4974 assert( opt_isnan->is_If(), "Expect an IfNode"); 4975 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4976 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4977 4978 set_control(iftrue); 4979 4980 static const jlong nan_bits = CONST64(0x7ff8000000000000); 4981 Node *slow_result = longcon(nan_bits); // return NaN 4982 phi->init_req(1, _gvn.transform( slow_result )); 4983 r->init_req(1, iftrue); 4984 4985 // Else fall through 4986 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4987 set_control(iffalse); 4988 4989 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 4990 r->init_req(2, iffalse); 4991 4992 // Post merge 4993 set_control(_gvn.transform(r)); 4994 record_for_igvn(r); 4995 4996 C->set_has_split_ifs(true); // Has chance for split-if optimization 4997 result = phi; 4998 assert(result->bottom_type()->isa_long(), "must be"); 4999 break; 5000 } 5001 5002 case vmIntrinsics::_floatToIntBits: { 5003 // two paths (plus control) merge in a wood 5004 RegionNode *r = new RegionNode(3); 5005 Node *phi = new PhiNode(r, TypeInt::INT); 5006 5007 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 5008 // Build the boolean node 5009 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 5010 5011 // Branch either way. 5012 // NaN case is less traveled, which makes all the difference. 5013 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 5014 Node *opt_isnan = _gvn.transform(ifisnan); 5015 assert( opt_isnan->is_If(), "Expect an IfNode"); 5016 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 5017 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 5018 5019 set_control(iftrue); 5020 5021 static const jint nan_bits = 0x7fc00000; 5022 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 5023 phi->init_req(1, _gvn.transform( slow_result )); 5024 r->init_req(1, iftrue); 5025 5026 // Else fall through 5027 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 5028 set_control(iffalse); 5029 5030 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 5031 r->init_req(2, iffalse); 5032 5033 // Post merge 5034 set_control(_gvn.transform(r)); 5035 record_for_igvn(r); 5036 5037 C->set_has_split_ifs(true); // Has chance for split-if optimization 5038 result = phi; 5039 assert(result->bottom_type()->isa_int(), "must be"); 5040 break; 5041 } 5042 5043 default: 5044 fatal_unexpected_iid(id); 5045 break; 5046 } 5047 set_result(_gvn.transform(result)); 5048 return true; 5049 } 5050 5051 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) { 5052 Node* arg = argument(0); 5053 Node* result = nullptr; 5054 5055 switch (id) { 5056 case vmIntrinsics::_floatIsInfinite: 5057 result = new IsInfiniteFNode(arg); 5058 break; 5059 case vmIntrinsics::_floatIsFinite: 5060 result = new IsFiniteFNode(arg); 5061 break; 5062 case vmIntrinsics::_doubleIsInfinite: 5063 result = new IsInfiniteDNode(arg); 5064 break; 5065 case vmIntrinsics::_doubleIsFinite: 5066 result = new IsFiniteDNode(arg); 5067 break; 5068 default: 5069 fatal_unexpected_iid(id); 5070 break; 5071 } 5072 set_result(_gvn.transform(result)); 5073 return true; 5074 } 5075 5076 //----------------------inline_unsafe_copyMemory------------------------- 5077 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 5078 5079 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) { 5080 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr(); 5081 const Type* base_t = gvn.type(base); 5082 5083 bool in_native = (base_t == TypePtr::NULL_PTR); 5084 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t); 5085 bool is_mixed = !in_heap && !in_native; 5086 5087 if (is_mixed) { 5088 return true; // mixed accesses can touch both on-heap and off-heap memory 5089 } 5090 if (in_heap) { 5091 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM); 5092 if (!is_prim_array) { 5093 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array, 5094 // there's not enough type information available to determine proper memory slice for it. 5095 return true; 5096 } 5097 } 5098 return false; 5099 } 5100 5101 bool LibraryCallKit::inline_unsafe_copyMemory() { 5102 if (callee()->is_static()) return false; // caller must have the capability! 5103 null_check_receiver(); // null-check receiver 5104 if (stopped()) return true; 5105 5106 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 5107 5108 Node* src_base = argument(1); // type: oop 5109 Node* src_off = ConvL2X(argument(2)); // type: long 5110 Node* dst_base = argument(4); // type: oop 5111 Node* dst_off = ConvL2X(argument(5)); // type: long 5112 Node* size = ConvL2X(argument(7)); // type: long 5113 5114 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 5115 "fieldOffset must be byte-scaled"); 5116 5117 Node* src_addr = make_unsafe_address(src_base, src_off); 5118 Node* dst_addr = make_unsafe_address(dst_base, dst_off); 5119 5120 Node* thread = _gvn.transform(new ThreadLocalNode()); 5121 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 5122 BasicType doing_unsafe_access_bt = T_BYTE; 5123 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 5124 5125 // update volatile field 5126 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered); 5127 5128 int flags = RC_LEAF | RC_NO_FP; 5129 5130 const TypePtr* dst_type = TypePtr::BOTTOM; 5131 5132 // Adjust memory effects of the runtime call based on input values. 5133 if (!has_wide_mem(_gvn, src_addr, src_base) && 5134 !has_wide_mem(_gvn, dst_addr, dst_base)) { 5135 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory 5136 5137 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr(); 5138 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) { 5139 flags |= RC_NARROW_MEM; // narrow in memory 5140 } 5141 } 5142 5143 // Call it. Note that the length argument is not scaled. 5144 make_runtime_call(flags, 5145 OptoRuntime::fast_arraycopy_Type(), 5146 StubRoutines::unsafe_arraycopy(), 5147 "unsafe_arraycopy", 5148 dst_type, 5149 src_addr, dst_addr, size XTOP); 5150 5151 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered); 5152 5153 return true; 5154 } 5155 5156 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value); 5157 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value' 5158 bool LibraryCallKit::inline_unsafe_setMemory() { 5159 if (callee()->is_static()) return false; // caller must have the capability! 5160 null_check_receiver(); // null-check receiver 5161 if (stopped()) return true; 5162 5163 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 5164 5165 Node* dst_base = argument(1); // type: oop 5166 Node* dst_off = ConvL2X(argument(2)); // type: long 5167 Node* size = ConvL2X(argument(4)); // type: long 5168 Node* byte = argument(6); // type: byte 5169 5170 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 5171 "fieldOffset must be byte-scaled"); 5172 5173 Node* dst_addr = make_unsafe_address(dst_base, dst_off); 5174 5175 Node* thread = _gvn.transform(new ThreadLocalNode()); 5176 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 5177 BasicType doing_unsafe_access_bt = T_BYTE; 5178 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 5179 5180 // update volatile field 5181 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered); 5182 5183 int flags = RC_LEAF | RC_NO_FP; 5184 5185 const TypePtr* dst_type = TypePtr::BOTTOM; 5186 5187 // Adjust memory effects of the runtime call based on input values. 5188 if (!has_wide_mem(_gvn, dst_addr, dst_base)) { 5189 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory 5190 5191 flags |= RC_NARROW_MEM; // narrow in memory 5192 } 5193 5194 // Call it. Note that the length argument is not scaled. 5195 make_runtime_call(flags, 5196 OptoRuntime::unsafe_setmemory_Type(), 5197 StubRoutines::unsafe_setmemory(), 5198 "unsafe_setmemory", 5199 dst_type, 5200 dst_addr, size XTOP, byte); 5201 5202 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered); 5203 5204 return true; 5205 } 5206 5207 #undef XTOP 5208 5209 //------------------------clone_coping----------------------------------- 5210 // Helper function for inline_native_clone. 5211 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) { 5212 assert(obj_size != nullptr, ""); 5213 Node* raw_obj = alloc_obj->in(1); 5214 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 5215 5216 AllocateNode* alloc = nullptr; 5217 if (ReduceBulkZeroing && 5218 // If we are implementing an array clone without knowing its source type 5219 // (can happen when compiling the array-guarded branch of a reflective 5220 // Object.clone() invocation), initialize the array within the allocation. 5221 // This is needed because some GCs (e.g. ZGC) might fall back in this case 5222 // to a runtime clone call that assumes fully initialized source arrays. 5223 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) { 5224 // We will be completely responsible for initializing this object - 5225 // mark Initialize node as complete. 5226 alloc = AllocateNode::Ideal_allocation(alloc_obj); 5227 // The object was just allocated - there should be no any stores! 5228 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), ""); 5229 // Mark as complete_with_arraycopy so that on AllocateNode 5230 // expansion, we know this AllocateNode is initialized by an array 5231 // copy and a StoreStore barrier exists after the array copy. 5232 alloc->initialization()->set_complete_with_arraycopy(); 5233 } 5234 5235 Node* size = _gvn.transform(obj_size); 5236 access_clone(obj, alloc_obj, size, is_array); 5237 5238 // Do not let reads from the cloned object float above the arraycopy. 5239 if (alloc != nullptr) { 5240 // Do not let stores that initialize this object be reordered with 5241 // a subsequent store that would make this object accessible by 5242 // other threads. 5243 // Record what AllocateNode this StoreStore protects so that 5244 // escape analysis can go from the MemBarStoreStoreNode to the 5245 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 5246 // based on the escape status of the AllocateNode. 5247 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 5248 } else { 5249 insert_mem_bar(Op_MemBarCPUOrder); 5250 } 5251 } 5252 5253 //------------------------inline_native_clone---------------------------- 5254 // protected native Object java.lang.Object.clone(); 5255 // 5256 // Here are the simple edge cases: 5257 // null receiver => normal trap 5258 // virtual and clone was overridden => slow path to out-of-line clone 5259 // not cloneable or finalizer => slow path to out-of-line Object.clone 5260 // 5261 // The general case has two steps, allocation and copying. 5262 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 5263 // 5264 // Copying also has two cases, oop arrays and everything else. 5265 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 5266 // Everything else uses the tight inline loop supplied by CopyArrayNode. 5267 // 5268 // These steps fold up nicely if and when the cloned object's klass 5269 // can be sharply typed as an object array, a type array, or an instance. 5270 // 5271 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 5272 PhiNode* result_val; 5273 5274 // Set the reexecute bit for the interpreter to reexecute 5275 // the bytecode that invokes Object.clone if deoptimization happens. 5276 { PreserveReexecuteState preexecs(this); 5277 jvms()->set_should_reexecute(true); 5278 5279 Node* obj = null_check_receiver(); 5280 if (stopped()) return true; 5281 5282 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 5283 5284 // If we are going to clone an instance, we need its exact type to 5285 // know the number and types of fields to convert the clone to 5286 // loads/stores. Maybe a speculative type can help us. 5287 if (!obj_type->klass_is_exact() && 5288 obj_type->speculative_type() != nullptr && 5289 obj_type->speculative_type()->is_instance_klass()) { 5290 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 5291 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 5292 !spec_ik->has_injected_fields()) { 5293 if (!obj_type->isa_instptr() || 5294 obj_type->is_instptr()->instance_klass()->has_subklass()) { 5295 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 5296 } 5297 } 5298 } 5299 5300 // Conservatively insert a memory barrier on all memory slices. 5301 // Do not let writes into the original float below the clone. 5302 insert_mem_bar(Op_MemBarCPUOrder); 5303 5304 // paths into result_reg: 5305 enum { 5306 _slow_path = 1, // out-of-line call to clone method (virtual or not) 5307 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 5308 _array_path, // plain array allocation, plus arrayof_long_arraycopy 5309 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 5310 PATH_LIMIT 5311 }; 5312 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 5313 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 5314 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 5315 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 5316 record_for_igvn(result_reg); 5317 5318 Node* obj_klass = load_object_klass(obj); 5319 Node* array_obj = obj; 5320 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj); 5321 if (array_ctl != nullptr) { 5322 // It's an array. 5323 PreserveJVMState pjvms(this); 5324 set_control(array_ctl); 5325 Node* obj_length = load_array_length(array_obj); 5326 Node* array_size = nullptr; // Size of the array without object alignment padding. 5327 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true); 5328 5329 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 5330 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) { 5331 // If it is an oop array, it requires very special treatment, 5332 // because gc barriers are required when accessing the array. 5333 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr); 5334 if (is_obja != nullptr) { 5335 PreserveJVMState pjvms2(this); 5336 set_control(is_obja); 5337 // Generate a direct call to the right arraycopy function(s). 5338 // Clones are always tightly coupled. 5339 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false); 5340 ac->set_clone_oop_array(); 5341 Node* n = _gvn.transform(ac); 5342 assert(n == ac, "cannot disappear"); 5343 ac->connect_outputs(this, /*deoptimize_on_exception=*/true); 5344 5345 result_reg->init_req(_objArray_path, control()); 5346 result_val->init_req(_objArray_path, alloc_obj); 5347 result_i_o ->set_req(_objArray_path, i_o()); 5348 result_mem ->set_req(_objArray_path, reset_memory()); 5349 } 5350 } 5351 // Otherwise, there are no barriers to worry about. 5352 // (We can dispense with card marks if we know the allocation 5353 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 5354 // causes the non-eden paths to take compensating steps to 5355 // simulate a fresh allocation, so that no further 5356 // card marks are required in compiled code to initialize 5357 // the object.) 5358 5359 if (!stopped()) { 5360 copy_to_clone(array_obj, alloc_obj, array_size, true); 5361 5362 // Present the results of the copy. 5363 result_reg->init_req(_array_path, control()); 5364 result_val->init_req(_array_path, alloc_obj); 5365 result_i_o ->set_req(_array_path, i_o()); 5366 result_mem ->set_req(_array_path, reset_memory()); 5367 } 5368 } 5369 5370 // We only go to the instance fast case code if we pass a number of guards. 5371 // The paths which do not pass are accumulated in the slow_region. 5372 RegionNode* slow_region = new RegionNode(1); 5373 record_for_igvn(slow_region); 5374 if (!stopped()) { 5375 // It's an instance (we did array above). Make the slow-path tests. 5376 // If this is a virtual call, we generate a funny guard. We grab 5377 // the vtable entry corresponding to clone() from the target object. 5378 // If the target method which we are calling happens to be the 5379 // Object clone() method, we pass the guard. We do not need this 5380 // guard for non-virtual calls; the caller is known to be the native 5381 // Object clone(). 5382 if (is_virtual) { 5383 generate_virtual_guard(obj_klass, slow_region); 5384 } 5385 5386 // The object must be easily cloneable and must not have a finalizer. 5387 // Both of these conditions may be checked in a single test. 5388 // We could optimize the test further, but we don't care. 5389 generate_misc_flags_guard(obj_klass, 5390 // Test both conditions: 5391 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer, 5392 // Must be cloneable but not finalizer: 5393 KlassFlags::_misc_is_cloneable_fast, 5394 slow_region); 5395 } 5396 5397 if (!stopped()) { 5398 // It's an instance, and it passed the slow-path tests. 5399 PreserveJVMState pjvms(this); 5400 Node* obj_size = nullptr; // Total object size, including object alignment padding. 5401 // Need to deoptimize on exception from allocation since Object.clone intrinsic 5402 // is reexecuted if deoptimization occurs and there could be problems when merging 5403 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 5404 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true); 5405 5406 copy_to_clone(obj, alloc_obj, obj_size, false); 5407 5408 // Present the results of the slow call. 5409 result_reg->init_req(_instance_path, control()); 5410 result_val->init_req(_instance_path, alloc_obj); 5411 result_i_o ->set_req(_instance_path, i_o()); 5412 result_mem ->set_req(_instance_path, reset_memory()); 5413 } 5414 5415 // Generate code for the slow case. We make a call to clone(). 5416 set_control(_gvn.transform(slow_region)); 5417 if (!stopped()) { 5418 PreserveJVMState pjvms(this); 5419 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true); 5420 // We need to deoptimize on exception (see comment above) 5421 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true); 5422 // this->control() comes from set_results_for_java_call 5423 result_reg->init_req(_slow_path, control()); 5424 result_val->init_req(_slow_path, slow_result); 5425 result_i_o ->set_req(_slow_path, i_o()); 5426 result_mem ->set_req(_slow_path, reset_memory()); 5427 } 5428 5429 // Return the combined state. 5430 set_control( _gvn.transform(result_reg)); 5431 set_i_o( _gvn.transform(result_i_o)); 5432 set_all_memory( _gvn.transform(result_mem)); 5433 } // original reexecute is set back here 5434 5435 set_result(_gvn.transform(result_val)); 5436 return true; 5437 } 5438 5439 // If we have a tightly coupled allocation, the arraycopy may take care 5440 // of the array initialization. If one of the guards we insert between 5441 // the allocation and the arraycopy causes a deoptimization, an 5442 // uninitialized array will escape the compiled method. To prevent that 5443 // we set the JVM state for uncommon traps between the allocation and 5444 // the arraycopy to the state before the allocation so, in case of 5445 // deoptimization, we'll reexecute the allocation and the 5446 // initialization. 5447 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 5448 if (alloc != nullptr) { 5449 ciMethod* trap_method = alloc->jvms()->method(); 5450 int trap_bci = alloc->jvms()->bci(); 5451 5452 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5453 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 5454 // Make sure there's no store between the allocation and the 5455 // arraycopy otherwise visible side effects could be rexecuted 5456 // in case of deoptimization and cause incorrect execution. 5457 bool no_interfering_store = true; 5458 Node* mem = alloc->in(TypeFunc::Memory); 5459 if (mem->is_MergeMem()) { 5460 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 5461 Node* n = mms.memory(); 5462 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 5463 assert(n->is_Store(), "what else?"); 5464 no_interfering_store = false; 5465 break; 5466 } 5467 } 5468 } else { 5469 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 5470 Node* n = mms.memory(); 5471 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 5472 assert(n->is_Store(), "what else?"); 5473 no_interfering_store = false; 5474 break; 5475 } 5476 } 5477 } 5478 5479 if (no_interfering_store) { 5480 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc); 5481 5482 JVMState* saved_jvms = jvms(); 5483 saved_reexecute_sp = _reexecute_sp; 5484 5485 set_jvms(sfpt->jvms()); 5486 _reexecute_sp = jvms()->sp(); 5487 5488 return saved_jvms; 5489 } 5490 } 5491 } 5492 return nullptr; 5493 } 5494 5495 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack 5496 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter. 5497 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const { 5498 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 5499 uint size = alloc->req(); 5500 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 5501 old_jvms->set_map(sfpt); 5502 for (uint i = 0; i < size; i++) { 5503 sfpt->init_req(i, alloc->in(i)); 5504 } 5505 // re-push array length for deoptimization 5506 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength)); 5507 old_jvms->set_sp(old_jvms->sp()+1); 5508 old_jvms->set_monoff(old_jvms->monoff()+1); 5509 old_jvms->set_scloff(old_jvms->scloff()+1); 5510 old_jvms->set_endoff(old_jvms->endoff()+1); 5511 old_jvms->set_should_reexecute(true); 5512 5513 sfpt->set_i_o(map()->i_o()); 5514 sfpt->set_memory(map()->memory()); 5515 sfpt->set_control(map()->control()); 5516 return sfpt; 5517 } 5518 5519 // In case of a deoptimization, we restart execution at the 5520 // allocation, allocating a new array. We would leave an uninitialized 5521 // array in the heap that GCs wouldn't expect. Move the allocation 5522 // after the traps so we don't allocate the array if we 5523 // deoptimize. This is possible because tightly_coupled_allocation() 5524 // guarantees there's no observer of the allocated array at this point 5525 // and the control flow is simple enough. 5526 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards, 5527 int saved_reexecute_sp, uint new_idx) { 5528 if (saved_jvms_before_guards != nullptr && !stopped()) { 5529 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards); 5530 5531 assert(alloc != nullptr, "only with a tightly coupled allocation"); 5532 // restore JVM state to the state at the arraycopy 5533 saved_jvms_before_guards->map()->set_control(map()->control()); 5534 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?"); 5535 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?"); 5536 // If we've improved the types of some nodes (null check) while 5537 // emitting the guards, propagate them to the current state 5538 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx); 5539 set_jvms(saved_jvms_before_guards); 5540 _reexecute_sp = saved_reexecute_sp; 5541 5542 // Remove the allocation from above the guards 5543 CallProjections callprojs; 5544 alloc->extract_projections(&callprojs, true); 5545 InitializeNode* init = alloc->initialization(); 5546 Node* alloc_mem = alloc->in(TypeFunc::Memory); 5547 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 5548 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 5549 5550 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below 5551 // the allocation (i.e. is only valid if the allocation succeeds): 5552 // 1) replace CastIINode with AllocateArrayNode's length here 5553 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method 5554 // 5555 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate 5556 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy) 5557 Node* init_control = init->proj_out(TypeFunc::Control); 5558 Node* alloc_length = alloc->Ideal_length(); 5559 #ifdef ASSERT 5560 Node* prev_cast = nullptr; 5561 #endif 5562 for (uint i = 0; i < init_control->outcnt(); i++) { 5563 Node* init_out = init_control->raw_out(i); 5564 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) { 5565 #ifdef ASSERT 5566 if (prev_cast == nullptr) { 5567 prev_cast = init_out; 5568 } else { 5569 if (prev_cast->cmp(*init_out) == false) { 5570 prev_cast->dump(); 5571 init_out->dump(); 5572 assert(false, "not equal CastIINode"); 5573 } 5574 } 5575 #endif 5576 C->gvn_replace_by(init_out, alloc_length); 5577 } 5578 } 5579 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 5580 5581 // move the allocation here (after the guards) 5582 _gvn.hash_delete(alloc); 5583 alloc->set_req(TypeFunc::Control, control()); 5584 alloc->set_req(TypeFunc::I_O, i_o()); 5585 Node *mem = reset_memory(); 5586 set_all_memory(mem); 5587 alloc->set_req(TypeFunc::Memory, mem); 5588 set_control(init->proj_out_or_null(TypeFunc::Control)); 5589 set_i_o(callprojs.fallthrough_ioproj); 5590 5591 // Update memory as done in GraphKit::set_output_for_allocation() 5592 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 5593 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 5594 if (ary_type->isa_aryptr() && length_type != nullptr) { 5595 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 5596 } 5597 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 5598 int elemidx = C->get_alias_index(telemref); 5599 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw); 5600 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx); 5601 5602 Node* allocx = _gvn.transform(alloc); 5603 assert(allocx == alloc, "where has the allocation gone?"); 5604 assert(dest->is_CheckCastPP(), "not an allocation result?"); 5605 5606 _gvn.hash_delete(dest); 5607 dest->set_req(0, control()); 5608 Node* destx = _gvn.transform(dest); 5609 assert(destx == dest, "where has the allocation result gone?"); 5610 5611 array_ideal_length(alloc, ary_type, true); 5612 } 5613 } 5614 5615 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(), 5616 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary 5617 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array 5618 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter, 5619 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in 5620 // the interpreter similar to what we are doing for the newly emitted guards for the array copy. 5621 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc, 5622 JVMState* saved_jvms_before_guards) { 5623 if (saved_jvms_before_guards->map()->control()->is_IfProj()) { 5624 // There is at least one unrelated uncommon trap which needs to be replaced. 5625 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc); 5626 5627 JVMState* saved_jvms = jvms(); 5628 const int saved_reexecute_sp = _reexecute_sp; 5629 set_jvms(sfpt->jvms()); 5630 _reexecute_sp = jvms()->sp(); 5631 5632 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards); 5633 5634 // Restore state 5635 set_jvms(saved_jvms); 5636 _reexecute_sp = saved_reexecute_sp; 5637 } 5638 } 5639 5640 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon 5641 // traps will have the state of the array allocation. Let the old uncommon trap nodes die. 5642 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) { 5643 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards 5644 while (if_proj->is_IfProj()) { 5645 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj); 5646 if (uncommon_trap != nullptr) { 5647 create_new_uncommon_trap(uncommon_trap); 5648 } 5649 assert(if_proj->in(0)->is_If(), "must be If"); 5650 if_proj = if_proj->in(0)->in(0); 5651 } 5652 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(), 5653 "must have reached control projection of init node"); 5654 } 5655 5656 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) { 5657 const int trap_request = uncommon_trap_call->uncommon_trap_request(); 5658 assert(trap_request != 0, "no valid UCT trap request"); 5659 PreserveJVMState pjvms(this); 5660 set_control(uncommon_trap_call->in(0)); 5661 uncommon_trap(Deoptimization::trap_request_reason(trap_request), 5662 Deoptimization::trap_request_action(trap_request)); 5663 assert(stopped(), "Should be stopped"); 5664 _gvn.hash_delete(uncommon_trap_call); 5665 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it 5666 } 5667 5668 // Common checks for array sorting intrinsics arguments. 5669 // Returns `true` if checks passed. 5670 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) { 5671 // check address of the class 5672 if (elementType == nullptr || elementType->is_top()) { 5673 return false; // dead path 5674 } 5675 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr(); 5676 if (elem_klass == nullptr) { 5677 return false; // dead path 5678 } 5679 // java_mirror_type() returns non-null for compile-time Class constants only 5680 ciType* elem_type = elem_klass->java_mirror_type(); 5681 if (elem_type == nullptr) { 5682 return false; 5683 } 5684 bt = elem_type->basic_type(); 5685 // Disable the intrinsic if the CPU does not support SIMD sort 5686 if (!Matcher::supports_simd_sort(bt)) { 5687 return false; 5688 } 5689 // check address of the array 5690 if (obj == nullptr || obj->is_top()) { 5691 return false; // dead path 5692 } 5693 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr(); 5694 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) { 5695 return false; // failed input validation 5696 } 5697 return true; 5698 } 5699 5700 //------------------------------inline_array_partition----------------------- 5701 bool LibraryCallKit::inline_array_partition() { 5702 address stubAddr = StubRoutines::select_array_partition_function(); 5703 if (stubAddr == nullptr) { 5704 return false; // Intrinsic's stub is not implemented on this platform 5705 } 5706 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)"); 5707 5708 // no receiver because it is a static method 5709 Node* elementType = argument(0); 5710 Node* obj = argument(1); 5711 Node* offset = argument(2); // long 5712 Node* fromIndex = argument(4); 5713 Node* toIndex = argument(5); 5714 Node* indexPivot1 = argument(6); 5715 Node* indexPivot2 = argument(7); 5716 // PartitionOperation: argument(8) is ignored 5717 5718 Node* pivotIndices = nullptr; 5719 BasicType bt = T_ILLEGAL; 5720 5721 if (!check_array_sort_arguments(elementType, obj, bt)) { 5722 return false; 5723 } 5724 null_check(obj); 5725 // If obj is dead, only null-path is taken. 5726 if (stopped()) { 5727 return true; 5728 } 5729 // Set the original stack and the reexecute bit for the interpreter to reexecute 5730 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens. 5731 { PreserveReexecuteState preexecs(this); 5732 jvms()->set_should_reexecute(true); 5733 5734 Node* obj_adr = make_unsafe_address(obj, offset); 5735 5736 // create the pivotIndices array of type int and size = 2 5737 Node* size = intcon(2); 5738 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT))); 5739 pivotIndices = new_array(klass_node, size, 0); // no arguments to push 5740 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices); 5741 guarantee(alloc != nullptr, "created above"); 5742 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT)); 5743 5744 // pass the basic type enum to the stub 5745 Node* elemType = intcon(bt); 5746 5747 // Call the stub 5748 const char *stubName = "array_partition_stub"; 5749 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(), 5750 stubAddr, stubName, TypePtr::BOTTOM, 5751 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr, 5752 indexPivot1, indexPivot2); 5753 5754 } // original reexecute is set back here 5755 5756 if (!stopped()) { 5757 set_result(pivotIndices); 5758 } 5759 5760 return true; 5761 } 5762 5763 5764 //------------------------------inline_array_sort----------------------- 5765 bool LibraryCallKit::inline_array_sort() { 5766 address stubAddr = StubRoutines::select_arraysort_function(); 5767 if (stubAddr == nullptr) { 5768 return false; // Intrinsic's stub is not implemented on this platform 5769 } 5770 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)"); 5771 5772 // no receiver because it is a static method 5773 Node* elementType = argument(0); 5774 Node* obj = argument(1); 5775 Node* offset = argument(2); // long 5776 Node* fromIndex = argument(4); 5777 Node* toIndex = argument(5); 5778 // SortOperation: argument(6) is ignored 5779 5780 BasicType bt = T_ILLEGAL; 5781 5782 if (!check_array_sort_arguments(elementType, obj, bt)) { 5783 return false; 5784 } 5785 null_check(obj); 5786 // If obj is dead, only null-path is taken. 5787 if (stopped()) { 5788 return true; 5789 } 5790 Node* obj_adr = make_unsafe_address(obj, offset); 5791 5792 // pass the basic type enum to the stub 5793 Node* elemType = intcon(bt); 5794 5795 // Call the stub. 5796 const char *stubName = "arraysort_stub"; 5797 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(), 5798 stubAddr, stubName, TypePtr::BOTTOM, 5799 obj_adr, elemType, fromIndex, toIndex); 5800 5801 return true; 5802 } 5803 5804 5805 //------------------------------inline_arraycopy----------------------- 5806 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 5807 // Object dest, int destPos, 5808 // int length); 5809 bool LibraryCallKit::inline_arraycopy() { 5810 // Get the arguments. 5811 Node* src = argument(0); // type: oop 5812 Node* src_offset = argument(1); // type: int 5813 Node* dest = argument(2); // type: oop 5814 Node* dest_offset = argument(3); // type: int 5815 Node* length = argument(4); // type: int 5816 5817 uint new_idx = C->unique(); 5818 5819 // Check for allocation before we add nodes that would confuse 5820 // tightly_coupled_allocation() 5821 AllocateArrayNode* alloc = tightly_coupled_allocation(dest); 5822 5823 int saved_reexecute_sp = -1; 5824 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 5825 // See arraycopy_restore_alloc_state() comment 5826 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards 5827 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation 5828 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards 5829 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr); 5830 5831 // The following tests must be performed 5832 // (1) src and dest are arrays. 5833 // (2) src and dest arrays must have elements of the same BasicType 5834 // (3) src and dest must not be null. 5835 // (4) src_offset must not be negative. 5836 // (5) dest_offset must not be negative. 5837 // (6) length must not be negative. 5838 // (7) src_offset + length must not exceed length of src. 5839 // (8) dest_offset + length must not exceed length of dest. 5840 // (9) each element of an oop array must be assignable 5841 5842 // (3) src and dest must not be null. 5843 // always do this here because we need the JVM state for uncommon traps 5844 Node* null_ctl = top(); 5845 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 5846 assert(null_ctl->is_top(), "no null control here"); 5847 dest = null_check(dest, T_ARRAY); 5848 5849 if (!can_emit_guards) { 5850 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any 5851 // guards but the arraycopy node could still take advantage of a 5852 // tightly allocated allocation. tightly_coupled_allocation() is 5853 // called again to make sure it takes the null check above into 5854 // account: the null check is mandatory and if it caused an 5855 // uncommon trap to be emitted then the allocation can't be 5856 // considered tightly coupled in this context. 5857 alloc = tightly_coupled_allocation(dest); 5858 } 5859 5860 bool validated = false; 5861 5862 const Type* src_type = _gvn.type(src); 5863 const Type* dest_type = _gvn.type(dest); 5864 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5865 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5866 5867 // Do we have the type of src? 5868 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM); 5869 // Do we have the type of dest? 5870 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM); 5871 // Is the type for src from speculation? 5872 bool src_spec = false; 5873 // Is the type for dest from speculation? 5874 bool dest_spec = false; 5875 5876 if ((!has_src || !has_dest) && can_emit_guards) { 5877 // We don't have sufficient type information, let's see if 5878 // speculative types can help. We need to have types for both src 5879 // and dest so that it pays off. 5880 5881 // Do we already have or could we have type information for src 5882 bool could_have_src = has_src; 5883 // Do we already have or could we have type information for dest 5884 bool could_have_dest = has_dest; 5885 5886 ciKlass* src_k = nullptr; 5887 if (!has_src) { 5888 src_k = src_type->speculative_type_not_null(); 5889 if (src_k != nullptr && src_k->is_array_klass()) { 5890 could_have_src = true; 5891 } 5892 } 5893 5894 ciKlass* dest_k = nullptr; 5895 if (!has_dest) { 5896 dest_k = dest_type->speculative_type_not_null(); 5897 if (dest_k != nullptr && dest_k->is_array_klass()) { 5898 could_have_dest = true; 5899 } 5900 } 5901 5902 if (could_have_src && could_have_dest) { 5903 // This is going to pay off so emit the required guards 5904 if (!has_src) { 5905 src = maybe_cast_profiled_obj(src, src_k, true); 5906 src_type = _gvn.type(src); 5907 top_src = src_type->isa_aryptr(); 5908 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM); 5909 src_spec = true; 5910 } 5911 if (!has_dest) { 5912 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5913 dest_type = _gvn.type(dest); 5914 top_dest = dest_type->isa_aryptr(); 5915 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM); 5916 dest_spec = true; 5917 } 5918 } 5919 } 5920 5921 if (has_src && has_dest && can_emit_guards) { 5922 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type(); 5923 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type(); 5924 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT; 5925 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT; 5926 5927 if (src_elem == dest_elem && src_elem == T_OBJECT) { 5928 // If both arrays are object arrays then having the exact types 5929 // for both will remove the need for a subtype check at runtime 5930 // before the call and may make it possible to pick a faster copy 5931 // routine (without a subtype check on every element) 5932 // Do we have the exact type of src? 5933 bool could_have_src = src_spec; 5934 // Do we have the exact type of dest? 5935 bool could_have_dest = dest_spec; 5936 ciKlass* src_k = nullptr; 5937 ciKlass* dest_k = nullptr; 5938 if (!src_spec) { 5939 src_k = src_type->speculative_type_not_null(); 5940 if (src_k != nullptr && src_k->is_array_klass()) { 5941 could_have_src = true; 5942 } 5943 } 5944 if (!dest_spec) { 5945 dest_k = dest_type->speculative_type_not_null(); 5946 if (dest_k != nullptr && dest_k->is_array_klass()) { 5947 could_have_dest = true; 5948 } 5949 } 5950 if (could_have_src && could_have_dest) { 5951 // If we can have both exact types, emit the missing guards 5952 if (could_have_src && !src_spec) { 5953 src = maybe_cast_profiled_obj(src, src_k, true); 5954 } 5955 if (could_have_dest && !dest_spec) { 5956 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5957 } 5958 } 5959 } 5960 } 5961 5962 ciMethod* trap_method = method(); 5963 int trap_bci = bci(); 5964 if (saved_jvms_before_guards != nullptr) { 5965 trap_method = alloc->jvms()->method(); 5966 trap_bci = alloc->jvms()->bci(); 5967 } 5968 5969 bool negative_length_guard_generated = false; 5970 5971 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5972 can_emit_guards && 5973 !src->is_top() && !dest->is_top()) { 5974 // validate arguments: enables transformation the ArrayCopyNode 5975 validated = true; 5976 5977 RegionNode* slow_region = new RegionNode(1); 5978 record_for_igvn(slow_region); 5979 5980 // (1) src and dest are arrays. 5981 generate_non_array_guard(load_object_klass(src), slow_region, &src); 5982 generate_non_array_guard(load_object_klass(dest), slow_region, &dest); 5983 5984 // (2) src and dest arrays must have elements of the same BasicType 5985 // done at macro expansion or at Ideal transformation time 5986 5987 // (4) src_offset must not be negative. 5988 generate_negative_guard(src_offset, slow_region); 5989 5990 // (5) dest_offset must not be negative. 5991 generate_negative_guard(dest_offset, slow_region); 5992 5993 // (7) src_offset + length must not exceed length of src. 5994 generate_limit_guard(src_offset, length, 5995 load_array_length(src), 5996 slow_region); 5997 5998 // (8) dest_offset + length must not exceed length of dest. 5999 generate_limit_guard(dest_offset, length, 6000 load_array_length(dest), 6001 slow_region); 6002 6003 // (6) length must not be negative. 6004 // This is also checked in generate_arraycopy() during macro expansion, but 6005 // we also have to check it here for the case where the ArrayCopyNode will 6006 // be eliminated by Escape Analysis. 6007 if (EliminateAllocations) { 6008 generate_negative_guard(length, slow_region); 6009 negative_length_guard_generated = true; 6010 } 6011 6012 // (9) each element of an oop array must be assignable 6013 Node* dest_klass = load_object_klass(dest); 6014 if (src != dest) { 6015 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass); 6016 6017 if (not_subtype_ctrl != top()) { 6018 PreserveJVMState pjvms(this); 6019 set_control(not_subtype_ctrl); 6020 uncommon_trap(Deoptimization::Reason_intrinsic, 6021 Deoptimization::Action_make_not_entrant); 6022 assert(stopped(), "Should be stopped"); 6023 } 6024 } 6025 { 6026 PreserveJVMState pjvms(this); 6027 set_control(_gvn.transform(slow_region)); 6028 uncommon_trap(Deoptimization::Reason_intrinsic, 6029 Deoptimization::Action_make_not_entrant); 6030 assert(stopped(), "Should be stopped"); 6031 } 6032 6033 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr(); 6034 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type(); 6035 src = _gvn.transform(new CheckCastPPNode(control(), src, toop)); 6036 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx); 6037 } 6038 6039 if (stopped()) { 6040 return true; 6041 } 6042 6043 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated, 6044 // Create LoadRange and LoadKlass nodes for use during macro expansion here 6045 // so the compiler has a chance to eliminate them: during macro expansion, 6046 // we have to set their control (CastPP nodes are eliminated). 6047 load_object_klass(src), load_object_klass(dest), 6048 load_array_length(src), load_array_length(dest)); 6049 6050 ac->set_arraycopy(validated); 6051 6052 Node* n = _gvn.transform(ac); 6053 if (n == ac) { 6054 ac->connect_outputs(this); 6055 } else { 6056 assert(validated, "shouldn't transform if all arguments not validated"); 6057 set_all_memory(n); 6058 } 6059 clear_upper_avx(); 6060 6061 6062 return true; 6063 } 6064 6065 6066 // Helper function which determines if an arraycopy immediately follows 6067 // an allocation, with no intervening tests or other escapes for the object. 6068 AllocateArrayNode* 6069 LibraryCallKit::tightly_coupled_allocation(Node* ptr) { 6070 if (stopped()) return nullptr; // no fast path 6071 if (!C->do_aliasing()) return nullptr; // no MergeMems around 6072 6073 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr); 6074 if (alloc == nullptr) return nullptr; 6075 6076 Node* rawmem = memory(Compile::AliasIdxRaw); 6077 // Is the allocation's memory state untouched? 6078 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 6079 // Bail out if there have been raw-memory effects since the allocation. 6080 // (Example: There might have been a call or safepoint.) 6081 return nullptr; 6082 } 6083 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 6084 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 6085 return nullptr; 6086 } 6087 6088 // There must be no unexpected observers of this allocation. 6089 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 6090 Node* obs = ptr->fast_out(i); 6091 if (obs != this->map()) { 6092 return nullptr; 6093 } 6094 } 6095 6096 // This arraycopy must unconditionally follow the allocation of the ptr. 6097 Node* alloc_ctl = ptr->in(0); 6098 Node* ctl = control(); 6099 while (ctl != alloc_ctl) { 6100 // There may be guards which feed into the slow_region. 6101 // Any other control flow means that we might not get a chance 6102 // to finish initializing the allocated object. 6103 // Various low-level checks bottom out in uncommon traps. These 6104 // are considered safe since we've already checked above that 6105 // there is no unexpected observer of this allocation. 6106 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) { 6107 assert(ctl->in(0)->is_If(), "must be If"); 6108 ctl = ctl->in(0)->in(0); 6109 } else { 6110 return nullptr; 6111 } 6112 } 6113 6114 // If we get this far, we have an allocation which immediately 6115 // precedes the arraycopy, and we can take over zeroing the new object. 6116 // The arraycopy will finish the initialization, and provide 6117 // a new control state to which we will anchor the destination pointer. 6118 6119 return alloc; 6120 } 6121 6122 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) { 6123 if (node->is_IfProj()) { 6124 Node* other_proj = node->as_IfProj()->other_if_proj(); 6125 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) { 6126 Node* obs = other_proj->fast_out(j); 6127 if (obs->in(0) == other_proj && obs->is_CallStaticJava() && 6128 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) { 6129 return obs->as_CallStaticJava(); 6130 } 6131 } 6132 } 6133 return nullptr; 6134 } 6135 6136 //-------------inline_encodeISOArray----------------------------------- 6137 // encode char[] to byte[] in ISO_8859_1 or ASCII 6138 bool LibraryCallKit::inline_encodeISOArray(bool ascii) { 6139 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 6140 // no receiver since it is static method 6141 Node *src = argument(0); 6142 Node *src_offset = argument(1); 6143 Node *dst = argument(2); 6144 Node *dst_offset = argument(3); 6145 Node *length = argument(4); 6146 6147 src = must_be_not_null(src, true); 6148 dst = must_be_not_null(dst, true); 6149 6150 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6151 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr(); 6152 if (src_type == nullptr || src_type->elem() == Type::BOTTOM || 6153 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) { 6154 // failed array check 6155 return false; 6156 } 6157 6158 // Figure out the size and type of the elements we will be copying. 6159 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6160 BasicType dst_elem = dst_type->elem()->array_element_basic_type(); 6161 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 6162 return false; 6163 } 6164 6165 Node* src_start = array_element_address(src, src_offset, T_CHAR); 6166 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 6167 // 'src_start' points to src array + scaled offset 6168 // 'dst_start' points to dst array + scaled offset 6169 6170 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 6171 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii); 6172 enc = _gvn.transform(enc); 6173 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 6174 set_memory(res_mem, mtype); 6175 set_result(enc); 6176 clear_upper_avx(); 6177 6178 return true; 6179 } 6180 6181 //-------------inline_multiplyToLen----------------------------------- 6182 bool LibraryCallKit::inline_multiplyToLen() { 6183 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 6184 6185 address stubAddr = StubRoutines::multiplyToLen(); 6186 if (stubAddr == nullptr) { 6187 return false; // Intrinsic's stub is not implemented on this platform 6188 } 6189 const char* stubName = "multiplyToLen"; 6190 6191 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 6192 6193 // no receiver because it is a static method 6194 Node* x = argument(0); 6195 Node* xlen = argument(1); 6196 Node* y = argument(2); 6197 Node* ylen = argument(3); 6198 Node* z = argument(4); 6199 6200 x = must_be_not_null(x, true); 6201 y = must_be_not_null(y, true); 6202 6203 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr(); 6204 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr(); 6205 if (x_type == nullptr || x_type->elem() == Type::BOTTOM || 6206 y_type == nullptr || y_type->elem() == Type::BOTTOM) { 6207 // failed array check 6208 return false; 6209 } 6210 6211 BasicType x_elem = x_type->elem()->array_element_basic_type(); 6212 BasicType y_elem = y_type->elem()->array_element_basic_type(); 6213 if (x_elem != T_INT || y_elem != T_INT) { 6214 return false; 6215 } 6216 6217 Node* x_start = array_element_address(x, intcon(0), x_elem); 6218 Node* y_start = array_element_address(y, intcon(0), y_elem); 6219 // 'x_start' points to x array + scaled xlen 6220 // 'y_start' points to y array + scaled ylen 6221 6222 Node* z_start = array_element_address(z, intcon(0), T_INT); 6223 6224 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6225 OptoRuntime::multiplyToLen_Type(), 6226 stubAddr, stubName, TypePtr::BOTTOM, 6227 x_start, xlen, y_start, ylen, z_start); 6228 6229 C->set_has_split_ifs(true); // Has chance for split-if optimization 6230 set_result(z); 6231 return true; 6232 } 6233 6234 //-------------inline_squareToLen------------------------------------ 6235 bool LibraryCallKit::inline_squareToLen() { 6236 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 6237 6238 address stubAddr = StubRoutines::squareToLen(); 6239 if (stubAddr == nullptr) { 6240 return false; // Intrinsic's stub is not implemented on this platform 6241 } 6242 const char* stubName = "squareToLen"; 6243 6244 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 6245 6246 Node* x = argument(0); 6247 Node* len = argument(1); 6248 Node* z = argument(2); 6249 Node* zlen = argument(3); 6250 6251 x = must_be_not_null(x, true); 6252 z = must_be_not_null(z, true); 6253 6254 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr(); 6255 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr(); 6256 if (x_type == nullptr || x_type->elem() == Type::BOTTOM || 6257 z_type == nullptr || z_type->elem() == Type::BOTTOM) { 6258 // failed array check 6259 return false; 6260 } 6261 6262 BasicType x_elem = x_type->elem()->array_element_basic_type(); 6263 BasicType z_elem = z_type->elem()->array_element_basic_type(); 6264 if (x_elem != T_INT || z_elem != T_INT) { 6265 return false; 6266 } 6267 6268 6269 Node* x_start = array_element_address(x, intcon(0), x_elem); 6270 Node* z_start = array_element_address(z, intcon(0), z_elem); 6271 6272 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6273 OptoRuntime::squareToLen_Type(), 6274 stubAddr, stubName, TypePtr::BOTTOM, 6275 x_start, len, z_start, zlen); 6276 6277 set_result(z); 6278 return true; 6279 } 6280 6281 //-------------inline_mulAdd------------------------------------------ 6282 bool LibraryCallKit::inline_mulAdd() { 6283 assert(UseMulAddIntrinsic, "not implemented on this platform"); 6284 6285 address stubAddr = StubRoutines::mulAdd(); 6286 if (stubAddr == nullptr) { 6287 return false; // Intrinsic's stub is not implemented on this platform 6288 } 6289 const char* stubName = "mulAdd"; 6290 6291 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 6292 6293 Node* out = argument(0); 6294 Node* in = argument(1); 6295 Node* offset = argument(2); 6296 Node* len = argument(3); 6297 Node* k = argument(4); 6298 6299 in = must_be_not_null(in, true); 6300 out = must_be_not_null(out, true); 6301 6302 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr(); 6303 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr(); 6304 if (out_type == nullptr || out_type->elem() == Type::BOTTOM || 6305 in_type == nullptr || in_type->elem() == Type::BOTTOM) { 6306 // failed array check 6307 return false; 6308 } 6309 6310 BasicType out_elem = out_type->elem()->array_element_basic_type(); 6311 BasicType in_elem = in_type->elem()->array_element_basic_type(); 6312 if (out_elem != T_INT || in_elem != T_INT) { 6313 return false; 6314 } 6315 6316 Node* outlen = load_array_length(out); 6317 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 6318 Node* out_start = array_element_address(out, intcon(0), out_elem); 6319 Node* in_start = array_element_address(in, intcon(0), in_elem); 6320 6321 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6322 OptoRuntime::mulAdd_Type(), 6323 stubAddr, stubName, TypePtr::BOTTOM, 6324 out_start,in_start, new_offset, len, k); 6325 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6326 set_result(result); 6327 return true; 6328 } 6329 6330 //-------------inline_montgomeryMultiply----------------------------------- 6331 bool LibraryCallKit::inline_montgomeryMultiply() { 6332 address stubAddr = StubRoutines::montgomeryMultiply(); 6333 if (stubAddr == nullptr) { 6334 return false; // Intrinsic's stub is not implemented on this platform 6335 } 6336 6337 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 6338 const char* stubName = "montgomery_multiply"; 6339 6340 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 6341 6342 Node* a = argument(0); 6343 Node* b = argument(1); 6344 Node* n = argument(2); 6345 Node* len = argument(3); 6346 Node* inv = argument(4); 6347 Node* m = argument(6); 6348 6349 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr(); 6350 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr(); 6351 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr(); 6352 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr(); 6353 if (a_type == nullptr || a_type->elem() == Type::BOTTOM || 6354 b_type == nullptr || b_type->elem() == Type::BOTTOM || 6355 n_type == nullptr || n_type->elem() == Type::BOTTOM || 6356 m_type == nullptr || m_type->elem() == Type::BOTTOM) { 6357 // failed array check 6358 return false; 6359 } 6360 6361 BasicType a_elem = a_type->elem()->array_element_basic_type(); 6362 BasicType b_elem = b_type->elem()->array_element_basic_type(); 6363 BasicType n_elem = n_type->elem()->array_element_basic_type(); 6364 BasicType m_elem = m_type->elem()->array_element_basic_type(); 6365 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 6366 return false; 6367 } 6368 6369 // Make the call 6370 { 6371 Node* a_start = array_element_address(a, intcon(0), a_elem); 6372 Node* b_start = array_element_address(b, intcon(0), b_elem); 6373 Node* n_start = array_element_address(n, intcon(0), n_elem); 6374 Node* m_start = array_element_address(m, intcon(0), m_elem); 6375 6376 Node* call = make_runtime_call(RC_LEAF, 6377 OptoRuntime::montgomeryMultiply_Type(), 6378 stubAddr, stubName, TypePtr::BOTTOM, 6379 a_start, b_start, n_start, len, inv, top(), 6380 m_start); 6381 set_result(m); 6382 } 6383 6384 return true; 6385 } 6386 6387 bool LibraryCallKit::inline_montgomerySquare() { 6388 address stubAddr = StubRoutines::montgomerySquare(); 6389 if (stubAddr == nullptr) { 6390 return false; // Intrinsic's stub is not implemented on this platform 6391 } 6392 6393 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 6394 const char* stubName = "montgomery_square"; 6395 6396 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 6397 6398 Node* a = argument(0); 6399 Node* n = argument(1); 6400 Node* len = argument(2); 6401 Node* inv = argument(3); 6402 Node* m = argument(5); 6403 6404 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr(); 6405 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr(); 6406 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr(); 6407 if (a_type == nullptr || a_type->elem() == Type::BOTTOM || 6408 n_type == nullptr || n_type->elem() == Type::BOTTOM || 6409 m_type == nullptr || m_type->elem() == Type::BOTTOM) { 6410 // failed array check 6411 return false; 6412 } 6413 6414 BasicType a_elem = a_type->elem()->array_element_basic_type(); 6415 BasicType n_elem = n_type->elem()->array_element_basic_type(); 6416 BasicType m_elem = m_type->elem()->array_element_basic_type(); 6417 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 6418 return false; 6419 } 6420 6421 // Make the call 6422 { 6423 Node* a_start = array_element_address(a, intcon(0), a_elem); 6424 Node* n_start = array_element_address(n, intcon(0), n_elem); 6425 Node* m_start = array_element_address(m, intcon(0), m_elem); 6426 6427 Node* call = make_runtime_call(RC_LEAF, 6428 OptoRuntime::montgomerySquare_Type(), 6429 stubAddr, stubName, TypePtr::BOTTOM, 6430 a_start, n_start, len, inv, top(), 6431 m_start); 6432 set_result(m); 6433 } 6434 6435 return true; 6436 } 6437 6438 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) { 6439 address stubAddr = nullptr; 6440 const char* stubName = nullptr; 6441 6442 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift(); 6443 if (stubAddr == nullptr) { 6444 return false; // Intrinsic's stub is not implemented on this platform 6445 } 6446 6447 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker"; 6448 6449 assert(callee()->signature()->size() == 5, "expected 5 arguments"); 6450 6451 Node* newArr = argument(0); 6452 Node* oldArr = argument(1); 6453 Node* newIdx = argument(2); 6454 Node* shiftCount = argument(3); 6455 Node* numIter = argument(4); 6456 6457 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr(); 6458 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr(); 6459 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM || 6460 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) { 6461 return false; 6462 } 6463 6464 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type(); 6465 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type(); 6466 if (newArr_elem != T_INT || oldArr_elem != T_INT) { 6467 return false; 6468 } 6469 6470 // Make the call 6471 { 6472 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem); 6473 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem); 6474 6475 Node* call = make_runtime_call(RC_LEAF, 6476 OptoRuntime::bigIntegerShift_Type(), 6477 stubAddr, 6478 stubName, 6479 TypePtr::BOTTOM, 6480 newArr_start, 6481 oldArr_start, 6482 newIdx, 6483 shiftCount, 6484 numIter); 6485 } 6486 6487 return true; 6488 } 6489 6490 //-------------inline_vectorizedMismatch------------------------------ 6491 bool LibraryCallKit::inline_vectorizedMismatch() { 6492 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform"); 6493 6494 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 6495 Node* obja = argument(0); // Object 6496 Node* aoffset = argument(1); // long 6497 Node* objb = argument(3); // Object 6498 Node* boffset = argument(4); // long 6499 Node* length = argument(6); // int 6500 Node* scale = argument(7); // int 6501 6502 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr(); 6503 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr(); 6504 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM || 6505 objb_t == nullptr || objb_t->elem() == Type::BOTTOM || 6506 scale == top()) { 6507 return false; // failed input validation 6508 } 6509 6510 Node* obja_adr = make_unsafe_address(obja, aoffset); 6511 Node* objb_adr = make_unsafe_address(objb, boffset); 6512 6513 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size. 6514 // 6515 // inline_limit = ArrayOperationPartialInlineSize / element_size; 6516 // if (length <= inline_limit) { 6517 // inline_path: 6518 // vmask = VectorMaskGen length 6519 // vload1 = LoadVectorMasked obja, vmask 6520 // vload2 = LoadVectorMasked objb, vmask 6521 // result1 = VectorCmpMasked vload1, vload2, vmask 6522 // } else { 6523 // call_stub_path: 6524 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale) 6525 // } 6526 // exit_block: 6527 // return Phi(result1, result2); 6528 // 6529 enum { inline_path = 1, // input is small enough to process it all at once 6530 stub_path = 2, // input is too large; call into the VM 6531 PATH_LIMIT = 3 6532 }; 6533 6534 Node* exit_block = new RegionNode(PATH_LIMIT); 6535 Node* result_phi = new PhiNode(exit_block, TypeInt::INT); 6536 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM); 6537 6538 Node* call_stub_path = control(); 6539 6540 BasicType elem_bt = T_ILLEGAL; 6541 6542 const TypeInt* scale_t = _gvn.type(scale)->is_int(); 6543 if (scale_t->is_con()) { 6544 switch (scale_t->get_con()) { 6545 case 0: elem_bt = T_BYTE; break; 6546 case 1: elem_bt = T_SHORT; break; 6547 case 2: elem_bt = T_INT; break; 6548 case 3: elem_bt = T_LONG; break; 6549 6550 default: elem_bt = T_ILLEGAL; break; // not supported 6551 } 6552 } 6553 6554 int inline_limit = 0; 6555 bool do_partial_inline = false; 6556 6557 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) { 6558 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt); 6559 do_partial_inline = inline_limit >= 16; 6560 } 6561 6562 if (do_partial_inline) { 6563 assert(elem_bt != T_ILLEGAL, "sanity"); 6564 6565 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) && 6566 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) && 6567 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) { 6568 6569 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit); 6570 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit))); 6571 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt)); 6572 6573 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN); 6574 6575 if (!stopped()) { 6576 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin))); 6577 6578 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr(); 6579 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr(); 6580 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t)); 6581 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t)); 6582 6583 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt)); 6584 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask)); 6585 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask)); 6586 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT)); 6587 6588 exit_block->init_req(inline_path, control()); 6589 memory_phi->init_req(inline_path, map()->memory()); 6590 result_phi->init_req(inline_path, result); 6591 6592 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size())); 6593 clear_upper_avx(); 6594 } 6595 } 6596 } 6597 6598 if (call_stub_path != nullptr) { 6599 set_control(call_stub_path); 6600 6601 Node* call = make_runtime_call(RC_LEAF, 6602 OptoRuntime::vectorizedMismatch_Type(), 6603 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM, 6604 obja_adr, objb_adr, length, scale); 6605 6606 exit_block->init_req(stub_path, control()); 6607 memory_phi->init_req(stub_path, map()->memory()); 6608 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms))); 6609 } 6610 6611 exit_block = _gvn.transform(exit_block); 6612 memory_phi = _gvn.transform(memory_phi); 6613 result_phi = _gvn.transform(result_phi); 6614 6615 record_for_igvn(exit_block); 6616 record_for_igvn(memory_phi); 6617 record_for_igvn(result_phi); 6618 6619 set_control(exit_block); 6620 set_all_memory(memory_phi); 6621 set_result(result_phi); 6622 6623 return true; 6624 } 6625 6626 //------------------------------inline_vectorizedHashcode---------------------------- 6627 bool LibraryCallKit::inline_vectorizedHashCode() { 6628 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform"); 6629 6630 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters"); 6631 Node* array = argument(0); 6632 Node* offset = argument(1); 6633 Node* length = argument(2); 6634 Node* initialValue = argument(3); 6635 Node* basic_type = argument(4); 6636 6637 if (basic_type == top()) { 6638 return false; // failed input validation 6639 } 6640 6641 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int(); 6642 if (!basic_type_t->is_con()) { 6643 return false; // Only intrinsify if mode argument is constant 6644 } 6645 6646 array = must_be_not_null(array, true); 6647 6648 BasicType bt = (BasicType)basic_type_t->get_con(); 6649 6650 // Resolve address of first element 6651 Node* array_start = array_element_address(array, offset, bt); 6652 6653 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)), 6654 array_start, length, initialValue, basic_type))); 6655 clear_upper_avx(); 6656 6657 return true; 6658 } 6659 6660 /** 6661 * Calculate CRC32 for byte. 6662 * int java.util.zip.CRC32.update(int crc, int b) 6663 */ 6664 bool LibraryCallKit::inline_updateCRC32() { 6665 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support"); 6666 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 6667 // no receiver since it is static method 6668 Node* crc = argument(0); // type: int 6669 Node* b = argument(1); // type: int 6670 6671 /* 6672 * int c = ~ crc; 6673 * b = timesXtoThe32[(b ^ c) & 0xFF]; 6674 * b = b ^ (c >>> 8); 6675 * crc = ~b; 6676 */ 6677 6678 Node* M1 = intcon(-1); 6679 crc = _gvn.transform(new XorINode(crc, M1)); 6680 Node* result = _gvn.transform(new XorINode(crc, b)); 6681 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 6682 6683 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 6684 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 6685 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 6686 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 6687 6688 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 6689 result = _gvn.transform(new XorINode(crc, result)); 6690 result = _gvn.transform(new XorINode(result, M1)); 6691 set_result(result); 6692 return true; 6693 } 6694 6695 /** 6696 * Calculate CRC32 for byte[] array. 6697 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 6698 */ 6699 bool LibraryCallKit::inline_updateBytesCRC32() { 6700 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support"); 6701 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6702 // no receiver since it is static method 6703 Node* crc = argument(0); // type: int 6704 Node* src = argument(1); // type: oop 6705 Node* offset = argument(2); // type: int 6706 Node* length = argument(3); // type: int 6707 6708 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6709 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6710 // failed array check 6711 return false; 6712 } 6713 6714 // Figure out the size and type of the elements we will be copying. 6715 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6716 if (src_elem != T_BYTE) { 6717 return false; 6718 } 6719 6720 // 'src_start' points to src array + scaled offset 6721 src = must_be_not_null(src, true); 6722 Node* src_start = array_element_address(src, offset, src_elem); 6723 6724 // We assume that range check is done by caller. 6725 // TODO: generate range check (offset+length < src.length) in debug VM. 6726 6727 // Call the stub. 6728 address stubAddr = StubRoutines::updateBytesCRC32(); 6729 const char *stubName = "updateBytesCRC32"; 6730 6731 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 6732 stubAddr, stubName, TypePtr::BOTTOM, 6733 crc, src_start, length); 6734 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6735 set_result(result); 6736 return true; 6737 } 6738 6739 /** 6740 * Calculate CRC32 for ByteBuffer. 6741 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 6742 */ 6743 bool LibraryCallKit::inline_updateByteBufferCRC32() { 6744 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support"); 6745 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 6746 // no receiver since it is static method 6747 Node* crc = argument(0); // type: int 6748 Node* src = argument(1); // type: long 6749 Node* offset = argument(3); // type: int 6750 Node* length = argument(4); // type: int 6751 6752 src = ConvL2X(src); // adjust Java long to machine word 6753 Node* base = _gvn.transform(new CastX2PNode(src)); 6754 offset = ConvI2X(offset); 6755 6756 // 'src_start' points to src array + scaled offset 6757 Node* src_start = basic_plus_adr(top(), base, offset); 6758 6759 // Call the stub. 6760 address stubAddr = StubRoutines::updateBytesCRC32(); 6761 const char *stubName = "updateBytesCRC32"; 6762 6763 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 6764 stubAddr, stubName, TypePtr::BOTTOM, 6765 crc, src_start, length); 6766 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6767 set_result(result); 6768 return true; 6769 } 6770 6771 //------------------------------get_table_from_crc32c_class----------------------- 6772 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 6773 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class); 6774 assert (table != nullptr, "wrong version of java.util.zip.CRC32C"); 6775 6776 return table; 6777 } 6778 6779 //------------------------------inline_updateBytesCRC32C----------------------- 6780 // 6781 // Calculate CRC32C for byte[] array. 6782 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 6783 // 6784 bool LibraryCallKit::inline_updateBytesCRC32C() { 6785 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6786 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6787 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6788 // no receiver since it is a static method 6789 Node* crc = argument(0); // type: int 6790 Node* src = argument(1); // type: oop 6791 Node* offset = argument(2); // type: int 6792 Node* end = argument(3); // type: int 6793 6794 Node* length = _gvn.transform(new SubINode(end, offset)); 6795 6796 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6797 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6798 // failed array check 6799 return false; 6800 } 6801 6802 // Figure out the size and type of the elements we will be copying. 6803 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6804 if (src_elem != T_BYTE) { 6805 return false; 6806 } 6807 6808 // 'src_start' points to src array + scaled offset 6809 src = must_be_not_null(src, true); 6810 Node* src_start = array_element_address(src, offset, src_elem); 6811 6812 // static final int[] byteTable in class CRC32C 6813 Node* table = get_table_from_crc32c_class(callee()->holder()); 6814 table = must_be_not_null(table, true); 6815 Node* table_start = array_element_address(table, intcon(0), T_INT); 6816 6817 // We assume that range check is done by caller. 6818 // TODO: generate range check (offset+length < src.length) in debug VM. 6819 6820 // Call the stub. 6821 address stubAddr = StubRoutines::updateBytesCRC32C(); 6822 const char *stubName = "updateBytesCRC32C"; 6823 6824 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 6825 stubAddr, stubName, TypePtr::BOTTOM, 6826 crc, src_start, length, table_start); 6827 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6828 set_result(result); 6829 return true; 6830 } 6831 6832 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 6833 // 6834 // Calculate CRC32C for DirectByteBuffer. 6835 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 6836 // 6837 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 6838 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6839 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 6840 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6841 // no receiver since it is a static method 6842 Node* crc = argument(0); // type: int 6843 Node* src = argument(1); // type: long 6844 Node* offset = argument(3); // type: int 6845 Node* end = argument(4); // type: int 6846 6847 Node* length = _gvn.transform(new SubINode(end, offset)); 6848 6849 src = ConvL2X(src); // adjust Java long to machine word 6850 Node* base = _gvn.transform(new CastX2PNode(src)); 6851 offset = ConvI2X(offset); 6852 6853 // 'src_start' points to src array + scaled offset 6854 Node* src_start = basic_plus_adr(top(), base, offset); 6855 6856 // static final int[] byteTable in class CRC32C 6857 Node* table = get_table_from_crc32c_class(callee()->holder()); 6858 table = must_be_not_null(table, true); 6859 Node* table_start = array_element_address(table, intcon(0), T_INT); 6860 6861 // Call the stub. 6862 address stubAddr = StubRoutines::updateBytesCRC32C(); 6863 const char *stubName = "updateBytesCRC32C"; 6864 6865 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 6866 stubAddr, stubName, TypePtr::BOTTOM, 6867 crc, src_start, length, table_start); 6868 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6869 set_result(result); 6870 return true; 6871 } 6872 6873 //------------------------------inline_updateBytesAdler32---------------------- 6874 // 6875 // Calculate Adler32 checksum for byte[] array. 6876 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 6877 // 6878 bool LibraryCallKit::inline_updateBytesAdler32() { 6879 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one 6880 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6881 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 6882 // no receiver since it is static method 6883 Node* crc = argument(0); // type: int 6884 Node* src = argument(1); // type: oop 6885 Node* offset = argument(2); // type: int 6886 Node* length = argument(3); // type: int 6887 6888 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6889 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6890 // failed array check 6891 return false; 6892 } 6893 6894 // Figure out the size and type of the elements we will be copying. 6895 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6896 if (src_elem != T_BYTE) { 6897 return false; 6898 } 6899 6900 // 'src_start' points to src array + scaled offset 6901 Node* src_start = array_element_address(src, offset, src_elem); 6902 6903 // We assume that range check is done by caller. 6904 // TODO: generate range check (offset+length < src.length) in debug VM. 6905 6906 // Call the stub. 6907 address stubAddr = StubRoutines::updateBytesAdler32(); 6908 const char *stubName = "updateBytesAdler32"; 6909 6910 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6911 stubAddr, stubName, TypePtr::BOTTOM, 6912 crc, src_start, length); 6913 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6914 set_result(result); 6915 return true; 6916 } 6917 6918 //------------------------------inline_updateByteBufferAdler32--------------- 6919 // 6920 // Calculate Adler32 checksum for DirectByteBuffer. 6921 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 6922 // 6923 bool LibraryCallKit::inline_updateByteBufferAdler32() { 6924 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one 6925 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 6926 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 6927 // no receiver since it is static method 6928 Node* crc = argument(0); // type: int 6929 Node* src = argument(1); // type: long 6930 Node* offset = argument(3); // type: int 6931 Node* length = argument(4); // type: int 6932 6933 src = ConvL2X(src); // adjust Java long to machine word 6934 Node* base = _gvn.transform(new CastX2PNode(src)); 6935 offset = ConvI2X(offset); 6936 6937 // 'src_start' points to src array + scaled offset 6938 Node* src_start = basic_plus_adr(top(), base, offset); 6939 6940 // Call the stub. 6941 address stubAddr = StubRoutines::updateBytesAdler32(); 6942 const char *stubName = "updateBytesAdler32"; 6943 6944 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6945 stubAddr, stubName, TypePtr::BOTTOM, 6946 crc, src_start, length); 6947 6948 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6949 set_result(result); 6950 return true; 6951 } 6952 6953 //----------------------------inline_reference_get0---------------------------- 6954 // public T java.lang.ref.Reference.get(); 6955 bool LibraryCallKit::inline_reference_get0() { 6956 const int referent_offset = java_lang_ref_Reference::referent_offset(); 6957 6958 // Get the argument: 6959 Node* reference_obj = null_check_receiver(); 6960 if (stopped()) return true; 6961 6962 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF; 6963 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;", 6964 decorators, /*is_static*/ false, nullptr); 6965 if (result == nullptr) return false; 6966 6967 // Add memory barrier to prevent commoning reads from this field 6968 // across safepoint since GC can change its value. 6969 insert_mem_bar(Op_MemBarCPUOrder); 6970 6971 set_result(result); 6972 return true; 6973 } 6974 6975 //----------------------------inline_reference_refersTo0---------------------------- 6976 // bool java.lang.ref.Reference.refersTo0(); 6977 // bool java.lang.ref.PhantomReference.refersTo0(); 6978 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) { 6979 // Get arguments: 6980 Node* reference_obj = null_check_receiver(); 6981 Node* other_obj = argument(1); 6982 if (stopped()) return true; 6983 6984 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE; 6985 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF); 6986 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;", 6987 decorators, /*is_static*/ false, nullptr); 6988 if (referent == nullptr) return false; 6989 6990 // Add memory barrier to prevent commoning reads from this field 6991 // across safepoint since GC can change its value. 6992 insert_mem_bar(Op_MemBarCPUOrder); 6993 6994 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj)); 6995 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 6996 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN); 6997 6998 RegionNode* region = new RegionNode(3); 6999 PhiNode* phi = new PhiNode(region, TypeInt::BOOL); 7000 7001 Node* if_true = _gvn.transform(new IfTrueNode(if_node)); 7002 region->init_req(1, if_true); 7003 phi->init_req(1, intcon(1)); 7004 7005 Node* if_false = _gvn.transform(new IfFalseNode(if_node)); 7006 region->init_req(2, if_false); 7007 phi->init_req(2, intcon(0)); 7008 7009 set_control(_gvn.transform(region)); 7010 record_for_igvn(region); 7011 set_result(_gvn.transform(phi)); 7012 return true; 7013 } 7014 7015 //----------------------------inline_reference_clear0---------------------------- 7016 // void java.lang.ref.Reference.clear0(); 7017 // void java.lang.ref.PhantomReference.clear0(); 7018 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) { 7019 // This matches the implementation in JVM_ReferenceClear, see the comments there. 7020 7021 // Get arguments 7022 Node* reference_obj = null_check_receiver(); 7023 if (stopped()) return true; 7024 7025 // Common access parameters 7026 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE; 7027 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF); 7028 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset()); 7029 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr(); 7030 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass()); 7031 7032 Node* referent = access_load_at(reference_obj, 7033 referent_field_addr, 7034 referent_field_addr_type, 7035 val_type, 7036 T_OBJECT, 7037 decorators); 7038 7039 IdealKit ideal(this); 7040 #define __ ideal. 7041 __ if_then(referent, BoolTest::ne, null()); 7042 sync_kit(ideal); 7043 access_store_at(reference_obj, 7044 referent_field_addr, 7045 referent_field_addr_type, 7046 null(), 7047 val_type, 7048 T_OBJECT, 7049 decorators); 7050 __ sync_kit(this); 7051 __ end_if(); 7052 final_sync(ideal); 7053 #undef __ 7054 7055 return true; 7056 } 7057 7058 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString, 7059 DecoratorSet decorators, bool is_static, 7060 ciInstanceKlass* fromKls) { 7061 if (fromKls == nullptr) { 7062 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 7063 assert(tinst != nullptr, "obj is null"); 7064 assert(tinst->is_loaded(), "obj is not loaded"); 7065 fromKls = tinst->instance_klass(); 7066 } else { 7067 assert(is_static, "only for static field access"); 7068 } 7069 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 7070 ciSymbol::make(fieldTypeString), 7071 is_static); 7072 7073 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName); 7074 if (field == nullptr) return (Node *) nullptr; 7075 7076 if (is_static) { 7077 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 7078 fromObj = makecon(tip); 7079 } 7080 7081 // Next code copied from Parse::do_get_xxx(): 7082 7083 // Compute address and memory type. 7084 int offset = field->offset_in_bytes(); 7085 bool is_vol = field->is_volatile(); 7086 ciType* field_klass = field->type(); 7087 assert(field_klass->is_loaded(), "should be loaded"); 7088 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 7089 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 7090 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()), 7091 "slice of address and input slice don't match"); 7092 BasicType bt = field->layout_type(); 7093 7094 // Build the resultant type of the load 7095 const Type *type; 7096 if (bt == T_OBJECT) { 7097 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 7098 } else { 7099 type = Type::get_const_basic_type(bt); 7100 } 7101 7102 if (is_vol) { 7103 decorators |= MO_SEQ_CST; 7104 } 7105 7106 return access_load_at(fromObj, adr, adr_type, type, bt, decorators); 7107 } 7108 7109 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 7110 bool is_exact /* true */, bool is_static /* false */, 7111 ciInstanceKlass * fromKls /* nullptr */) { 7112 if (fromKls == nullptr) { 7113 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 7114 assert(tinst != nullptr, "obj is null"); 7115 assert(tinst->is_loaded(), "obj is not loaded"); 7116 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 7117 fromKls = tinst->instance_klass(); 7118 } 7119 else { 7120 assert(is_static, "only for static field access"); 7121 } 7122 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 7123 ciSymbol::make(fieldTypeString), 7124 is_static); 7125 7126 assert(field != nullptr, "undefined field"); 7127 assert(!field->is_volatile(), "not defined for volatile fields"); 7128 7129 if (is_static) { 7130 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 7131 fromObj = makecon(tip); 7132 } 7133 7134 // Next code copied from Parse::do_get_xxx(): 7135 7136 // Compute address and memory type. 7137 int offset = field->offset_in_bytes(); 7138 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 7139 7140 return adr; 7141 } 7142 7143 //------------------------------inline_aescrypt_Block----------------------- 7144 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 7145 address stubAddr = nullptr; 7146 const char *stubName; 7147 assert(UseAES, "need AES instruction support"); 7148 7149 switch(id) { 7150 case vmIntrinsics::_aescrypt_encryptBlock: 7151 stubAddr = StubRoutines::aescrypt_encryptBlock(); 7152 stubName = "aescrypt_encryptBlock"; 7153 break; 7154 case vmIntrinsics::_aescrypt_decryptBlock: 7155 stubAddr = StubRoutines::aescrypt_decryptBlock(); 7156 stubName = "aescrypt_decryptBlock"; 7157 break; 7158 default: 7159 break; 7160 } 7161 if (stubAddr == nullptr) return false; 7162 7163 Node* aescrypt_object = argument(0); 7164 Node* src = argument(1); 7165 Node* src_offset = argument(2); 7166 Node* dest = argument(3); 7167 Node* dest_offset = argument(4); 7168 7169 src = must_be_not_null(src, true); 7170 dest = must_be_not_null(dest, true); 7171 7172 // (1) src and dest are arrays. 7173 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7174 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7175 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7176 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7177 7178 // for the quick and dirty code we will skip all the checks. 7179 // we are just trying to get the call to be generated. 7180 Node* src_start = src; 7181 Node* dest_start = dest; 7182 if (src_offset != nullptr || dest_offset != nullptr) { 7183 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7184 src_start = array_element_address(src, src_offset, T_BYTE); 7185 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7186 } 7187 7188 // now need to get the start of its expanded key array 7189 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7190 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7191 if (k_start == nullptr) return false; 7192 7193 // Call the stub. 7194 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 7195 stubAddr, stubName, TypePtr::BOTTOM, 7196 src_start, dest_start, k_start); 7197 7198 return true; 7199 } 7200 7201 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 7202 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 7203 address stubAddr = nullptr; 7204 const char *stubName = nullptr; 7205 7206 assert(UseAES, "need AES instruction support"); 7207 7208 switch(id) { 7209 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 7210 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 7211 stubName = "cipherBlockChaining_encryptAESCrypt"; 7212 break; 7213 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 7214 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 7215 stubName = "cipherBlockChaining_decryptAESCrypt"; 7216 break; 7217 default: 7218 break; 7219 } 7220 if (stubAddr == nullptr) return false; 7221 7222 Node* cipherBlockChaining_object = argument(0); 7223 Node* src = argument(1); 7224 Node* src_offset = argument(2); 7225 Node* len = argument(3); 7226 Node* dest = argument(4); 7227 Node* dest_offset = argument(5); 7228 7229 src = must_be_not_null(src, false); 7230 dest = must_be_not_null(dest, false); 7231 7232 // (1) src and dest are arrays. 7233 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7234 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7235 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7236 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7237 7238 // checks are the responsibility of the caller 7239 Node* src_start = src; 7240 Node* dest_start = dest; 7241 if (src_offset != nullptr || dest_offset != nullptr) { 7242 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7243 src_start = array_element_address(src, src_offset, T_BYTE); 7244 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7245 } 7246 7247 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7248 // (because of the predicated logic executed earlier). 7249 // so we cast it here safely. 7250 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7251 7252 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7253 if (embeddedCipherObj == nullptr) return false; 7254 7255 // cast it to what we know it will be at runtime 7256 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 7257 assert(tinst != nullptr, "CBC obj is null"); 7258 assert(tinst->is_loaded(), "CBC obj is not loaded"); 7259 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7260 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7261 7262 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7263 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7264 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7265 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7266 aescrypt_object = _gvn.transform(aescrypt_object); 7267 7268 // we need to get the start of the aescrypt_object's expanded key array 7269 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7270 if (k_start == nullptr) return false; 7271 7272 // similarly, get the start address of the r vector 7273 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B"); 7274 if (objRvec == nullptr) return false; 7275 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 7276 7277 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7278 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 7279 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 7280 stubAddr, stubName, TypePtr::BOTTOM, 7281 src_start, dest_start, k_start, r_start, len); 7282 7283 // return cipher length (int) 7284 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 7285 set_result(retvalue); 7286 return true; 7287 } 7288 7289 //------------------------------inline_electronicCodeBook_AESCrypt----------------------- 7290 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) { 7291 address stubAddr = nullptr; 7292 const char *stubName = nullptr; 7293 7294 assert(UseAES, "need AES instruction support"); 7295 7296 switch (id) { 7297 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 7298 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt(); 7299 stubName = "electronicCodeBook_encryptAESCrypt"; 7300 break; 7301 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 7302 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt(); 7303 stubName = "electronicCodeBook_decryptAESCrypt"; 7304 break; 7305 default: 7306 break; 7307 } 7308 7309 if (stubAddr == nullptr) return false; 7310 7311 Node* electronicCodeBook_object = argument(0); 7312 Node* src = argument(1); 7313 Node* src_offset = argument(2); 7314 Node* len = argument(3); 7315 Node* dest = argument(4); 7316 Node* dest_offset = argument(5); 7317 7318 // (1) src and dest are arrays. 7319 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7320 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7321 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7322 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7323 7324 // checks are the responsibility of the caller 7325 Node* src_start = src; 7326 Node* dest_start = dest; 7327 if (src_offset != nullptr || dest_offset != nullptr) { 7328 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7329 src_start = array_element_address(src, src_offset, T_BYTE); 7330 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7331 } 7332 7333 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7334 // (because of the predicated logic executed earlier). 7335 // so we cast it here safely. 7336 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7337 7338 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7339 if (embeddedCipherObj == nullptr) return false; 7340 7341 // cast it to what we know it will be at runtime 7342 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr(); 7343 assert(tinst != nullptr, "ECB obj is null"); 7344 assert(tinst->is_loaded(), "ECB obj is not loaded"); 7345 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7346 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7347 7348 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7349 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7350 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7351 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7352 aescrypt_object = _gvn.transform(aescrypt_object); 7353 7354 // we need to get the start of the aescrypt_object's expanded key array 7355 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7356 if (k_start == nullptr) return false; 7357 7358 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7359 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP, 7360 OptoRuntime::electronicCodeBook_aescrypt_Type(), 7361 stubAddr, stubName, TypePtr::BOTTOM, 7362 src_start, dest_start, k_start, len); 7363 7364 // return cipher length (int) 7365 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms)); 7366 set_result(retvalue); 7367 return true; 7368 } 7369 7370 //------------------------------inline_counterMode_AESCrypt----------------------- 7371 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 7372 assert(UseAES, "need AES instruction support"); 7373 if (!UseAESCTRIntrinsics) return false; 7374 7375 address stubAddr = nullptr; 7376 const char *stubName = nullptr; 7377 if (id == vmIntrinsics::_counterMode_AESCrypt) { 7378 stubAddr = StubRoutines::counterMode_AESCrypt(); 7379 stubName = "counterMode_AESCrypt"; 7380 } 7381 if (stubAddr == nullptr) return false; 7382 7383 Node* counterMode_object = argument(0); 7384 Node* src = argument(1); 7385 Node* src_offset = argument(2); 7386 Node* len = argument(3); 7387 Node* dest = argument(4); 7388 Node* dest_offset = argument(5); 7389 7390 // (1) src and dest are arrays. 7391 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7392 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7393 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7394 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7395 7396 // checks are the responsibility of the caller 7397 Node* src_start = src; 7398 Node* dest_start = dest; 7399 if (src_offset != nullptr || dest_offset != nullptr) { 7400 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7401 src_start = array_element_address(src, src_offset, T_BYTE); 7402 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7403 } 7404 7405 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7406 // (because of the predicated logic executed earlier). 7407 // so we cast it here safely. 7408 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7409 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7410 if (embeddedCipherObj == nullptr) return false; 7411 // cast it to what we know it will be at runtime 7412 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 7413 assert(tinst != nullptr, "CTR obj is null"); 7414 assert(tinst->is_loaded(), "CTR obj is not loaded"); 7415 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7416 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7417 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7418 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7419 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7420 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7421 aescrypt_object = _gvn.transform(aescrypt_object); 7422 // we need to get the start of the aescrypt_object's expanded key array 7423 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7424 if (k_start == nullptr) return false; 7425 // similarly, get the start address of the r vector 7426 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B"); 7427 if (obj_counter == nullptr) return false; 7428 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 7429 7430 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B"); 7431 if (saved_encCounter == nullptr) return false; 7432 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 7433 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 7434 7435 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7436 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 7437 OptoRuntime::counterMode_aescrypt_Type(), 7438 stubAddr, stubName, TypePtr::BOTTOM, 7439 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 7440 7441 // return cipher length (int) 7442 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 7443 set_result(retvalue); 7444 return true; 7445 } 7446 7447 //------------------------------get_key_start_from_aescrypt_object----------------------- 7448 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 7449 #if defined(PPC64) || defined(S390) || defined(RISCV64) 7450 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys. 7451 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns. 7452 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption. 7453 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]). 7454 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I"); 7455 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt"); 7456 if (objSessionK == nullptr) { 7457 return (Node *) nullptr; 7458 } 7459 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true); 7460 #else 7461 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I"); 7462 #endif // PPC64 7463 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt"); 7464 if (objAESCryptKey == nullptr) return (Node *) nullptr; 7465 7466 // now have the array, need to get the start address of the K array 7467 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 7468 return k_start; 7469 } 7470 7471 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 7472 // Return node representing slow path of predicate check. 7473 // the pseudo code we want to emulate with this predicate is: 7474 // for encryption: 7475 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7476 // for decryption: 7477 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7478 // note cipher==plain is more conservative than the original java code but that's OK 7479 // 7480 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 7481 // The receiver was checked for null already. 7482 Node* objCBC = argument(0); 7483 7484 Node* src = argument(1); 7485 Node* dest = argument(4); 7486 7487 // Load embeddedCipher field of CipherBlockChaining object. 7488 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7489 7490 // get AESCrypt klass for instanceOf check 7491 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7492 // will have same classloader as CipherBlockChaining object 7493 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 7494 assert(tinst != nullptr, "CBCobj is null"); 7495 assert(tinst->is_loaded(), "CBCobj is not loaded"); 7496 7497 // we want to do an instanceof comparison against the AESCrypt class 7498 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7499 if (!klass_AESCrypt->is_loaded()) { 7500 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7501 Node* ctrl = control(); 7502 set_control(top()); // no regular fast path 7503 return ctrl; 7504 } 7505 7506 src = must_be_not_null(src, true); 7507 dest = must_be_not_null(dest, true); 7508 7509 // Resolve oops to stable for CmpP below. 7510 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7511 7512 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7513 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7514 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7515 7516 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7517 7518 // for encryption, we are done 7519 if (!decrypting) 7520 return instof_false; // even if it is null 7521 7522 // for decryption, we need to add a further check to avoid 7523 // taking the intrinsic path when cipher and plain are the same 7524 // see the original java code for why. 7525 RegionNode* region = new RegionNode(3); 7526 region->init_req(1, instof_false); 7527 7528 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 7529 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 7530 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN); 7531 region->init_req(2, src_dest_conjoint); 7532 7533 record_for_igvn(region); 7534 return _gvn.transform(region); 7535 } 7536 7537 //----------------------------inline_electronicCodeBook_AESCrypt_predicate---------------------------- 7538 // Return node representing slow path of predicate check. 7539 // the pseudo code we want to emulate with this predicate is: 7540 // for encryption: 7541 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7542 // for decryption: 7543 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7544 // note cipher==plain is more conservative than the original java code but that's OK 7545 // 7546 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) { 7547 // The receiver was checked for null already. 7548 Node* objECB = argument(0); 7549 7550 // Load embeddedCipher field of ElectronicCodeBook object. 7551 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7552 7553 // get AESCrypt klass for instanceOf check 7554 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7555 // will have same classloader as ElectronicCodeBook object 7556 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr(); 7557 assert(tinst != nullptr, "ECBobj is null"); 7558 assert(tinst->is_loaded(), "ECBobj is not loaded"); 7559 7560 // we want to do an instanceof comparison against the AESCrypt class 7561 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7562 if (!klass_AESCrypt->is_loaded()) { 7563 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7564 Node* ctrl = control(); 7565 set_control(top()); // no regular fast path 7566 return ctrl; 7567 } 7568 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7569 7570 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7571 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7572 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7573 7574 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7575 7576 // for encryption, we are done 7577 if (!decrypting) 7578 return instof_false; // even if it is null 7579 7580 // for decryption, we need to add a further check to avoid 7581 // taking the intrinsic path when cipher and plain are the same 7582 // see the original java code for why. 7583 RegionNode* region = new RegionNode(3); 7584 region->init_req(1, instof_false); 7585 Node* src = argument(1); 7586 Node* dest = argument(4); 7587 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 7588 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 7589 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN); 7590 region->init_req(2, src_dest_conjoint); 7591 7592 record_for_igvn(region); 7593 return _gvn.transform(region); 7594 } 7595 7596 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 7597 // Return node representing slow path of predicate check. 7598 // the pseudo code we want to emulate with this predicate is: 7599 // for encryption: 7600 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7601 // for decryption: 7602 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7603 // note cipher==plain is more conservative than the original java code but that's OK 7604 // 7605 7606 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 7607 // The receiver was checked for null already. 7608 Node* objCTR = argument(0); 7609 7610 // Load embeddedCipher field of CipherBlockChaining object. 7611 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7612 7613 // get AESCrypt klass for instanceOf check 7614 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7615 // will have same classloader as CipherBlockChaining object 7616 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 7617 assert(tinst != nullptr, "CTRobj is null"); 7618 assert(tinst->is_loaded(), "CTRobj is not loaded"); 7619 7620 // we want to do an instanceof comparison against the AESCrypt class 7621 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7622 if (!klass_AESCrypt->is_loaded()) { 7623 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7624 Node* ctrl = control(); 7625 set_control(top()); // no regular fast path 7626 return ctrl; 7627 } 7628 7629 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7630 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7631 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7632 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7633 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7634 7635 return instof_false; // even if it is null 7636 } 7637 7638 //------------------------------inline_ghash_processBlocks 7639 bool LibraryCallKit::inline_ghash_processBlocks() { 7640 address stubAddr; 7641 const char *stubName; 7642 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 7643 7644 stubAddr = StubRoutines::ghash_processBlocks(); 7645 stubName = "ghash_processBlocks"; 7646 7647 Node* data = argument(0); 7648 Node* offset = argument(1); 7649 Node* len = argument(2); 7650 Node* state = argument(3); 7651 Node* subkeyH = argument(4); 7652 7653 state = must_be_not_null(state, true); 7654 subkeyH = must_be_not_null(subkeyH, true); 7655 data = must_be_not_null(data, true); 7656 7657 Node* state_start = array_element_address(state, intcon(0), T_LONG); 7658 assert(state_start, "state is null"); 7659 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 7660 assert(subkeyH_start, "subkeyH is null"); 7661 Node* data_start = array_element_address(data, offset, T_BYTE); 7662 assert(data_start, "data is null"); 7663 7664 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 7665 OptoRuntime::ghash_processBlocks_Type(), 7666 stubAddr, stubName, TypePtr::BOTTOM, 7667 state_start, subkeyH_start, data_start, len); 7668 return true; 7669 } 7670 7671 //------------------------------inline_chacha20Block 7672 bool LibraryCallKit::inline_chacha20Block() { 7673 address stubAddr; 7674 const char *stubName; 7675 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support"); 7676 7677 stubAddr = StubRoutines::chacha20Block(); 7678 stubName = "chacha20Block"; 7679 7680 Node* state = argument(0); 7681 Node* result = argument(1); 7682 7683 state = must_be_not_null(state, true); 7684 result = must_be_not_null(result, true); 7685 7686 Node* state_start = array_element_address(state, intcon(0), T_INT); 7687 assert(state_start, "state is null"); 7688 Node* result_start = array_element_address(result, intcon(0), T_BYTE); 7689 assert(result_start, "result is null"); 7690 7691 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP, 7692 OptoRuntime::chacha20Block_Type(), 7693 stubAddr, stubName, TypePtr::BOTTOM, 7694 state_start, result_start); 7695 // return key stream length (int) 7696 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms)); 7697 set_result(retvalue); 7698 return true; 7699 } 7700 7701 //------------------------------inline_kyberNtt 7702 bool LibraryCallKit::inline_kyberNtt() { 7703 address stubAddr; 7704 const char *stubName; 7705 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7706 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters"); 7707 7708 stubAddr = StubRoutines::kyberNtt(); 7709 stubName = "kyberNtt"; 7710 if (!stubAddr) return false; 7711 7712 Node* coeffs = argument(0); 7713 Node* ntt_zetas = argument(1); 7714 7715 coeffs = must_be_not_null(coeffs, true); 7716 ntt_zetas = must_be_not_null(ntt_zetas, true); 7717 7718 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT); 7719 assert(coeffs_start, "coeffs is null"); 7720 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT); 7721 assert(ntt_zetas_start, "ntt_zetas is null"); 7722 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 7723 OptoRuntime::kyberNtt_Type(), 7724 stubAddr, stubName, TypePtr::BOTTOM, 7725 coeffs_start, ntt_zetas_start); 7726 // return an int 7727 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms)); 7728 set_result(retvalue); 7729 return true; 7730 } 7731 7732 //------------------------------inline_kyberInverseNtt 7733 bool LibraryCallKit::inline_kyberInverseNtt() { 7734 address stubAddr; 7735 const char *stubName; 7736 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7737 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters"); 7738 7739 stubAddr = StubRoutines::kyberInverseNtt(); 7740 stubName = "kyberInverseNtt"; 7741 if (!stubAddr) return false; 7742 7743 Node* coeffs = argument(0); 7744 Node* zetas = argument(1); 7745 7746 coeffs = must_be_not_null(coeffs, true); 7747 zetas = must_be_not_null(zetas, true); 7748 7749 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT); 7750 assert(coeffs_start, "coeffs is null"); 7751 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT); 7752 assert(zetas_start, "inverseNtt_zetas is null"); 7753 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 7754 OptoRuntime::kyberInverseNtt_Type(), 7755 stubAddr, stubName, TypePtr::BOTTOM, 7756 coeffs_start, zetas_start); 7757 7758 // return an int 7759 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms)); 7760 set_result(retvalue); 7761 return true; 7762 } 7763 7764 //------------------------------inline_kyberNttMult 7765 bool LibraryCallKit::inline_kyberNttMult() { 7766 address stubAddr; 7767 const char *stubName; 7768 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7769 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters"); 7770 7771 stubAddr = StubRoutines::kyberNttMult(); 7772 stubName = "kyberNttMult"; 7773 if (!stubAddr) return false; 7774 7775 Node* result = argument(0); 7776 Node* ntta = argument(1); 7777 Node* nttb = argument(2); 7778 Node* zetas = argument(3); 7779 7780 result = must_be_not_null(result, true); 7781 ntta = must_be_not_null(ntta, true); 7782 nttb = must_be_not_null(nttb, true); 7783 zetas = must_be_not_null(zetas, true); 7784 7785 Node* result_start = array_element_address(result, intcon(0), T_SHORT); 7786 assert(result_start, "result is null"); 7787 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT); 7788 assert(ntta_start, "ntta is null"); 7789 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT); 7790 assert(nttb_start, "nttb is null"); 7791 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT); 7792 assert(zetas_start, "nttMult_zetas is null"); 7793 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP, 7794 OptoRuntime::kyberNttMult_Type(), 7795 stubAddr, stubName, TypePtr::BOTTOM, 7796 result_start, ntta_start, nttb_start, 7797 zetas_start); 7798 7799 // return an int 7800 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms)); 7801 set_result(retvalue); 7802 7803 return true; 7804 } 7805 7806 //------------------------------inline_kyberAddPoly_2 7807 bool LibraryCallKit::inline_kyberAddPoly_2() { 7808 address stubAddr; 7809 const char *stubName; 7810 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7811 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters"); 7812 7813 stubAddr = StubRoutines::kyberAddPoly_2(); 7814 stubName = "kyberAddPoly_2"; 7815 if (!stubAddr) return false; 7816 7817 Node* result = argument(0); 7818 Node* a = argument(1); 7819 Node* b = argument(2); 7820 7821 result = must_be_not_null(result, true); 7822 a = must_be_not_null(a, true); 7823 b = must_be_not_null(b, true); 7824 7825 Node* result_start = array_element_address(result, intcon(0), T_SHORT); 7826 assert(result_start, "result is null"); 7827 Node* a_start = array_element_address(a, intcon(0), T_SHORT); 7828 assert(a_start, "a is null"); 7829 Node* b_start = array_element_address(b, intcon(0), T_SHORT); 7830 assert(b_start, "b is null"); 7831 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP, 7832 OptoRuntime::kyberAddPoly_2_Type(), 7833 stubAddr, stubName, TypePtr::BOTTOM, 7834 result_start, a_start, b_start); 7835 // return an int 7836 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms)); 7837 set_result(retvalue); 7838 return true; 7839 } 7840 7841 //------------------------------inline_kyberAddPoly_3 7842 bool LibraryCallKit::inline_kyberAddPoly_3() { 7843 address stubAddr; 7844 const char *stubName; 7845 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7846 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters"); 7847 7848 stubAddr = StubRoutines::kyberAddPoly_3(); 7849 stubName = "kyberAddPoly_3"; 7850 if (!stubAddr) return false; 7851 7852 Node* result = argument(0); 7853 Node* a = argument(1); 7854 Node* b = argument(2); 7855 Node* c = argument(3); 7856 7857 result = must_be_not_null(result, true); 7858 a = must_be_not_null(a, true); 7859 b = must_be_not_null(b, true); 7860 c = must_be_not_null(c, true); 7861 7862 Node* result_start = array_element_address(result, intcon(0), T_SHORT); 7863 assert(result_start, "result is null"); 7864 Node* a_start = array_element_address(a, intcon(0), T_SHORT); 7865 assert(a_start, "a is null"); 7866 Node* b_start = array_element_address(b, intcon(0), T_SHORT); 7867 assert(b_start, "b is null"); 7868 Node* c_start = array_element_address(c, intcon(0), T_SHORT); 7869 assert(c_start, "c is null"); 7870 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP, 7871 OptoRuntime::kyberAddPoly_3_Type(), 7872 stubAddr, stubName, TypePtr::BOTTOM, 7873 result_start, a_start, b_start, c_start); 7874 // return an int 7875 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms)); 7876 set_result(retvalue); 7877 return true; 7878 } 7879 7880 //------------------------------inline_kyber12To16 7881 bool LibraryCallKit::inline_kyber12To16() { 7882 address stubAddr; 7883 const char *stubName; 7884 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7885 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters"); 7886 7887 stubAddr = StubRoutines::kyber12To16(); 7888 stubName = "kyber12To16"; 7889 if (!stubAddr) return false; 7890 7891 Node* condensed = argument(0); 7892 Node* condensedOffs = argument(1); 7893 Node* parsed = argument(2); 7894 Node* parsedLength = argument(3); 7895 7896 condensed = must_be_not_null(condensed, true); 7897 parsed = must_be_not_null(parsed, true); 7898 7899 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE); 7900 assert(condensed_start, "condensed is null"); 7901 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT); 7902 assert(parsed_start, "parsed is null"); 7903 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP, 7904 OptoRuntime::kyber12To16_Type(), 7905 stubAddr, stubName, TypePtr::BOTTOM, 7906 condensed_start, condensedOffs, parsed_start, parsedLength); 7907 // return an int 7908 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms)); 7909 set_result(retvalue); 7910 return true; 7911 7912 } 7913 7914 //------------------------------inline_kyberBarrettReduce 7915 bool LibraryCallKit::inline_kyberBarrettReduce() { 7916 address stubAddr; 7917 const char *stubName; 7918 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7919 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters"); 7920 7921 stubAddr = StubRoutines::kyberBarrettReduce(); 7922 stubName = "kyberBarrettReduce"; 7923 if (!stubAddr) return false; 7924 7925 Node* coeffs = argument(0); 7926 7927 coeffs = must_be_not_null(coeffs, true); 7928 7929 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT); 7930 assert(coeffs_start, "coeffs is null"); 7931 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP, 7932 OptoRuntime::kyberBarrettReduce_Type(), 7933 stubAddr, stubName, TypePtr::BOTTOM, 7934 coeffs_start); 7935 // return an int 7936 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms)); 7937 set_result(retvalue); 7938 return true; 7939 } 7940 7941 //------------------------------inline_dilithiumAlmostNtt 7942 bool LibraryCallKit::inline_dilithiumAlmostNtt() { 7943 address stubAddr; 7944 const char *stubName; 7945 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 7946 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters"); 7947 7948 stubAddr = StubRoutines::dilithiumAlmostNtt(); 7949 stubName = "dilithiumAlmostNtt"; 7950 if (!stubAddr) return false; 7951 7952 Node* coeffs = argument(0); 7953 Node* ntt_zetas = argument(1); 7954 7955 coeffs = must_be_not_null(coeffs, true); 7956 ntt_zetas = must_be_not_null(ntt_zetas, true); 7957 7958 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT); 7959 assert(coeffs_start, "coeffs is null"); 7960 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT); 7961 assert(ntt_zetas_start, "ntt_zetas is null"); 7962 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 7963 OptoRuntime::dilithiumAlmostNtt_Type(), 7964 stubAddr, stubName, TypePtr::BOTTOM, 7965 coeffs_start, ntt_zetas_start); 7966 // return an int 7967 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms)); 7968 set_result(retvalue); 7969 return true; 7970 } 7971 7972 //------------------------------inline_dilithiumAlmostInverseNtt 7973 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() { 7974 address stubAddr; 7975 const char *stubName; 7976 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 7977 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters"); 7978 7979 stubAddr = StubRoutines::dilithiumAlmostInverseNtt(); 7980 stubName = "dilithiumAlmostInverseNtt"; 7981 if (!stubAddr) return false; 7982 7983 Node* coeffs = argument(0); 7984 Node* zetas = argument(1); 7985 7986 coeffs = must_be_not_null(coeffs, true); 7987 zetas = must_be_not_null(zetas, true); 7988 7989 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT); 7990 assert(coeffs_start, "coeffs is null"); 7991 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT); 7992 assert(zetas_start, "inverseNtt_zetas is null"); 7993 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 7994 OptoRuntime::dilithiumAlmostInverseNtt_Type(), 7995 stubAddr, stubName, TypePtr::BOTTOM, 7996 coeffs_start, zetas_start); 7997 // return an int 7998 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms)); 7999 set_result(retvalue); 8000 return true; 8001 } 8002 8003 //------------------------------inline_dilithiumNttMult 8004 bool LibraryCallKit::inline_dilithiumNttMult() { 8005 address stubAddr; 8006 const char *stubName; 8007 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 8008 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters"); 8009 8010 stubAddr = StubRoutines::dilithiumNttMult(); 8011 stubName = "dilithiumNttMult"; 8012 if (!stubAddr) return false; 8013 8014 Node* result = argument(0); 8015 Node* ntta = argument(1); 8016 Node* nttb = argument(2); 8017 Node* zetas = argument(3); 8018 8019 result = must_be_not_null(result, true); 8020 ntta = must_be_not_null(ntta, true); 8021 nttb = must_be_not_null(nttb, true); 8022 zetas = must_be_not_null(zetas, true); 8023 8024 Node* result_start = array_element_address(result, intcon(0), T_INT); 8025 assert(result_start, "result is null"); 8026 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT); 8027 assert(ntta_start, "ntta is null"); 8028 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT); 8029 assert(nttb_start, "nttb is null"); 8030 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP, 8031 OptoRuntime::dilithiumNttMult_Type(), 8032 stubAddr, stubName, TypePtr::BOTTOM, 8033 result_start, ntta_start, nttb_start); 8034 8035 // return an int 8036 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms)); 8037 set_result(retvalue); 8038 8039 return true; 8040 } 8041 8042 //------------------------------inline_dilithiumMontMulByConstant 8043 bool LibraryCallKit::inline_dilithiumMontMulByConstant() { 8044 address stubAddr; 8045 const char *stubName; 8046 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 8047 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters"); 8048 8049 stubAddr = StubRoutines::dilithiumMontMulByConstant(); 8050 stubName = "dilithiumMontMulByConstant"; 8051 if (!stubAddr) return false; 8052 8053 Node* coeffs = argument(0); 8054 Node* constant = argument(1); 8055 8056 coeffs = must_be_not_null(coeffs, true); 8057 8058 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT); 8059 assert(coeffs_start, "coeffs is null"); 8060 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP, 8061 OptoRuntime::dilithiumMontMulByConstant_Type(), 8062 stubAddr, stubName, TypePtr::BOTTOM, 8063 coeffs_start, constant); 8064 8065 // return an int 8066 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms)); 8067 set_result(retvalue); 8068 return true; 8069 } 8070 8071 8072 //------------------------------inline_dilithiumDecomposePoly 8073 bool LibraryCallKit::inline_dilithiumDecomposePoly() { 8074 address stubAddr; 8075 const char *stubName; 8076 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 8077 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters"); 8078 8079 stubAddr = StubRoutines::dilithiumDecomposePoly(); 8080 stubName = "dilithiumDecomposePoly"; 8081 if (!stubAddr) return false; 8082 8083 Node* input = argument(0); 8084 Node* lowPart = argument(1); 8085 Node* highPart = argument(2); 8086 Node* twoGamma2 = argument(3); 8087 Node* multiplier = argument(4); 8088 8089 input = must_be_not_null(input, true); 8090 lowPart = must_be_not_null(lowPart, true); 8091 highPart = must_be_not_null(highPart, true); 8092 8093 Node* input_start = array_element_address(input, intcon(0), T_INT); 8094 assert(input_start, "input is null"); 8095 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT); 8096 assert(lowPart_start, "lowPart is null"); 8097 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT); 8098 assert(highPart_start, "highPart is null"); 8099 8100 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP, 8101 OptoRuntime::dilithiumDecomposePoly_Type(), 8102 stubAddr, stubName, TypePtr::BOTTOM, 8103 input_start, lowPart_start, highPart_start, 8104 twoGamma2, multiplier); 8105 8106 // return an int 8107 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms)); 8108 set_result(retvalue); 8109 return true; 8110 } 8111 8112 bool LibraryCallKit::inline_base64_encodeBlock() { 8113 address stubAddr; 8114 const char *stubName; 8115 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 8116 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters"); 8117 stubAddr = StubRoutines::base64_encodeBlock(); 8118 stubName = "encodeBlock"; 8119 8120 if (!stubAddr) return false; 8121 Node* base64obj = argument(0); 8122 Node* src = argument(1); 8123 Node* offset = argument(2); 8124 Node* len = argument(3); 8125 Node* dest = argument(4); 8126 Node* dp = argument(5); 8127 Node* isURL = argument(6); 8128 8129 src = must_be_not_null(src, true); 8130 dest = must_be_not_null(dest, true); 8131 8132 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 8133 assert(src_start, "source array is null"); 8134 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 8135 assert(dest_start, "destination array is null"); 8136 8137 Node* base64 = make_runtime_call(RC_LEAF, 8138 OptoRuntime::base64_encodeBlock_Type(), 8139 stubAddr, stubName, TypePtr::BOTTOM, 8140 src_start, offset, len, dest_start, dp, isURL); 8141 return true; 8142 } 8143 8144 bool LibraryCallKit::inline_base64_decodeBlock() { 8145 address stubAddr; 8146 const char *stubName; 8147 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 8148 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters"); 8149 stubAddr = StubRoutines::base64_decodeBlock(); 8150 stubName = "decodeBlock"; 8151 8152 if (!stubAddr) return false; 8153 Node* base64obj = argument(0); 8154 Node* src = argument(1); 8155 Node* src_offset = argument(2); 8156 Node* len = argument(3); 8157 Node* dest = argument(4); 8158 Node* dest_offset = argument(5); 8159 Node* isURL = argument(6); 8160 Node* isMIME = argument(7); 8161 8162 src = must_be_not_null(src, true); 8163 dest = must_be_not_null(dest, true); 8164 8165 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 8166 assert(src_start, "source array is null"); 8167 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 8168 assert(dest_start, "destination array is null"); 8169 8170 Node* call = make_runtime_call(RC_LEAF, 8171 OptoRuntime::base64_decodeBlock_Type(), 8172 stubAddr, stubName, TypePtr::BOTTOM, 8173 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME); 8174 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 8175 set_result(result); 8176 return true; 8177 } 8178 8179 bool LibraryCallKit::inline_poly1305_processBlocks() { 8180 address stubAddr; 8181 const char *stubName; 8182 assert(UsePoly1305Intrinsics, "need Poly intrinsics support"); 8183 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size()); 8184 stubAddr = StubRoutines::poly1305_processBlocks(); 8185 stubName = "poly1305_processBlocks"; 8186 8187 if (!stubAddr) return false; 8188 null_check_receiver(); // null-check receiver 8189 if (stopped()) return true; 8190 8191 Node* input = argument(1); 8192 Node* input_offset = argument(2); 8193 Node* len = argument(3); 8194 Node* alimbs = argument(4); 8195 Node* rlimbs = argument(5); 8196 8197 input = must_be_not_null(input, true); 8198 alimbs = must_be_not_null(alimbs, true); 8199 rlimbs = must_be_not_null(rlimbs, true); 8200 8201 Node* input_start = array_element_address(input, input_offset, T_BYTE); 8202 assert(input_start, "input array is null"); 8203 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG); 8204 assert(acc_start, "acc array is null"); 8205 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG); 8206 assert(r_start, "r array is null"); 8207 8208 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP, 8209 OptoRuntime::poly1305_processBlocks_Type(), 8210 stubAddr, stubName, TypePtr::BOTTOM, 8211 input_start, len, acc_start, r_start); 8212 return true; 8213 } 8214 8215 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() { 8216 address stubAddr; 8217 const char *stubName; 8218 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support"); 8219 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size()); 8220 stubAddr = StubRoutines::intpoly_montgomeryMult_P256(); 8221 stubName = "intpoly_montgomeryMult_P256"; 8222 8223 if (!stubAddr) return false; 8224 null_check_receiver(); // null-check receiver 8225 if (stopped()) return true; 8226 8227 Node* a = argument(1); 8228 Node* b = argument(2); 8229 Node* r = argument(3); 8230 8231 a = must_be_not_null(a, true); 8232 b = must_be_not_null(b, true); 8233 r = must_be_not_null(r, true); 8234 8235 Node* a_start = array_element_address(a, intcon(0), T_LONG); 8236 assert(a_start, "a array is null"); 8237 Node* b_start = array_element_address(b, intcon(0), T_LONG); 8238 assert(b_start, "b array is null"); 8239 Node* r_start = array_element_address(r, intcon(0), T_LONG); 8240 assert(r_start, "r array is null"); 8241 8242 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP, 8243 OptoRuntime::intpoly_montgomeryMult_P256_Type(), 8244 stubAddr, stubName, TypePtr::BOTTOM, 8245 a_start, b_start, r_start); 8246 return true; 8247 } 8248 8249 bool LibraryCallKit::inline_intpoly_assign() { 8250 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support"); 8251 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size()); 8252 const char *stubName = "intpoly_assign"; 8253 address stubAddr = StubRoutines::intpoly_assign(); 8254 if (!stubAddr) return false; 8255 8256 Node* set = argument(0); 8257 Node* a = argument(1); 8258 Node* b = argument(2); 8259 Node* arr_length = load_array_length(a); 8260 8261 a = must_be_not_null(a, true); 8262 b = must_be_not_null(b, true); 8263 8264 Node* a_start = array_element_address(a, intcon(0), T_LONG); 8265 assert(a_start, "a array is null"); 8266 Node* b_start = array_element_address(b, intcon(0), T_LONG); 8267 assert(b_start, "b array is null"); 8268 8269 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP, 8270 OptoRuntime::intpoly_assign_Type(), 8271 stubAddr, stubName, TypePtr::BOTTOM, 8272 set, a_start, b_start, arr_length); 8273 return true; 8274 } 8275 8276 //------------------------------inline_digestBase_implCompress----------------------- 8277 // 8278 // Calculate MD5 for single-block byte[] array. 8279 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs) 8280 // 8281 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 8282 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 8283 // 8284 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 8285 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 8286 // 8287 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 8288 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 8289 // 8290 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array. 8291 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs) 8292 // 8293 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) { 8294 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 8295 8296 Node* digestBase_obj = argument(0); 8297 Node* src = argument(1); // type oop 8298 Node* ofs = argument(2); // type int 8299 8300 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 8301 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 8302 // failed array check 8303 return false; 8304 } 8305 // Figure out the size and type of the elements we will be copying. 8306 BasicType src_elem = src_type->elem()->array_element_basic_type(); 8307 if (src_elem != T_BYTE) { 8308 return false; 8309 } 8310 // 'src_start' points to src array + offset 8311 src = must_be_not_null(src, true); 8312 Node* src_start = array_element_address(src, ofs, src_elem); 8313 Node* state = nullptr; 8314 Node* block_size = nullptr; 8315 address stubAddr; 8316 const char *stubName; 8317 8318 switch(id) { 8319 case vmIntrinsics::_md5_implCompress: 8320 assert(UseMD5Intrinsics, "need MD5 instruction support"); 8321 state = get_state_from_digest_object(digestBase_obj, T_INT); 8322 stubAddr = StubRoutines::md5_implCompress(); 8323 stubName = "md5_implCompress"; 8324 break; 8325 case vmIntrinsics::_sha_implCompress: 8326 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 8327 state = get_state_from_digest_object(digestBase_obj, T_INT); 8328 stubAddr = StubRoutines::sha1_implCompress(); 8329 stubName = "sha1_implCompress"; 8330 break; 8331 case vmIntrinsics::_sha2_implCompress: 8332 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 8333 state = get_state_from_digest_object(digestBase_obj, T_INT); 8334 stubAddr = StubRoutines::sha256_implCompress(); 8335 stubName = "sha256_implCompress"; 8336 break; 8337 case vmIntrinsics::_sha5_implCompress: 8338 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 8339 state = get_state_from_digest_object(digestBase_obj, T_LONG); 8340 stubAddr = StubRoutines::sha512_implCompress(); 8341 stubName = "sha512_implCompress"; 8342 break; 8343 case vmIntrinsics::_sha3_implCompress: 8344 assert(UseSHA3Intrinsics, "need SHA3 instruction support"); 8345 state = get_state_from_digest_object(digestBase_obj, T_LONG); 8346 stubAddr = StubRoutines::sha3_implCompress(); 8347 stubName = "sha3_implCompress"; 8348 block_size = get_block_size_from_digest_object(digestBase_obj); 8349 if (block_size == nullptr) return false; 8350 break; 8351 default: 8352 fatal_unexpected_iid(id); 8353 return false; 8354 } 8355 if (state == nullptr) return false; 8356 8357 assert(stubAddr != nullptr, "Stub %s is not generated", stubName); 8358 if (stubAddr == nullptr) return false; 8359 8360 // Call the stub. 8361 Node* call; 8362 if (block_size == nullptr) { 8363 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false), 8364 stubAddr, stubName, TypePtr::BOTTOM, 8365 src_start, state); 8366 } else { 8367 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true), 8368 stubAddr, stubName, TypePtr::BOTTOM, 8369 src_start, state, block_size); 8370 } 8371 8372 return true; 8373 } 8374 8375 //------------------------------inline_double_keccak 8376 bool LibraryCallKit::inline_double_keccak() { 8377 address stubAddr; 8378 const char *stubName; 8379 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support"); 8380 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters"); 8381 8382 stubAddr = StubRoutines::double_keccak(); 8383 stubName = "double_keccak"; 8384 if (!stubAddr) return false; 8385 8386 Node* status0 = argument(0); 8387 Node* status1 = argument(1); 8388 8389 status0 = must_be_not_null(status0, true); 8390 status1 = must_be_not_null(status1, true); 8391 8392 Node* status0_start = array_element_address(status0, intcon(0), T_LONG); 8393 assert(status0_start, "status0 is null"); 8394 Node* status1_start = array_element_address(status1, intcon(0), T_LONG); 8395 assert(status1_start, "status1 is null"); 8396 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP, 8397 OptoRuntime::double_keccak_Type(), 8398 stubAddr, stubName, TypePtr::BOTTOM, 8399 status0_start, status1_start); 8400 // return an int 8401 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms)); 8402 set_result(retvalue); 8403 return true; 8404 } 8405 8406 8407 //------------------------------inline_digestBase_implCompressMB----------------------- 8408 // 8409 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array. 8410 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 8411 // 8412 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 8413 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics, 8414 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support"); 8415 assert((uint)predicate < 5, "sanity"); 8416 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 8417 8418 Node* digestBase_obj = argument(0); // The receiver was checked for null already. 8419 Node* src = argument(1); // byte[] array 8420 Node* ofs = argument(2); // type int 8421 Node* limit = argument(3); // type int 8422 8423 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 8424 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 8425 // failed array check 8426 return false; 8427 } 8428 // Figure out the size and type of the elements we will be copying. 8429 BasicType src_elem = src_type->elem()->array_element_basic_type(); 8430 if (src_elem != T_BYTE) { 8431 return false; 8432 } 8433 // 'src_start' points to src array + offset 8434 src = must_be_not_null(src, false); 8435 Node* src_start = array_element_address(src, ofs, src_elem); 8436 8437 const char* klass_digestBase_name = nullptr; 8438 const char* stub_name = nullptr; 8439 address stub_addr = nullptr; 8440 BasicType elem_type = T_INT; 8441 8442 switch (predicate) { 8443 case 0: 8444 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) { 8445 klass_digestBase_name = "sun/security/provider/MD5"; 8446 stub_name = "md5_implCompressMB"; 8447 stub_addr = StubRoutines::md5_implCompressMB(); 8448 } 8449 break; 8450 case 1: 8451 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) { 8452 klass_digestBase_name = "sun/security/provider/SHA"; 8453 stub_name = "sha1_implCompressMB"; 8454 stub_addr = StubRoutines::sha1_implCompressMB(); 8455 } 8456 break; 8457 case 2: 8458 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) { 8459 klass_digestBase_name = "sun/security/provider/SHA2"; 8460 stub_name = "sha256_implCompressMB"; 8461 stub_addr = StubRoutines::sha256_implCompressMB(); 8462 } 8463 break; 8464 case 3: 8465 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) { 8466 klass_digestBase_name = "sun/security/provider/SHA5"; 8467 stub_name = "sha512_implCompressMB"; 8468 stub_addr = StubRoutines::sha512_implCompressMB(); 8469 elem_type = T_LONG; 8470 } 8471 break; 8472 case 4: 8473 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) { 8474 klass_digestBase_name = "sun/security/provider/SHA3"; 8475 stub_name = "sha3_implCompressMB"; 8476 stub_addr = StubRoutines::sha3_implCompressMB(); 8477 elem_type = T_LONG; 8478 } 8479 break; 8480 default: 8481 fatal("unknown DigestBase intrinsic predicate: %d", predicate); 8482 } 8483 if (klass_digestBase_name != nullptr) { 8484 assert(stub_addr != nullptr, "Stub is generated"); 8485 if (stub_addr == nullptr) return false; 8486 8487 // get DigestBase klass to lookup for SHA klass 8488 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 8489 assert(tinst != nullptr, "digestBase_obj is not instance???"); 8490 assert(tinst->is_loaded(), "DigestBase is not loaded"); 8491 8492 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name)); 8493 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded"); 8494 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass(); 8495 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit); 8496 } 8497 return false; 8498 } 8499 8500 //------------------------------inline_digestBase_implCompressMB----------------------- 8501 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase, 8502 BasicType elem_type, address stubAddr, const char *stubName, 8503 Node* src_start, Node* ofs, Node* limit) { 8504 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase); 8505 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 8506 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 8507 digest_obj = _gvn.transform(digest_obj); 8508 8509 Node* state = get_state_from_digest_object(digest_obj, elem_type); 8510 if (state == nullptr) return false; 8511 8512 Node* block_size = nullptr; 8513 if (strcmp("sha3_implCompressMB", stubName) == 0) { 8514 block_size = get_block_size_from_digest_object(digest_obj); 8515 if (block_size == nullptr) return false; 8516 } 8517 8518 // Call the stub. 8519 Node* call; 8520 if (block_size == nullptr) { 8521 call = make_runtime_call(RC_LEAF|RC_NO_FP, 8522 OptoRuntime::digestBase_implCompressMB_Type(false), 8523 stubAddr, stubName, TypePtr::BOTTOM, 8524 src_start, state, ofs, limit); 8525 } else { 8526 call = make_runtime_call(RC_LEAF|RC_NO_FP, 8527 OptoRuntime::digestBase_implCompressMB_Type(true), 8528 stubAddr, stubName, TypePtr::BOTTOM, 8529 src_start, state, block_size, ofs, limit); 8530 } 8531 8532 // return ofs (int) 8533 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 8534 set_result(result); 8535 8536 return true; 8537 } 8538 8539 //------------------------------inline_galoisCounterMode_AESCrypt----------------------- 8540 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() { 8541 assert(UseAES, "need AES instruction support"); 8542 address stubAddr = nullptr; 8543 const char *stubName = nullptr; 8544 stubAddr = StubRoutines::galoisCounterMode_AESCrypt(); 8545 stubName = "galoisCounterMode_AESCrypt"; 8546 8547 if (stubAddr == nullptr) return false; 8548 8549 Node* in = argument(0); 8550 Node* inOfs = argument(1); 8551 Node* len = argument(2); 8552 Node* ct = argument(3); 8553 Node* ctOfs = argument(4); 8554 Node* out = argument(5); 8555 Node* outOfs = argument(6); 8556 Node* gctr_object = argument(7); 8557 Node* ghash_object = argument(8); 8558 8559 // (1) in, ct and out are arrays. 8560 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr(); 8561 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr(); 8562 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr(); 8563 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM && 8564 ct_type != nullptr && ct_type->elem() != Type::BOTTOM && 8565 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange"); 8566 8567 // checks are the responsibility of the caller 8568 Node* in_start = in; 8569 Node* ct_start = ct; 8570 Node* out_start = out; 8571 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) { 8572 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, ""); 8573 in_start = array_element_address(in, inOfs, T_BYTE); 8574 ct_start = array_element_address(ct, ctOfs, T_BYTE); 8575 out_start = array_element_address(out, outOfs, T_BYTE); 8576 } 8577 8578 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 8579 // (because of the predicated logic executed earlier). 8580 // so we cast it here safely. 8581 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 8582 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 8583 Node* counter = load_field_from_object(gctr_object, "counter", "[B"); 8584 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J"); 8585 Node* state = load_field_from_object(ghash_object, "state", "[J"); 8586 8587 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) { 8588 return false; 8589 } 8590 // cast it to what we know it will be at runtime 8591 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr(); 8592 assert(tinst != nullptr, "GCTR obj is null"); 8593 assert(tinst->is_loaded(), "GCTR obj is not loaded"); 8594 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 8595 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 8596 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 8597 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 8598 const TypeOopPtr* xtype = aklass->as_instance_type(); 8599 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 8600 aescrypt_object = _gvn.transform(aescrypt_object); 8601 // we need to get the start of the aescrypt_object's expanded key array 8602 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 8603 if (k_start == nullptr) return false; 8604 // similarly, get the start address of the r vector 8605 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE); 8606 Node* state_start = array_element_address(state, intcon(0), T_LONG); 8607 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG); 8608 8609 8610 // Call the stub, passing params 8611 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 8612 OptoRuntime::galoisCounterMode_aescrypt_Type(), 8613 stubAddr, stubName, TypePtr::BOTTOM, 8614 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start); 8615 8616 // return cipher length (int) 8617 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms)); 8618 set_result(retvalue); 8619 8620 return true; 8621 } 8622 8623 //----------------------------inline_galoisCounterMode_AESCrypt_predicate---------------------------- 8624 // Return node representing slow path of predicate check. 8625 // the pseudo code we want to emulate with this predicate is: 8626 // for encryption: 8627 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 8628 // for decryption: 8629 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 8630 // note cipher==plain is more conservative than the original java code but that's OK 8631 // 8632 8633 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() { 8634 // The receiver was checked for null already. 8635 Node* objGCTR = argument(7); 8636 // Load embeddedCipher field of GCTR object. 8637 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 8638 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null"); 8639 8640 // get AESCrypt klass for instanceOf check 8641 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 8642 // will have same classloader as CipherBlockChaining object 8643 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr(); 8644 assert(tinst != nullptr, "GCTR obj is null"); 8645 assert(tinst->is_loaded(), "GCTR obj is not loaded"); 8646 8647 // we want to do an instanceof comparison against the AESCrypt class 8648 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 8649 if (!klass_AESCrypt->is_loaded()) { 8650 // if AESCrypt is not even loaded, we never take the intrinsic fast path 8651 Node* ctrl = control(); 8652 set_control(top()); // no regular fast path 8653 return ctrl; 8654 } 8655 8656 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 8657 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 8658 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 8659 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 8660 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 8661 8662 return instof_false; // even if it is null 8663 } 8664 8665 //------------------------------get_state_from_digest_object----------------------- 8666 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) { 8667 const char* state_type; 8668 switch (elem_type) { 8669 case T_BYTE: state_type = "[B"; break; 8670 case T_INT: state_type = "[I"; break; 8671 case T_LONG: state_type = "[J"; break; 8672 default: ShouldNotReachHere(); 8673 } 8674 Node* digest_state = load_field_from_object(digest_object, "state", state_type); 8675 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3"); 8676 if (digest_state == nullptr) return (Node *) nullptr; 8677 8678 // now have the array, need to get the start address of the state array 8679 Node* state = array_element_address(digest_state, intcon(0), elem_type); 8680 return state; 8681 } 8682 8683 //------------------------------get_block_size_from_sha3_object---------------------------------- 8684 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) { 8685 Node* block_size = load_field_from_object(digest_object, "blockSize", "I"); 8686 assert (block_size != nullptr, "sanity"); 8687 return block_size; 8688 } 8689 8690 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 8691 // Return node representing slow path of predicate check. 8692 // the pseudo code we want to emulate with this predicate is: 8693 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath 8694 // 8695 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 8696 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics, 8697 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support"); 8698 assert((uint)predicate < 5, "sanity"); 8699 8700 // The receiver was checked for null already. 8701 Node* digestBaseObj = argument(0); 8702 8703 // get DigestBase klass for instanceOf check 8704 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 8705 assert(tinst != nullptr, "digestBaseObj is null"); 8706 assert(tinst->is_loaded(), "DigestBase is not loaded"); 8707 8708 const char* klass_name = nullptr; 8709 switch (predicate) { 8710 case 0: 8711 if (UseMD5Intrinsics) { 8712 // we want to do an instanceof comparison against the MD5 class 8713 klass_name = "sun/security/provider/MD5"; 8714 } 8715 break; 8716 case 1: 8717 if (UseSHA1Intrinsics) { 8718 // we want to do an instanceof comparison against the SHA class 8719 klass_name = "sun/security/provider/SHA"; 8720 } 8721 break; 8722 case 2: 8723 if (UseSHA256Intrinsics) { 8724 // we want to do an instanceof comparison against the SHA2 class 8725 klass_name = "sun/security/provider/SHA2"; 8726 } 8727 break; 8728 case 3: 8729 if (UseSHA512Intrinsics) { 8730 // we want to do an instanceof comparison against the SHA5 class 8731 klass_name = "sun/security/provider/SHA5"; 8732 } 8733 break; 8734 case 4: 8735 if (UseSHA3Intrinsics) { 8736 // we want to do an instanceof comparison against the SHA3 class 8737 klass_name = "sun/security/provider/SHA3"; 8738 } 8739 break; 8740 default: 8741 fatal("unknown SHA intrinsic predicate: %d", predicate); 8742 } 8743 8744 ciKlass* klass = nullptr; 8745 if (klass_name != nullptr) { 8746 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name)); 8747 } 8748 if ((klass == nullptr) || !klass->is_loaded()) { 8749 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 8750 Node* ctrl = control(); 8751 set_control(top()); // no intrinsic path 8752 return ctrl; 8753 } 8754 ciInstanceKlass* instklass = klass->as_instance_klass(); 8755 8756 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass))); 8757 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 8758 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 8759 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 8760 8761 return instof_false; // even if it is null 8762 } 8763 8764 //-------------inline_fma----------------------------------- 8765 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) { 8766 Node *a = nullptr; 8767 Node *b = nullptr; 8768 Node *c = nullptr; 8769 Node* result = nullptr; 8770 switch (id) { 8771 case vmIntrinsics::_fmaD: 8772 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each."); 8773 // no receiver since it is static method 8774 a = argument(0); 8775 b = argument(2); 8776 c = argument(4); 8777 result = _gvn.transform(new FmaDNode(a, b, c)); 8778 break; 8779 case vmIntrinsics::_fmaF: 8780 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each."); 8781 a = argument(0); 8782 b = argument(1); 8783 c = argument(2); 8784 result = _gvn.transform(new FmaFNode(a, b, c)); 8785 break; 8786 default: 8787 fatal_unexpected_iid(id); break; 8788 } 8789 set_result(result); 8790 return true; 8791 } 8792 8793 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) { 8794 // argument(0) is receiver 8795 Node* codePoint = argument(1); 8796 Node* n = nullptr; 8797 8798 switch (id) { 8799 case vmIntrinsics::_isDigit : 8800 n = new DigitNode(control(), codePoint); 8801 break; 8802 case vmIntrinsics::_isLowerCase : 8803 n = new LowerCaseNode(control(), codePoint); 8804 break; 8805 case vmIntrinsics::_isUpperCase : 8806 n = new UpperCaseNode(control(), codePoint); 8807 break; 8808 case vmIntrinsics::_isWhitespace : 8809 n = new WhitespaceNode(control(), codePoint); 8810 break; 8811 default: 8812 fatal_unexpected_iid(id); 8813 } 8814 8815 set_result(_gvn.transform(n)); 8816 return true; 8817 } 8818 8819 bool LibraryCallKit::inline_profileBoolean() { 8820 Node* counts = argument(1); 8821 const TypeAryPtr* ary = nullptr; 8822 ciArray* aobj = nullptr; 8823 if (counts->is_Con() 8824 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr 8825 && (aobj = ary->const_oop()->as_array()) != nullptr 8826 && (aobj->length() == 2)) { 8827 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 8828 jint false_cnt = aobj->element_value(0).as_int(); 8829 jint true_cnt = aobj->element_value(1).as_int(); 8830 8831 if (C->log() != nullptr) { 8832 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 8833 false_cnt, true_cnt); 8834 } 8835 8836 if (false_cnt + true_cnt == 0) { 8837 // According to profile, never executed. 8838 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 8839 Deoptimization::Action_reinterpret); 8840 return true; 8841 } 8842 8843 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 8844 // is a number of each value occurrences. 8845 Node* result = argument(0); 8846 if (false_cnt == 0 || true_cnt == 0) { 8847 // According to profile, one value has been never seen. 8848 int expected_val = (false_cnt == 0) ? 1 : 0; 8849 8850 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 8851 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 8852 8853 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 8854 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 8855 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 8856 8857 { // Slow path: uncommon trap for never seen value and then reexecute 8858 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 8859 // the value has been seen at least once. 8860 PreserveJVMState pjvms(this); 8861 PreserveReexecuteState preexecs(this); 8862 jvms()->set_should_reexecute(true); 8863 8864 set_control(slow_path); 8865 set_i_o(i_o()); 8866 8867 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 8868 Deoptimization::Action_reinterpret); 8869 } 8870 // The guard for never seen value enables sharpening of the result and 8871 // returning a constant. It allows to eliminate branches on the same value 8872 // later on. 8873 set_control(fast_path); 8874 result = intcon(expected_val); 8875 } 8876 // Stop profiling. 8877 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 8878 // By replacing method body with profile data (represented as ProfileBooleanNode 8879 // on IR level) we effectively disable profiling. 8880 // It enables full speed execution once optimized code is generated. 8881 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 8882 C->record_for_igvn(profile); 8883 set_result(profile); 8884 return true; 8885 } else { 8886 // Continue profiling. 8887 // Profile data isn't available at the moment. So, execute method's bytecode version. 8888 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 8889 // is compiled and counters aren't available since corresponding MethodHandle 8890 // isn't a compile-time constant. 8891 return false; 8892 } 8893 } 8894 8895 bool LibraryCallKit::inline_isCompileConstant() { 8896 Node* n = argument(0); 8897 set_result(n->is_Con() ? intcon(1) : intcon(0)); 8898 return true; 8899 } 8900 8901 //------------------------------- inline_getObjectSize -------------------------------------- 8902 // 8903 // Calculate the runtime size of the object/array. 8904 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize); 8905 // 8906 bool LibraryCallKit::inline_getObjectSize() { 8907 Node* obj = argument(3); 8908 Node* klass_node = load_object_klass(obj); 8909 8910 jint layout_con = Klass::_lh_neutral_value; 8911 Node* layout_val = get_layout_helper(klass_node, layout_con); 8912 int layout_is_con = (layout_val == nullptr); 8913 8914 if (layout_is_con) { 8915 // Layout helper is constant, can figure out things at compile time. 8916 8917 if (Klass::layout_helper_is_instance(layout_con)) { 8918 // Instance case: layout_con contains the size itself. 8919 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con)); 8920 set_result(size); 8921 } else { 8922 // Array case: size is round(header + element_size*arraylength). 8923 // Since arraylength is different for every array instance, we have to 8924 // compute the whole thing at runtime. 8925 8926 Node* arr_length = load_array_length(obj); 8927 8928 int round_mask = MinObjAlignmentInBytes - 1; 8929 int hsize = Klass::layout_helper_header_size(layout_con); 8930 int eshift = Klass::layout_helper_log2_element_size(layout_con); 8931 8932 if ((round_mask & ~right_n_bits(eshift)) == 0) { 8933 round_mask = 0; // strength-reduce it if it goes away completely 8934 } 8935 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded"); 8936 Node* header_size = intcon(hsize + round_mask); 8937 8938 Node* lengthx = ConvI2X(arr_length); 8939 Node* headerx = ConvI2X(header_size); 8940 8941 Node* abody = lengthx; 8942 if (eshift != 0) { 8943 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift))); 8944 } 8945 Node* size = _gvn.transform( new AddXNode(headerx, abody) ); 8946 if (round_mask != 0) { 8947 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) ); 8948 } 8949 size = ConvX2L(size); 8950 set_result(size); 8951 } 8952 } else { 8953 // Layout helper is not constant, need to test for array-ness at runtime. 8954 8955 enum { _instance_path = 1, _array_path, PATH_LIMIT }; 8956 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 8957 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG); 8958 record_for_igvn(result_reg); 8959 8960 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj); 8961 if (array_ctl != nullptr) { 8962 // Array case: size is round(header + element_size*arraylength). 8963 // Since arraylength is different for every array instance, we have to 8964 // compute the whole thing at runtime. 8965 8966 PreserveJVMState pjvms(this); 8967 set_control(array_ctl); 8968 Node* arr_length = load_array_length(obj); 8969 8970 int round_mask = MinObjAlignmentInBytes - 1; 8971 Node* mask = intcon(round_mask); 8972 8973 Node* hss = intcon(Klass::_lh_header_size_shift); 8974 Node* hsm = intcon(Klass::_lh_header_size_mask); 8975 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss)); 8976 header_size = _gvn.transform(new AndINode(header_size, hsm)); 8977 header_size = _gvn.transform(new AddINode(header_size, mask)); 8978 8979 // There is no need to mask or shift this value. 8980 // The semantics of LShiftINode include an implicit mask to 0x1F. 8981 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place"); 8982 Node* elem_shift = layout_val; 8983 8984 Node* lengthx = ConvI2X(arr_length); 8985 Node* headerx = ConvI2X(header_size); 8986 8987 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift)); 8988 Node* size = _gvn.transform(new AddXNode(headerx, abody)); 8989 if (round_mask != 0) { 8990 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask))); 8991 } 8992 size = ConvX2L(size); 8993 8994 result_reg->init_req(_array_path, control()); 8995 result_val->init_req(_array_path, size); 8996 } 8997 8998 if (!stopped()) { 8999 // Instance case: the layout helper gives us instance size almost directly, 9000 // but we need to mask out the _lh_instance_slow_path_bit. 9001 Node* size = ConvI2X(layout_val); 9002 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit"); 9003 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong)); 9004 size = _gvn.transform(new AndXNode(size, mask)); 9005 size = ConvX2L(size); 9006 9007 result_reg->init_req(_instance_path, control()); 9008 result_val->init_req(_instance_path, size); 9009 } 9010 9011 set_result(result_reg, result_val); 9012 } 9013 9014 return true; 9015 } 9016 9017 //------------------------------- inline_blackhole -------------------------------------- 9018 // 9019 // Make sure all arguments to this node are alive. 9020 // This matches methods that were requested to be blackholed through compile commands. 9021 // 9022 bool LibraryCallKit::inline_blackhole() { 9023 assert(callee()->is_static(), "Should have been checked before: only static methods here"); 9024 assert(callee()->is_empty(), "Should have been checked before: only empty methods here"); 9025 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here"); 9026 9027 // Blackhole node pinches only the control, not memory. This allows 9028 // the blackhole to be pinned in the loop that computes blackholed 9029 // values, but have no other side effects, like breaking the optimizations 9030 // across the blackhole. 9031 9032 Node* bh = _gvn.transform(new BlackholeNode(control())); 9033 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control))); 9034 9035 // Bind call arguments as blackhole arguments to keep them alive 9036 uint nargs = callee()->arg_size(); 9037 for (uint i = 0; i < nargs; i++) { 9038 bh->add_req(argument(i)); 9039 } 9040 9041 return true; 9042 } 9043 9044 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) { 9045 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr(); 9046 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) { 9047 return nullptr; // box klass is not Float16 9048 } 9049 9050 // Null check; get notnull casted pointer 9051 Node* null_ctl = top(); 9052 Node* not_null_box = null_check_oop(box, &null_ctl, true); 9053 // If not_null_box is dead, only null-path is taken 9054 if (stopped()) { 9055 set_control(null_ctl); 9056 return nullptr; 9057 } 9058 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, ""); 9059 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 9060 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes()); 9061 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP); 9062 } 9063 9064 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) { 9065 PreserveReexecuteState preexecs(this); 9066 jvms()->set_should_reexecute(true); 9067 9068 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type(); 9069 Node* klass_node = makecon(klass_type); 9070 Node* box = new_instance(klass_node); 9071 9072 Node* value_field = basic_plus_adr(box, field->offset_in_bytes()); 9073 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr(); 9074 9075 Node* field_store = _gvn.transform(access_store_at(box, 9076 value_field, 9077 value_adr_type, 9078 value, 9079 TypeInt::SHORT, 9080 T_SHORT, 9081 IN_HEAP)); 9082 set_memory(field_store, value_adr_type); 9083 return box; 9084 } 9085 9086 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) { 9087 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) || 9088 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) { 9089 return false; 9090 } 9091 9092 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr(); 9093 if (box_type == nullptr || box_type->const_oop() == nullptr) { 9094 return false; 9095 } 9096 9097 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass(); 9098 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass); 9099 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(), 9100 ciSymbols::short_signature(), 9101 false); 9102 assert(field != nullptr, ""); 9103 9104 // Transformed nodes 9105 Node* fld1 = nullptr; 9106 Node* fld2 = nullptr; 9107 Node* fld3 = nullptr; 9108 switch(num_args) { 9109 case 3: 9110 fld3 = unbox_fp16_value(float16_box_type, field, argument(3)); 9111 if (fld3 == nullptr) { 9112 return false; 9113 } 9114 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3)); 9115 // fall-through 9116 case 2: 9117 fld2 = unbox_fp16_value(float16_box_type, field, argument(2)); 9118 if (fld2 == nullptr) { 9119 return false; 9120 } 9121 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2)); 9122 // fall-through 9123 case 1: 9124 fld1 = unbox_fp16_value(float16_box_type, field, argument(1)); 9125 if (fld1 == nullptr) { 9126 return false; 9127 } 9128 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1)); 9129 break; 9130 default: fatal("Unsupported number of arguments %d", num_args); 9131 } 9132 9133 Node* result = nullptr; 9134 switch (id) { 9135 // Unary operations 9136 case vmIntrinsics::_sqrt_float16: 9137 result = _gvn.transform(new SqrtHFNode(C, control(), fld1)); 9138 break; 9139 // Ternary operations 9140 case vmIntrinsics::_fma_float16: 9141 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3)); 9142 break; 9143 default: 9144 fatal_unexpected_iid(id); 9145 break; 9146 } 9147 result = _gvn.transform(new ReinterpretHF2SNode(result)); 9148 set_result(box_fp16_value(float16_box_type, field, result)); 9149 return true; 9150 } 9151