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::_dsinh: 249 case vmIntrinsics::_dtanh: 250 case vmIntrinsics::_dcbrt: 251 case vmIntrinsics::_dabs: 252 case vmIntrinsics::_fabs: 253 case vmIntrinsics::_iabs: 254 case vmIntrinsics::_labs: 255 case vmIntrinsics::_datan2: 256 case vmIntrinsics::_dsqrt: 257 case vmIntrinsics::_dsqrt_strict: 258 case vmIntrinsics::_dexp: 259 case vmIntrinsics::_dlog: 260 case vmIntrinsics::_dlog10: 261 case vmIntrinsics::_dpow: 262 case vmIntrinsics::_dcopySign: 263 case vmIntrinsics::_fcopySign: 264 case vmIntrinsics::_dsignum: 265 case vmIntrinsics::_roundF: 266 case vmIntrinsics::_roundD: 267 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id()); 268 269 case vmIntrinsics::_notify: 270 case vmIntrinsics::_notifyAll: 271 return inline_notify(intrinsic_id()); 272 273 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 274 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 275 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 276 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 277 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 278 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 279 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 280 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 281 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh(); 282 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh(); 283 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 284 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 285 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 286 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 287 288 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 289 290 case vmIntrinsics::_arraySort: return inline_array_sort(); 291 case vmIntrinsics::_arrayPartition: return inline_array_partition(); 292 293 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 294 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 295 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 296 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 297 298 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 299 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 300 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 301 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 302 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 303 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 304 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U); 305 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L); 306 307 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 308 309 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode(); 310 311 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 312 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 313 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 314 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 315 316 case vmIntrinsics::_compressStringC: 317 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 318 case vmIntrinsics::_inflateStringC: 319 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 320 321 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false); 322 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false); 323 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false); 324 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false); 325 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false); 326 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false); 327 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false); 328 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false); 329 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false); 330 331 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false); 332 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false); 333 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false); 334 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false); 335 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false); 336 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false); 337 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false); 338 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false); 339 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false); 340 341 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false); 342 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false); 343 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false); 344 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false); 345 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false); 346 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false); 347 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false); 348 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false); 349 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false); 350 351 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false); 352 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false); 353 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false); 354 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false); 355 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false); 356 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false); 357 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false); 358 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false); 359 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false); 360 361 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true); 362 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true); 363 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true); 364 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true); 365 366 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true); 367 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true); 368 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true); 369 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true); 370 371 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false); 372 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false); 373 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false); 374 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false); 375 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false); 376 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false); 377 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false); 378 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false); 379 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false); 380 381 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false); 382 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false); 383 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false); 384 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false); 385 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false); 386 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false); 387 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false); 388 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false); 389 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false); 390 391 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false); 392 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false); 393 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false); 394 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false); 395 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false); 396 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false); 397 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false); 398 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false); 399 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false); 400 401 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false); 402 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false); 403 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false); 404 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false); 405 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false); 406 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false); 407 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false); 408 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false); 409 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false); 410 411 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 412 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile); 413 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile); 414 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 415 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 416 417 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 418 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 419 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 420 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile); 421 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed); 422 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire); 423 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release); 424 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile); 425 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed); 426 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire); 427 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release); 428 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile); 429 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 430 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 431 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 432 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile); 433 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 434 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 435 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 436 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile); 437 438 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 439 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 440 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 441 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile); 442 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire); 443 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release); 444 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile); 445 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire); 446 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release); 447 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 448 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 449 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 450 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 451 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 452 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 453 454 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile); 455 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile); 456 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 457 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 458 459 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile); 460 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile); 461 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 462 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 463 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 464 465 case vmIntrinsics::_loadFence: 466 case vmIntrinsics::_storeFence: 467 case vmIntrinsics::_storeStoreFence: 468 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 469 470 case vmIntrinsics::_onSpinWait: return inline_onspinwait(); 471 472 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread(); 473 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 474 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread(); 475 476 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache(); 477 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache(); 478 479 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false); 480 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true); 481 482 #if INCLUDE_JVMTI 483 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()), 484 "notifyJvmtiStart", true, false); 485 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()), 486 "notifyJvmtiEnd", false, true); 487 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()), 488 "notifyJvmtiMount", false, false); 489 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()), 490 "notifyJvmtiUnmount", false, false); 491 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync(); 492 #endif 493 494 #ifdef JFR_HAVE_INTRINSICS 495 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime"); 496 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter(); 497 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit(); 498 #endif 499 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 500 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 501 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0(); 502 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true); 503 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false); 504 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 505 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 506 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory(); 507 case vmIntrinsics::_getLength: return inline_native_getLength(); 508 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 509 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 510 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 511 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 512 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT); 513 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG); 514 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 515 516 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true); 517 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false); 518 519 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 520 521 case vmIntrinsics::_isInstance: 522 case vmIntrinsics::_isHidden: 523 case vmIntrinsics::_getSuperclass: 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 case vmIntrinsics::_shipilev_magic_timestamp: 784 return inline_timestamp(false); 785 case vmIntrinsics::_shipilev_magic_timestamp_serial: 786 return inline_timestamp(true); 787 case vmIntrinsics::_shipilev_magic_sizeOf: 788 return inline_sizeOf(); 789 case vmIntrinsics::_shipilev_magic_addressOf: 790 return inline_addressOf(); 791 792 default: 793 // If you get here, it may be that someone has added a new intrinsic 794 // to the list in vmIntrinsics.hpp without implementing it here. 795 #ifndef PRODUCT 796 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 797 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 798 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id())); 799 } 800 #endif 801 return false; 802 } 803 } 804 805 Node* LibraryCallKit::try_to_predicate(int predicate) { 806 if (!jvms()->has_method()) { 807 // Root JVMState has a null method. 808 assert(map()->memory()->Opcode() == Op_Parm, ""); 809 // Insert the memory aliasing node 810 set_all_memory(reset_memory()); 811 } 812 assert(merged_memory(), ""); 813 814 switch (intrinsic_id()) { 815 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 816 return inline_cipherBlockChaining_AESCrypt_predicate(false); 817 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 818 return inline_cipherBlockChaining_AESCrypt_predicate(true); 819 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 820 return inline_electronicCodeBook_AESCrypt_predicate(false); 821 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 822 return inline_electronicCodeBook_AESCrypt_predicate(true); 823 case vmIntrinsics::_counterMode_AESCrypt: 824 return inline_counterMode_AESCrypt_predicate(); 825 case vmIntrinsics::_digestBase_implCompressMB: 826 return inline_digestBase_implCompressMB_predicate(predicate); 827 case vmIntrinsics::_galoisCounterMode_AESCrypt: 828 return inline_galoisCounterMode_AESCrypt_predicate(); 829 830 default: 831 // If you get here, it may be that someone has added a new intrinsic 832 // to the list in vmIntrinsics.hpp without implementing it here. 833 #ifndef PRODUCT 834 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 835 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 836 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id())); 837 } 838 #endif 839 Node* slow_ctl = control(); 840 set_control(top()); // No fast path intrinsic 841 return slow_ctl; 842 } 843 } 844 845 //------------------------------set_result------------------------------- 846 // Helper function for finishing intrinsics. 847 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 848 record_for_igvn(region); 849 set_control(_gvn.transform(region)); 850 set_result( _gvn.transform(value)); 851 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 852 } 853 854 //------------------------------generate_guard--------------------------- 855 // Helper function for generating guarded fast-slow graph structures. 856 // The given 'test', if true, guards a slow path. If the test fails 857 // then a fast path can be taken. (We generally hope it fails.) 858 // In all cases, GraphKit::control() is updated to the fast path. 859 // The returned value represents the control for the slow path. 860 // The return value is never 'top'; it is either a valid control 861 // or null if it is obvious that the slow path can never be taken. 862 // Also, if region and the slow control are not null, the slow edge 863 // is appended to the region. 864 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 865 if (stopped()) { 866 // Already short circuited. 867 return nullptr; 868 } 869 870 // Build an if node and its projections. 871 // If test is true we take the slow path, which we assume is uncommon. 872 if (_gvn.type(test) == TypeInt::ZERO) { 873 // The slow branch is never taken. No need to build this guard. 874 return nullptr; 875 } 876 877 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 878 879 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 880 if (if_slow == top()) { 881 // The slow branch is never taken. No need to build this guard. 882 return nullptr; 883 } 884 885 if (region != nullptr) 886 region->add_req(if_slow); 887 888 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 889 set_control(if_fast); 890 891 return if_slow; 892 } 893 894 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 895 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 896 } 897 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 898 return generate_guard(test, region, PROB_FAIR); 899 } 900 901 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 902 Node* *pos_index) { 903 if (stopped()) 904 return nullptr; // already stopped 905 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 906 return nullptr; // index is already adequately typed 907 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 908 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 909 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 910 if (is_neg != nullptr && pos_index != nullptr) { 911 // Emulate effect of Parse::adjust_map_after_if. 912 Node* ccast = new CastIINode(control(), index, TypeInt::POS); 913 (*pos_index) = _gvn.transform(ccast); 914 } 915 return is_neg; 916 } 917 918 // Make sure that 'position' is a valid limit index, in [0..length]. 919 // There are two equivalent plans for checking this: 920 // A. (offset + copyLength) unsigned<= arrayLength 921 // B. offset <= (arrayLength - copyLength) 922 // We require that all of the values above, except for the sum and 923 // difference, are already known to be non-negative. 924 // Plan A is robust in the face of overflow, if offset and copyLength 925 // are both hugely positive. 926 // 927 // Plan B is less direct and intuitive, but it does not overflow at 928 // all, since the difference of two non-negatives is always 929 // representable. Whenever Java methods must perform the equivalent 930 // check they generally use Plan B instead of Plan A. 931 // For the moment we use Plan A. 932 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 933 Node* subseq_length, 934 Node* array_length, 935 RegionNode* region) { 936 if (stopped()) 937 return nullptr; // already stopped 938 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 939 if (zero_offset && subseq_length->eqv_uncast(array_length)) 940 return nullptr; // common case of whole-array copy 941 Node* last = subseq_length; 942 if (!zero_offset) // last += offset 943 last = _gvn.transform(new AddINode(last, offset)); 944 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 945 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 946 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 947 return is_over; 948 } 949 950 // Emit range checks for the given String.value byte array 951 void LibraryCallKit::generate_string_range_check(Node* array, 952 Node* offset, 953 Node* count, 954 bool char_count, 955 bool halt_on_oob) { 956 if (stopped()) { 957 return; // already stopped 958 } 959 RegionNode* bailout = new RegionNode(1); 960 record_for_igvn(bailout); 961 if (char_count) { 962 // Convert char count to byte count 963 count = _gvn.transform(new LShiftINode(count, intcon(1))); 964 } 965 966 // Offset and count must not be negative 967 generate_negative_guard(offset, bailout); 968 generate_negative_guard(count, bailout); 969 // Offset + count must not exceed length of array 970 generate_limit_guard(offset, count, load_array_length(array), bailout); 971 972 if (bailout->req() > 1) { 973 if (halt_on_oob) { 974 bailout = _gvn.transform(bailout)->as_Region(); 975 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr)); 976 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic")); 977 C->root()->add_req(halt); 978 } else { 979 PreserveJVMState pjvms(this); 980 set_control(_gvn.transform(bailout)); 981 uncommon_trap(Deoptimization::Reason_intrinsic, 982 Deoptimization::Action_maybe_recompile); 983 } 984 } 985 } 986 987 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset, 988 bool is_immutable) { 989 ciKlass* thread_klass = env()->Thread_klass(); 990 const Type* thread_type 991 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 992 993 Node* thread = _gvn.transform(new ThreadLocalNode()); 994 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset)); 995 tls_output = thread; 996 997 Node* thread_obj_handle 998 = (is_immutable 999 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(), 1000 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered) 1001 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered)); 1002 thread_obj_handle = _gvn.transform(thread_obj_handle); 1003 1004 DecoratorSet decorators = IN_NATIVE; 1005 if (is_immutable) { 1006 decorators |= C2_IMMUTABLE_MEMORY; 1007 } 1008 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators); 1009 } 1010 1011 //--------------------------generate_current_thread-------------------- 1012 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 1013 return current_thread_helper(tls_output, JavaThread::threadObj_offset(), 1014 /*is_immutable*/false); 1015 } 1016 1017 //--------------------------generate_virtual_thread-------------------- 1018 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) { 1019 return current_thread_helper(tls_output, JavaThread::vthread_offset(), 1020 !C->method()->changes_current_thread()); 1021 } 1022 1023 //------------------------------make_string_method_node------------------------ 1024 // Helper method for String intrinsic functions. This version is called with 1025 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 1026 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 1027 // containing the lengths of str1 and str2. 1028 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 1029 Node* result = nullptr; 1030 switch (opcode) { 1031 case Op_StrIndexOf: 1032 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 1033 str1_start, cnt1, str2_start, cnt2, ae); 1034 break; 1035 case Op_StrComp: 1036 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 1037 str1_start, cnt1, str2_start, cnt2, ae); 1038 break; 1039 case Op_StrEquals: 1040 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1041 // Use the constant length if there is one because optimized match rule may exist. 1042 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1043 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1044 break; 1045 default: 1046 ShouldNotReachHere(); 1047 return nullptr; 1048 } 1049 1050 // All these intrinsics have checks. 1051 C->set_has_split_ifs(true); // Has chance for split-if optimization 1052 clear_upper_avx(); 1053 1054 return _gvn.transform(result); 1055 } 1056 1057 //------------------------------inline_string_compareTo------------------------ 1058 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1059 Node* arg1 = argument(0); 1060 Node* arg2 = argument(1); 1061 1062 arg1 = must_be_not_null(arg1, true); 1063 arg2 = must_be_not_null(arg2, true); 1064 1065 // Get start addr and length of first argument 1066 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1067 Node* arg1_cnt = load_array_length(arg1); 1068 1069 // Get start addr and length of second argument 1070 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1071 Node* arg2_cnt = load_array_length(arg2); 1072 1073 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1074 set_result(result); 1075 return true; 1076 } 1077 1078 //------------------------------inline_string_equals------------------------ 1079 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1080 Node* arg1 = argument(0); 1081 Node* arg2 = argument(1); 1082 1083 // paths (plus control) merge 1084 RegionNode* region = new RegionNode(3); 1085 Node* phi = new PhiNode(region, TypeInt::BOOL); 1086 1087 if (!stopped()) { 1088 1089 arg1 = must_be_not_null(arg1, true); 1090 arg2 = must_be_not_null(arg2, true); 1091 1092 // Get start addr and length of first argument 1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1094 Node* arg1_cnt = load_array_length(arg1); 1095 1096 // Get start addr and length of second argument 1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1098 Node* arg2_cnt = load_array_length(arg2); 1099 1100 // Check for arg1_cnt != arg2_cnt 1101 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1102 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1103 Node* if_ne = generate_slow_guard(bol, nullptr); 1104 if (if_ne != nullptr) { 1105 phi->init_req(2, intcon(0)); 1106 region->init_req(2, if_ne); 1107 } 1108 1109 // Check for count == 0 is done by assembler code for StrEquals. 1110 1111 if (!stopped()) { 1112 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1113 phi->init_req(1, equals); 1114 region->init_req(1, control()); 1115 } 1116 } 1117 1118 // post merge 1119 set_control(_gvn.transform(region)); 1120 record_for_igvn(region); 1121 1122 set_result(_gvn.transform(phi)); 1123 return true; 1124 } 1125 1126 //------------------------------inline_array_equals---------------------------- 1127 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1128 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1129 Node* arg1 = argument(0); 1130 Node* arg2 = argument(1); 1131 1132 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1133 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1134 clear_upper_avx(); 1135 1136 return true; 1137 } 1138 1139 1140 //------------------------------inline_countPositives------------------------------ 1141 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len) 1142 bool LibraryCallKit::inline_countPositives() { 1143 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1144 return false; 1145 } 1146 1147 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters"); 1148 // no receiver since it is static method 1149 Node* ba = argument(0); 1150 Node* offset = argument(1); 1151 Node* len = argument(2); 1152 1153 if (VerifyIntrinsicChecks) { 1154 ba = must_be_not_null(ba, true); 1155 generate_string_range_check(ba, offset, len, false, true); 1156 if (stopped()) { 1157 return true; 1158 } 1159 } 1160 1161 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1162 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1163 set_result(_gvn.transform(result)); 1164 clear_upper_avx(); 1165 return true; 1166 } 1167 1168 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) { 1169 Node* index = argument(0); 1170 Node* length = bt == T_INT ? argument(1) : argument(2); 1171 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1172 return false; 1173 } 1174 1175 // check that length is positive 1176 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt)); 1177 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1178 1179 { 1180 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1181 uncommon_trap(Deoptimization::Reason_intrinsic, 1182 Deoptimization::Action_make_not_entrant); 1183 } 1184 1185 if (stopped()) { 1186 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success 1187 return true; 1188 } 1189 1190 // length is now known positive, add a cast node to make this explicit 1191 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long(); 1192 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type( 1193 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), 1194 ConstraintCastNode::RegularDependency, bt); 1195 casted_length = _gvn.transform(casted_length); 1196 replace_in_map(length, casted_length); 1197 length = casted_length; 1198 1199 // Use an unsigned comparison for the range check itself 1200 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true)); 1201 BoolTest::mask btest = BoolTest::lt; 1202 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1203 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1204 _gvn.set_type(rc, rc->Value(&_gvn)); 1205 if (!rc_bool->is_Con()) { 1206 record_for_igvn(rc); 1207 } 1208 set_control(_gvn.transform(new IfTrueNode(rc))); 1209 { 1210 PreserveJVMState pjvms(this); 1211 set_control(_gvn.transform(new IfFalseNode(rc))); 1212 uncommon_trap(Deoptimization::Reason_range_check, 1213 Deoptimization::Action_make_not_entrant); 1214 } 1215 1216 if (stopped()) { 1217 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success 1218 return true; 1219 } 1220 1221 // index is now known to be >= 0 and < length, cast it 1222 Node* result = ConstraintCastNode::make_cast_for_basic_type( 1223 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), 1224 ConstraintCastNode::RegularDependency, bt); 1225 result = _gvn.transform(result); 1226 set_result(result); 1227 replace_in_map(index, result); 1228 return true; 1229 } 1230 1231 //------------------------------inline_string_indexOf------------------------ 1232 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1233 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1234 return false; 1235 } 1236 Node* src = argument(0); 1237 Node* tgt = argument(1); 1238 1239 // Make the merge point 1240 RegionNode* result_rgn = new RegionNode(4); 1241 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 1242 1243 src = must_be_not_null(src, true); 1244 tgt = must_be_not_null(tgt, true); 1245 1246 // Get start addr and length of source string 1247 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1248 Node* src_count = load_array_length(src); 1249 1250 // Get start addr and length of substring 1251 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1252 Node* tgt_count = load_array_length(tgt); 1253 1254 Node* result = nullptr; 1255 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr); 1256 1257 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1258 // Divide src size by 2 if String is UTF16 encoded 1259 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1260 } 1261 if (ae == StrIntrinsicNode::UU) { 1262 // Divide substring size by 2 if String is UTF16 encoded 1263 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1264 } 1265 1266 if (call_opt_stub) { 1267 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(), 1268 StubRoutines::_string_indexof_array[ae], 1269 "stringIndexOf", TypePtr::BOTTOM, src_start, 1270 src_count, tgt_start, tgt_count); 1271 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 1272 } else { 1273 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, 1274 result_rgn, result_phi, ae); 1275 } 1276 if (result != nullptr) { 1277 result_phi->init_req(3, result); 1278 result_rgn->init_req(3, control()); 1279 } 1280 set_control(_gvn.transform(result_rgn)); 1281 record_for_igvn(result_rgn); 1282 set_result(_gvn.transform(result_phi)); 1283 1284 return true; 1285 } 1286 1287 //-----------------------------inline_string_indexOfI----------------------- 1288 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1289 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1290 return false; 1291 } 1292 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1293 return false; 1294 } 1295 1296 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1297 Node* src = argument(0); // byte[] 1298 Node* src_count = argument(1); // char count 1299 Node* tgt = argument(2); // byte[] 1300 Node* tgt_count = argument(3); // char count 1301 Node* from_index = argument(4); // char index 1302 1303 src = must_be_not_null(src, true); 1304 tgt = must_be_not_null(tgt, true); 1305 1306 // Multiply byte array index by 2 if String is UTF16 encoded 1307 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1308 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1309 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1310 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1311 1312 // Range checks 1313 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1314 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1315 if (stopped()) { 1316 return true; 1317 } 1318 1319 RegionNode* region = new RegionNode(5); 1320 Node* phi = new PhiNode(region, TypeInt::INT); 1321 Node* result = nullptr; 1322 1323 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr); 1324 1325 if (call_opt_stub) { 1326 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(), 1327 StubRoutines::_string_indexof_array[ae], 1328 "stringIndexOf", TypePtr::BOTTOM, src_start, 1329 src_count, tgt_start, tgt_count); 1330 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 1331 } else { 1332 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, 1333 region, phi, ae); 1334 } 1335 if (result != nullptr) { 1336 // The result is index relative to from_index if substring was found, -1 otherwise. 1337 // Generate code which will fold into cmove. 1338 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1339 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1340 1341 Node* if_lt = generate_slow_guard(bol, nullptr); 1342 if (if_lt != nullptr) { 1343 // result == -1 1344 phi->init_req(3, result); 1345 region->init_req(3, if_lt); 1346 } 1347 if (!stopped()) { 1348 result = _gvn.transform(new AddINode(result, from_index)); 1349 phi->init_req(4, result); 1350 region->init_req(4, control()); 1351 } 1352 } 1353 1354 set_control(_gvn.transform(region)); 1355 record_for_igvn(region); 1356 set_result(_gvn.transform(phi)); 1357 clear_upper_avx(); 1358 1359 return true; 1360 } 1361 1362 // Create StrIndexOfNode with fast path checks 1363 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1364 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1365 // Check for substr count > string count 1366 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1367 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1368 Node* if_gt = generate_slow_guard(bol, nullptr); 1369 if (if_gt != nullptr) { 1370 phi->init_req(1, intcon(-1)); 1371 region->init_req(1, if_gt); 1372 } 1373 if (!stopped()) { 1374 // Check for substr count == 0 1375 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1376 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1377 Node* if_zero = generate_slow_guard(bol, nullptr); 1378 if (if_zero != nullptr) { 1379 phi->init_req(2, intcon(0)); 1380 region->init_req(2, if_zero); 1381 } 1382 } 1383 if (!stopped()) { 1384 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1385 } 1386 return nullptr; 1387 } 1388 1389 //-----------------------------inline_string_indexOfChar----------------------- 1390 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) { 1391 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1392 return false; 1393 } 1394 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1395 return false; 1396 } 1397 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1398 Node* src = argument(0); // byte[] 1399 Node* int_ch = argument(1); 1400 Node* from_index = argument(2); 1401 Node* max = argument(3); 1402 1403 src = must_be_not_null(src, true); 1404 1405 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1406 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1407 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1408 1409 // Range checks 1410 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U); 1411 1412 // Check for int_ch >= 0 1413 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0))); 1414 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge)); 1415 { 1416 BuildCutout unless(this, int_ch_bol, PROB_MAX); 1417 uncommon_trap(Deoptimization::Reason_intrinsic, 1418 Deoptimization::Action_maybe_recompile); 1419 } 1420 if (stopped()) { 1421 return true; 1422 } 1423 1424 RegionNode* region = new RegionNode(3); 1425 Node* phi = new PhiNode(region, TypeInt::INT); 1426 1427 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae); 1428 C->set_has_split_ifs(true); // Has chance for split-if optimization 1429 _gvn.transform(result); 1430 1431 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1432 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1433 1434 Node* if_lt = generate_slow_guard(bol, nullptr); 1435 if (if_lt != nullptr) { 1436 // result == -1 1437 phi->init_req(2, result); 1438 region->init_req(2, if_lt); 1439 } 1440 if (!stopped()) { 1441 result = _gvn.transform(new AddINode(result, from_index)); 1442 phi->init_req(1, result); 1443 region->init_req(1, control()); 1444 } 1445 set_control(_gvn.transform(region)); 1446 record_for_igvn(region); 1447 set_result(_gvn.transform(phi)); 1448 clear_upper_avx(); 1449 1450 return true; 1451 } 1452 //---------------------------inline_string_copy--------------------- 1453 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1454 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1455 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1456 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1457 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1458 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1459 bool LibraryCallKit::inline_string_copy(bool compress) { 1460 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1461 return false; 1462 } 1463 int nargs = 5; // 2 oops, 3 ints 1464 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1465 1466 Node* src = argument(0); 1467 Node* src_offset = argument(1); 1468 Node* dst = argument(2); 1469 Node* dst_offset = argument(3); 1470 Node* length = argument(4); 1471 1472 // Check for allocation before we add nodes that would confuse 1473 // tightly_coupled_allocation() 1474 AllocateArrayNode* alloc = tightly_coupled_allocation(dst); 1475 1476 // Figure out the size and type of the elements we will be copying. 1477 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 1478 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr(); 1479 if (src_type == nullptr || dst_type == nullptr) { 1480 return false; 1481 } 1482 BasicType src_elem = src_type->elem()->array_element_basic_type(); 1483 BasicType dst_elem = dst_type->elem()->array_element_basic_type(); 1484 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1485 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1486 "Unsupported array types for inline_string_copy"); 1487 1488 src = must_be_not_null(src, true); 1489 dst = must_be_not_null(dst, true); 1490 1491 // Convert char[] offsets to byte[] offsets 1492 bool convert_src = (compress && src_elem == T_BYTE); 1493 bool convert_dst = (!compress && dst_elem == T_BYTE); 1494 if (convert_src) { 1495 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1496 } else if (convert_dst) { 1497 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1498 } 1499 1500 // Range checks 1501 generate_string_range_check(src, src_offset, length, convert_src); 1502 generate_string_range_check(dst, dst_offset, length, convert_dst); 1503 if (stopped()) { 1504 return true; 1505 } 1506 1507 Node* src_start = array_element_address(src, src_offset, src_elem); 1508 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1509 // 'src_start' points to src array + scaled offset 1510 // 'dst_start' points to dst array + scaled offset 1511 Node* count = nullptr; 1512 if (compress) { 1513 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1514 } else { 1515 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1516 } 1517 1518 if (alloc != nullptr) { 1519 if (alloc->maybe_set_complete(&_gvn)) { 1520 // "You break it, you buy it." 1521 InitializeNode* init = alloc->initialization(); 1522 assert(init->is_complete(), "we just did this"); 1523 init->set_complete_with_arraycopy(); 1524 assert(dst->is_CheckCastPP(), "sanity"); 1525 assert(dst->in(0)->in(0) == init, "dest pinned"); 1526 } 1527 // Do not let stores that initialize this object be reordered with 1528 // a subsequent store that would make this object accessible by 1529 // other threads. 1530 // Record what AllocateNode this StoreStore protects so that 1531 // escape analysis can go from the MemBarStoreStoreNode to the 1532 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1533 // based on the escape status of the AllocateNode. 1534 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1535 } 1536 if (compress) { 1537 set_result(_gvn.transform(count)); 1538 } 1539 clear_upper_avx(); 1540 1541 return true; 1542 } 1543 1544 #ifdef _LP64 1545 #define XTOP ,top() /*additional argument*/ 1546 #else //_LP64 1547 #define XTOP /*no additional argument*/ 1548 #endif //_LP64 1549 1550 //------------------------inline_string_toBytesU-------------------------- 1551 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1552 bool LibraryCallKit::inline_string_toBytesU() { 1553 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1554 return false; 1555 } 1556 // Get the arguments. 1557 Node* value = argument(0); 1558 Node* offset = argument(1); 1559 Node* length = argument(2); 1560 1561 Node* newcopy = nullptr; 1562 1563 // Set the original stack and the reexecute bit for the interpreter to reexecute 1564 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1565 { PreserveReexecuteState preexecs(this); 1566 jvms()->set_should_reexecute(true); 1567 1568 // Check if a null path was taken unconditionally. 1569 value = null_check(value); 1570 1571 RegionNode* bailout = new RegionNode(1); 1572 record_for_igvn(bailout); 1573 1574 // Range checks 1575 generate_negative_guard(offset, bailout); 1576 generate_negative_guard(length, bailout); 1577 generate_limit_guard(offset, length, load_array_length(value), bailout); 1578 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1579 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1580 1581 if (bailout->req() > 1) { 1582 PreserveJVMState pjvms(this); 1583 set_control(_gvn.transform(bailout)); 1584 uncommon_trap(Deoptimization::Reason_intrinsic, 1585 Deoptimization::Action_maybe_recompile); 1586 } 1587 if (stopped()) { 1588 return true; 1589 } 1590 1591 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1592 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1593 newcopy = new_array(klass_node, size, 0); // no arguments to push 1594 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy); 1595 guarantee(alloc != nullptr, "created above"); 1596 1597 // Calculate starting addresses. 1598 Node* src_start = array_element_address(value, offset, T_CHAR); 1599 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1600 1601 // Check if dst array address is aligned to HeapWordSize 1602 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0); 1603 // If true, then check if src array address is aligned to HeapWordSize 1604 if (aligned) { 1605 const TypeInt* toffset = gvn().type(offset)->is_int(); 1606 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + 1607 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1608 } 1609 1610 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1611 const char* copyfunc_name = "arraycopy"; 1612 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1613 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1614 OptoRuntime::fast_arraycopy_Type(), 1615 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1616 src_start, dst_start, ConvI2X(length) XTOP); 1617 // Do not let reads from the cloned object float above the arraycopy. 1618 if (alloc->maybe_set_complete(&_gvn)) { 1619 // "You break it, you buy it." 1620 InitializeNode* init = alloc->initialization(); 1621 assert(init->is_complete(), "we just did this"); 1622 init->set_complete_with_arraycopy(); 1623 assert(newcopy->is_CheckCastPP(), "sanity"); 1624 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1625 } 1626 // Do not let stores that initialize this object be reordered with 1627 // a subsequent store that would make this object accessible by 1628 // other threads. 1629 // Record what AllocateNode this StoreStore protects so that 1630 // escape analysis can go from the MemBarStoreStoreNode to the 1631 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1632 // based on the escape status of the AllocateNode. 1633 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1634 } // original reexecute is set back here 1635 1636 C->set_has_split_ifs(true); // Has chance for split-if optimization 1637 if (!stopped()) { 1638 set_result(newcopy); 1639 } 1640 clear_upper_avx(); 1641 1642 return true; 1643 } 1644 1645 //------------------------inline_string_getCharsU-------------------------- 1646 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1647 bool LibraryCallKit::inline_string_getCharsU() { 1648 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1649 return false; 1650 } 1651 1652 // Get the arguments. 1653 Node* src = argument(0); 1654 Node* src_begin = argument(1); 1655 Node* src_end = argument(2); // exclusive offset (i < src_end) 1656 Node* dst = argument(3); 1657 Node* dst_begin = argument(4); 1658 1659 // Check for allocation before we add nodes that would confuse 1660 // tightly_coupled_allocation() 1661 AllocateArrayNode* alloc = tightly_coupled_allocation(dst); 1662 1663 // Check if a null path was taken unconditionally. 1664 src = null_check(src); 1665 dst = null_check(dst); 1666 if (stopped()) { 1667 return true; 1668 } 1669 1670 // Get length and convert char[] offset to byte[] offset 1671 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1672 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1673 1674 // Range checks 1675 generate_string_range_check(src, src_begin, length, true); 1676 generate_string_range_check(dst, dst_begin, length, false); 1677 if (stopped()) { 1678 return true; 1679 } 1680 1681 if (!stopped()) { 1682 // Calculate starting addresses. 1683 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1684 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1685 1686 // Check if array addresses are aligned to HeapWordSize 1687 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1688 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1689 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1690 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1691 1692 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1693 const char* copyfunc_name = "arraycopy"; 1694 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1695 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1696 OptoRuntime::fast_arraycopy_Type(), 1697 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1698 src_start, dst_start, ConvI2X(length) XTOP); 1699 // Do not let reads from the cloned object float above the arraycopy. 1700 if (alloc != nullptr) { 1701 if (alloc->maybe_set_complete(&_gvn)) { 1702 // "You break it, you buy it." 1703 InitializeNode* init = alloc->initialization(); 1704 assert(init->is_complete(), "we just did this"); 1705 init->set_complete_with_arraycopy(); 1706 assert(dst->is_CheckCastPP(), "sanity"); 1707 assert(dst->in(0)->in(0) == init, "dest pinned"); 1708 } 1709 // Do not let stores that initialize this object be reordered with 1710 // a subsequent store that would make this object accessible by 1711 // other threads. 1712 // Record what AllocateNode this StoreStore protects so that 1713 // escape analysis can go from the MemBarStoreStoreNode to the 1714 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1715 // based on the escape status of the AllocateNode. 1716 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1717 } else { 1718 insert_mem_bar(Op_MemBarCPUOrder); 1719 } 1720 } 1721 1722 C->set_has_split_ifs(true); // Has chance for split-if optimization 1723 return true; 1724 } 1725 1726 //----------------------inline_string_char_access---------------------------- 1727 // Store/Load char to/from byte[] array. 1728 // static void StringUTF16.putChar(byte[] val, int index, int c) 1729 // static char StringUTF16.getChar(byte[] val, int index) 1730 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1731 Node* value = argument(0); 1732 Node* index = argument(1); 1733 Node* ch = is_store ? argument(2) : nullptr; 1734 1735 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1736 // correctly requires matched array shapes. 1737 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1738 "sanity: byte[] and char[] bases agree"); 1739 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1740 "sanity: byte[] and char[] scales agree"); 1741 1742 // Bail when getChar over constants is requested: constant folding would 1743 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1744 // Java method would constant fold nicely instead. 1745 if (!is_store && value->is_Con() && index->is_Con()) { 1746 return false; 1747 } 1748 1749 // Save state and restore on bailout 1750 SavedState old_state(this); 1751 1752 value = must_be_not_null(value, true); 1753 1754 Node* adr = array_element_address(value, index, T_CHAR); 1755 if (adr->is_top()) { 1756 return false; 1757 } 1758 old_state.discard(); 1759 if (is_store) { 1760 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED); 1761 } else { 1762 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); 1763 set_result(ch); 1764 } 1765 return true; 1766 } 1767 1768 1769 //------------------------------inline_math----------------------------------- 1770 // public static double Math.abs(double) 1771 // public static double Math.sqrt(double) 1772 // public static double Math.log(double) 1773 // public static double Math.log10(double) 1774 // public static double Math.round(double) 1775 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) { 1776 Node* arg = argument(0); 1777 Node* n = nullptr; 1778 switch (id) { 1779 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1780 case vmIntrinsics::_dsqrt: 1781 case vmIntrinsics::_dsqrt_strict: 1782 n = new SqrtDNode(C, control(), arg); break; 1783 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break; 1784 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break; 1785 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break; 1786 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break; 1787 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break; 1788 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break; 1789 default: fatal_unexpected_iid(id); break; 1790 } 1791 set_result(_gvn.transform(n)); 1792 return true; 1793 } 1794 1795 //------------------------------inline_math----------------------------------- 1796 // public static float Math.abs(float) 1797 // public static int Math.abs(int) 1798 // public static long Math.abs(long) 1799 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1800 Node* arg = argument(0); 1801 Node* n = nullptr; 1802 switch (id) { 1803 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break; 1804 case vmIntrinsics::_iabs: n = new AbsINode( arg); break; 1805 case vmIntrinsics::_labs: n = new AbsLNode( arg); break; 1806 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break; 1807 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break; 1808 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break; 1809 default: fatal_unexpected_iid(id); break; 1810 } 1811 set_result(_gvn.transform(n)); 1812 return true; 1813 } 1814 1815 //------------------------------runtime_math----------------------------- 1816 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1817 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1818 "must be (DD)D or (D)D type"); 1819 1820 // Inputs 1821 Node* a = argument(0); 1822 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr; 1823 1824 const TypePtr* no_memory_effects = nullptr; 1825 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName, 1826 no_memory_effects, 1827 a, top(), b, b ? top() : nullptr); 1828 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1829 #ifdef ASSERT 1830 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1831 assert(value_top == top(), "second value must be top"); 1832 #endif 1833 1834 set_result(value); 1835 return true; 1836 } 1837 1838 //------------------------------inline_math_pow----------------------------- 1839 bool LibraryCallKit::inline_math_pow() { 1840 Node* exp = argument(2); 1841 const TypeD* d = _gvn.type(exp)->isa_double_constant(); 1842 if (d != nullptr) { 1843 if (d->getd() == 2.0) { 1844 // Special case: pow(x, 2.0) => x * x 1845 Node* base = argument(0); 1846 set_result(_gvn.transform(new MulDNode(base, base))); 1847 return true; 1848 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) { 1849 // Special case: pow(x, 0.5) => sqrt(x) 1850 Node* base = argument(0); 1851 Node* zero = _gvn.zerocon(T_DOUBLE); 1852 1853 RegionNode* region = new RegionNode(3); 1854 Node* phi = new PhiNode(region, Type::DOUBLE); 1855 1856 Node* cmp = _gvn.transform(new CmpDNode(base, zero)); 1857 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0. 1858 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0). 1859 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0. 1860 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le)); 1861 1862 Node* if_pow = generate_slow_guard(test, nullptr); 1863 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base)); 1864 phi->init_req(1, value_sqrt); 1865 region->init_req(1, control()); 1866 1867 if (if_pow != nullptr) { 1868 set_control(if_pow); 1869 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() : 1870 CAST_FROM_FN_PTR(address, SharedRuntime::dpow); 1871 const TypePtr* no_memory_effects = nullptr; 1872 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW", 1873 no_memory_effects, base, top(), exp, top()); 1874 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1875 #ifdef ASSERT 1876 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1877 assert(value_top == top(), "second value must be top"); 1878 #endif 1879 phi->init_req(2, value_pow); 1880 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control))); 1881 } 1882 1883 C->set_has_split_ifs(true); // Has chance for split-if optimization 1884 set_control(_gvn.transform(region)); 1885 record_for_igvn(region); 1886 set_result(_gvn.transform(phi)); 1887 1888 return true; 1889 } 1890 } 1891 1892 return StubRoutines::dpow() != nullptr ? 1893 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1894 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW"); 1895 } 1896 1897 //------------------------------inline_math_native----------------------------- 1898 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1899 switch (id) { 1900 case vmIntrinsics::_dsin: 1901 return StubRoutines::dsin() != nullptr ? 1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN"); 1904 case vmIntrinsics::_dcos: 1905 return StubRoutines::dcos() != nullptr ? 1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1907 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS"); 1908 case vmIntrinsics::_dtan: 1909 return StubRoutines::dtan() != nullptr ? 1910 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") : 1911 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN"); 1912 case vmIntrinsics::_dsinh: 1913 return StubRoutines::dsinh() != nullptr ? 1914 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false; 1915 case vmIntrinsics::_dtanh: 1916 return StubRoutines::dtanh() != nullptr ? 1917 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false; 1918 case vmIntrinsics::_dcbrt: 1919 return StubRoutines::dcbrt() != nullptr ? 1920 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false; 1921 case vmIntrinsics::_dexp: 1922 return StubRoutines::dexp() != nullptr ? 1923 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1924 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP"); 1925 case vmIntrinsics::_dlog: 1926 return StubRoutines::dlog() != nullptr ? 1927 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1928 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG"); 1929 case vmIntrinsics::_dlog10: 1930 return StubRoutines::dlog10() != nullptr ? 1931 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") : 1932 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10"); 1933 1934 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false; 1935 case vmIntrinsics::_ceil: 1936 case vmIntrinsics::_floor: 1937 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false; 1938 1939 case vmIntrinsics::_dsqrt: 1940 case vmIntrinsics::_dsqrt_strict: 1941 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false; 1942 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false; 1943 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false; 1944 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false; 1945 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false; 1946 1947 case vmIntrinsics::_dpow: return inline_math_pow(); 1948 case vmIntrinsics::_dcopySign: return inline_double_math(id); 1949 case vmIntrinsics::_fcopySign: return inline_math(id); 1950 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false; 1951 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false; 1952 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false; 1953 1954 // These intrinsics are not yet correctly implemented 1955 case vmIntrinsics::_datan2: 1956 return false; 1957 1958 default: 1959 fatal_unexpected_iid(id); 1960 return false; 1961 } 1962 } 1963 1964 //----------------------------inline_notify-----------------------------------* 1965 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1966 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1967 address func; 1968 if (id == vmIntrinsics::_notify) { 1969 func = OptoRuntime::monitor_notify_Java(); 1970 } else { 1971 func = OptoRuntime::monitor_notifyAll_Java(); 1972 } 1973 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0)); 1974 make_slow_call_ex(call, env()->Throwable_klass(), false); 1975 return true; 1976 } 1977 1978 1979 //----------------------------inline_min_max----------------------------------- 1980 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1981 Node* a = nullptr; 1982 Node* b = nullptr; 1983 Node* n = nullptr; 1984 switch (id) { 1985 case vmIntrinsics::_min: 1986 case vmIntrinsics::_max: 1987 case vmIntrinsics::_minF: 1988 case vmIntrinsics::_maxF: 1989 case vmIntrinsics::_minF_strict: 1990 case vmIntrinsics::_maxF_strict: 1991 case vmIntrinsics::_min_strict: 1992 case vmIntrinsics::_max_strict: 1993 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each."); 1994 a = argument(0); 1995 b = argument(1); 1996 break; 1997 case vmIntrinsics::_minD: 1998 case vmIntrinsics::_maxD: 1999 case vmIntrinsics::_minD_strict: 2000 case vmIntrinsics::_maxD_strict: 2001 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each."); 2002 a = argument(0); 2003 b = argument(2); 2004 break; 2005 case vmIntrinsics::_minL: 2006 case vmIntrinsics::_maxL: 2007 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each."); 2008 a = argument(0); 2009 b = argument(2); 2010 break; 2011 default: 2012 fatal_unexpected_iid(id); 2013 break; 2014 } 2015 2016 switch (id) { 2017 case vmIntrinsics::_min: 2018 case vmIntrinsics::_min_strict: 2019 n = new MinINode(a, b); 2020 break; 2021 case vmIntrinsics::_max: 2022 case vmIntrinsics::_max_strict: 2023 n = new MaxINode(a, b); 2024 break; 2025 case vmIntrinsics::_minF: 2026 case vmIntrinsics::_minF_strict: 2027 n = new MinFNode(a, b); 2028 break; 2029 case vmIntrinsics::_maxF: 2030 case vmIntrinsics::_maxF_strict: 2031 n = new MaxFNode(a, b); 2032 break; 2033 case vmIntrinsics::_minD: 2034 case vmIntrinsics::_minD_strict: 2035 n = new MinDNode(a, b); 2036 break; 2037 case vmIntrinsics::_maxD: 2038 case vmIntrinsics::_maxD_strict: 2039 n = new MaxDNode(a, b); 2040 break; 2041 case vmIntrinsics::_minL: 2042 n = new MinLNode(_gvn.C, a, b); 2043 break; 2044 case vmIntrinsics::_maxL: 2045 n = new MaxLNode(_gvn.C, a, b); 2046 break; 2047 default: 2048 fatal_unexpected_iid(id); 2049 break; 2050 } 2051 2052 set_result(_gvn.transform(n)); 2053 return true; 2054 } 2055 2056 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) { 2057 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic, 2058 env()->ArithmeticException_instance())) { 2059 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces), 2060 // so let's bail out intrinsic rather than risking deopting again. 2061 return false; 2062 } 2063 2064 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 2065 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 2066 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 2067 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 2068 2069 { 2070 PreserveJVMState pjvms(this); 2071 PreserveReexecuteState preexecs(this); 2072 jvms()->set_should_reexecute(true); 2073 2074 set_control(slow_path); 2075 set_i_o(i_o()); 2076 2077 builtin_throw(Deoptimization::Reason_intrinsic, 2078 env()->ArithmeticException_instance(), 2079 /*allow_too_many_traps*/ false); 2080 } 2081 2082 set_control(fast_path); 2083 set_result(math); 2084 return true; 2085 } 2086 2087 template <typename OverflowOp> 2088 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 2089 typedef typename OverflowOp::MathOp MathOp; 2090 2091 MathOp* mathOp = new MathOp(arg1, arg2); 2092 Node* operation = _gvn.transform( mathOp ); 2093 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 2094 return inline_math_mathExact(operation, ofcheck); 2095 } 2096 2097 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 2098 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 2099 } 2100 2101 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 2102 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 2103 } 2104 2105 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 2106 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 2107 } 2108 2109 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 2110 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 2111 } 2112 2113 bool LibraryCallKit::inline_math_negateExactI() { 2114 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 2115 } 2116 2117 bool LibraryCallKit::inline_math_negateExactL() { 2118 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 2119 } 2120 2121 bool LibraryCallKit::inline_math_multiplyExactI() { 2122 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 2123 } 2124 2125 bool LibraryCallKit::inline_math_multiplyExactL() { 2126 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 2127 } 2128 2129 bool LibraryCallKit::inline_math_multiplyHigh() { 2130 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2)))); 2131 return true; 2132 } 2133 2134 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() { 2135 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2)))); 2136 return true; 2137 } 2138 2139 inline int 2140 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) { 2141 const TypePtr* base_type = TypePtr::NULL_PTR; 2142 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr(); 2143 if (base_type == nullptr) { 2144 // Unknown type. 2145 return Type::AnyPtr; 2146 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) { 2147 // Since this is a null+long form, we have to switch to a rawptr. 2148 base = _gvn.transform(new CastX2PNode(offset)); 2149 offset = MakeConX(0); 2150 return Type::RawPtr; 2151 } else if (base_type->base() == Type::RawPtr) { 2152 return Type::RawPtr; 2153 } else if (base_type->isa_oopptr()) { 2154 // Base is never null => always a heap address. 2155 if (!TypePtr::NULL_PTR->higher_equal(base_type)) { 2156 return Type::OopPtr; 2157 } 2158 // Offset is small => always a heap address. 2159 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2160 if (offset_type != nullptr && 2161 base_type->offset() == 0 && // (should always be?) 2162 offset_type->_lo >= 0 && 2163 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2164 return Type::OopPtr; 2165 } else if (type == T_OBJECT) { 2166 // off heap access to an oop doesn't make any sense. Has to be on 2167 // heap. 2168 return Type::OopPtr; 2169 } 2170 // Otherwise, it might either be oop+off or null+addr. 2171 return Type::AnyPtr; 2172 } else { 2173 // No information: 2174 return Type::AnyPtr; 2175 } 2176 } 2177 2178 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) { 2179 Node* uncasted_base = base; 2180 int kind = classify_unsafe_addr(uncasted_base, offset, type); 2181 if (kind == Type::RawPtr) { 2182 return basic_plus_adr(top(), uncasted_base, offset); 2183 } else if (kind == Type::AnyPtr) { 2184 assert(base == uncasted_base, "unexpected base change"); 2185 if (can_cast) { 2186 if (!_gvn.type(base)->speculative_maybe_null() && 2187 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 2188 // According to profiling, this access is always on 2189 // heap. Casting the base to not null and thus avoiding membars 2190 // around the access should allow better optimizations 2191 Node* null_ctl = top(); 2192 base = null_check_oop(base, &null_ctl, true, true, true); 2193 assert(null_ctl->is_top(), "no null control here"); 2194 return basic_plus_adr(base, offset); 2195 } else if (_gvn.type(base)->speculative_always_null() && 2196 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 2197 // According to profiling, this access is always off 2198 // heap. 2199 base = null_assert(base); 2200 Node* raw_base = _gvn.transform(new CastX2PNode(offset)); 2201 offset = MakeConX(0); 2202 return basic_plus_adr(top(), raw_base, offset); 2203 } 2204 } 2205 // We don't know if it's an on heap or off heap access. Fall back 2206 // to raw memory access. 2207 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM)); 2208 return basic_plus_adr(top(), raw, offset); 2209 } else { 2210 assert(base == uncasted_base, "unexpected base change"); 2211 // We know it's an on heap access so base can't be null 2212 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) { 2213 base = must_be_not_null(base, true); 2214 } 2215 return basic_plus_adr(base, offset); 2216 } 2217 } 2218 2219 //--------------------------inline_number_methods----------------------------- 2220 // inline int Integer.numberOfLeadingZeros(int) 2221 // inline int Long.numberOfLeadingZeros(long) 2222 // 2223 // inline int Integer.numberOfTrailingZeros(int) 2224 // inline int Long.numberOfTrailingZeros(long) 2225 // 2226 // inline int Integer.bitCount(int) 2227 // inline int Long.bitCount(long) 2228 // 2229 // inline char Character.reverseBytes(char) 2230 // inline short Short.reverseBytes(short) 2231 // inline int Integer.reverseBytes(int) 2232 // inline long Long.reverseBytes(long) 2233 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2234 Node* arg = argument(0); 2235 Node* n = nullptr; 2236 switch (id) { 2237 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2238 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2239 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2240 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2241 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2242 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2243 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break; 2244 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break; 2245 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break; 2246 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break; 2247 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break; 2248 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break; 2249 default: fatal_unexpected_iid(id); break; 2250 } 2251 set_result(_gvn.transform(n)); 2252 return true; 2253 } 2254 2255 //--------------------------inline_bitshuffle_methods----------------------------- 2256 // inline int Integer.compress(int, int) 2257 // inline int Integer.expand(int, int) 2258 // inline long Long.compress(long, long) 2259 // inline long Long.expand(long, long) 2260 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) { 2261 Node* n = nullptr; 2262 switch (id) { 2263 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break; 2264 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break; 2265 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break; 2266 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break; 2267 default: fatal_unexpected_iid(id); break; 2268 } 2269 set_result(_gvn.transform(n)); 2270 return true; 2271 } 2272 2273 //--------------------------inline_number_methods----------------------------- 2274 // inline int Integer.compareUnsigned(int, int) 2275 // inline int Long.compareUnsigned(long, long) 2276 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) { 2277 Node* arg1 = argument(0); 2278 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1); 2279 Node* n = nullptr; 2280 switch (id) { 2281 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break; 2282 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break; 2283 default: fatal_unexpected_iid(id); break; 2284 } 2285 set_result(_gvn.transform(n)); 2286 return true; 2287 } 2288 2289 //--------------------------inline_unsigned_divmod_methods----------------------------- 2290 // inline int Integer.divideUnsigned(int, int) 2291 // inline int Integer.remainderUnsigned(int, int) 2292 // inline long Long.divideUnsigned(long, long) 2293 // inline long Long.remainderUnsigned(long, long) 2294 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) { 2295 Node* n = nullptr; 2296 switch (id) { 2297 case vmIntrinsics::_divideUnsigned_i: { 2298 zero_check_int(argument(1)); 2299 // Compile-time detect of null-exception 2300 if (stopped()) { 2301 return true; // keep the graph constructed so far 2302 } 2303 n = new UDivINode(control(), argument(0), argument(1)); 2304 break; 2305 } 2306 case vmIntrinsics::_divideUnsigned_l: { 2307 zero_check_long(argument(2)); 2308 // Compile-time detect of null-exception 2309 if (stopped()) { 2310 return true; // keep the graph constructed so far 2311 } 2312 n = new UDivLNode(control(), argument(0), argument(2)); 2313 break; 2314 } 2315 case vmIntrinsics::_remainderUnsigned_i: { 2316 zero_check_int(argument(1)); 2317 // Compile-time detect of null-exception 2318 if (stopped()) { 2319 return true; // keep the graph constructed so far 2320 } 2321 n = new UModINode(control(), argument(0), argument(1)); 2322 break; 2323 } 2324 case vmIntrinsics::_remainderUnsigned_l: { 2325 zero_check_long(argument(2)); 2326 // Compile-time detect of null-exception 2327 if (stopped()) { 2328 return true; // keep the graph constructed so far 2329 } 2330 n = new UModLNode(control(), argument(0), argument(2)); 2331 break; 2332 } 2333 default: fatal_unexpected_iid(id); break; 2334 } 2335 set_result(_gvn.transform(n)); 2336 return true; 2337 } 2338 2339 //----------------------------inline_unsafe_access---------------------------- 2340 2341 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) { 2342 // Attempt to infer a sharper value type from the offset and base type. 2343 ciKlass* sharpened_klass = nullptr; 2344 2345 // See if it is an instance field, with an object type. 2346 if (alias_type->field() != nullptr) { 2347 if (alias_type->field()->type()->is_klass()) { 2348 sharpened_klass = alias_type->field()->type()->as_klass(); 2349 } 2350 } 2351 2352 const TypeOopPtr* result = nullptr; 2353 // See if it is a narrow oop array. 2354 if (adr_type->isa_aryptr()) { 2355 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2356 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr(); 2357 if (elem_type != nullptr && elem_type->is_loaded()) { 2358 // Sharpen the value type. 2359 result = elem_type; 2360 } 2361 } 2362 } 2363 2364 // The sharpened class might be unloaded if there is no class loader 2365 // contraint in place. 2366 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) { 2367 // Sharpen the value type. 2368 result = TypeOopPtr::make_from_klass(sharpened_klass); 2369 } 2370 if (result != nullptr) { 2371 #ifndef PRODUCT 2372 if (C->print_intrinsics() || C->print_inlining()) { 2373 tty->print(" from base type: "); adr_type->dump(); tty->cr(); 2374 tty->print(" sharpened value: "); result->dump(); tty->cr(); 2375 } 2376 #endif 2377 } 2378 return result; 2379 } 2380 2381 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) { 2382 switch (kind) { 2383 case Relaxed: 2384 return MO_UNORDERED; 2385 case Opaque: 2386 return MO_RELAXED; 2387 case Acquire: 2388 return MO_ACQUIRE; 2389 case Release: 2390 return MO_RELEASE; 2391 case Volatile: 2392 return MO_SEQ_CST; 2393 default: 2394 ShouldNotReachHere(); 2395 return 0; 2396 } 2397 } 2398 2399 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) : 2400 _kit(kit), 2401 _sp(kit->sp()), 2402 _jvms(kit->jvms()), 2403 _map(kit->clone_map()), 2404 _discarded(false) 2405 { 2406 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) { 2407 Node* out = kit->control()->fast_out(i); 2408 if (out->is_CFG()) { 2409 _ctrl_succ.push(out); 2410 } 2411 } 2412 } 2413 2414 LibraryCallKit::SavedState::~SavedState() { 2415 if (_discarded) { 2416 _kit->destruct_map_clone(_map); 2417 return; 2418 } 2419 _kit->jvms()->set_map(_map); 2420 _kit->jvms()->set_sp(_sp); 2421 _map->set_jvms(_kit->jvms()); 2422 _kit->set_map(_map); 2423 _kit->set_sp(_sp); 2424 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) { 2425 Node* out = _kit->control()->fast_out(i); 2426 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) { 2427 _kit->_gvn.hash_delete(out); 2428 out->set_req(0, _kit->C->top()); 2429 _kit->C->record_for_igvn(out); 2430 --i; --imax; 2431 _kit->_gvn.hash_find_insert(out); 2432 } 2433 } 2434 } 2435 2436 void LibraryCallKit::SavedState::discard() { 2437 _discarded = true; 2438 } 2439 2440 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) { 2441 if (callee()->is_static()) return false; // caller must have the capability! 2442 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2443 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2444 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2445 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2446 2447 if (is_reference_type(type)) { 2448 decorators |= ON_UNKNOWN_OOP_REF; 2449 } 2450 2451 if (unaligned) { 2452 decorators |= C2_UNALIGNED; 2453 } 2454 2455 #ifndef PRODUCT 2456 { 2457 ResourceMark rm; 2458 // Check the signatures. 2459 ciSignature* sig = callee()->signature(); 2460 #ifdef ASSERT 2461 if (!is_store) { 2462 // Object getReference(Object base, int/long offset), etc. 2463 BasicType rtype = sig->return_type()->basic_type(); 2464 assert(rtype == type, "getter must return the expected value"); 2465 assert(sig->count() == 2, "oop getter has 2 arguments"); 2466 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2467 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2468 } else { 2469 // void putReference(Object base, int/long offset, Object x), etc. 2470 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2471 assert(sig->count() == 3, "oop putter has 3 arguments"); 2472 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2473 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2474 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2475 assert(vtype == type, "putter must accept the expected value"); 2476 } 2477 #endif // ASSERT 2478 } 2479 #endif //PRODUCT 2480 2481 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2482 2483 Node* receiver = argument(0); // type: oop 2484 2485 // Build address expression. 2486 Node* heap_base_oop = top(); 2487 2488 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2489 Node* base = argument(1); // type: oop 2490 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2491 Node* offset = argument(2); // type: long 2492 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2493 // to be plain byte offsets, which are also the same as those accepted 2494 // by oopDesc::field_addr. 2495 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2496 "fieldOffset must be byte-scaled"); 2497 // 32-bit machines ignore the high half! 2498 offset = ConvL2X(offset); 2499 2500 // Save state and restore on bailout 2501 SavedState old_state(this); 2502 2503 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed); 2504 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly"); 2505 2506 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) { 2507 if (type != T_OBJECT) { 2508 decorators |= IN_NATIVE; // off-heap primitive access 2509 } else { 2510 return false; // off-heap oop accesses are not supported 2511 } 2512 } else { 2513 heap_base_oop = base; // on-heap or mixed access 2514 } 2515 2516 // Can base be null? Otherwise, always on-heap access. 2517 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base)); 2518 2519 if (!can_access_non_heap) { 2520 decorators |= IN_HEAP; 2521 } 2522 2523 Node* val = is_store ? argument(4) : nullptr; 2524 2525 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr(); 2526 if (adr_type == TypePtr::NULL_PTR) { 2527 return false; // off-heap access with zero address 2528 } 2529 2530 // Try to categorize the address. 2531 Compile::AliasType* alias_type = C->alias_type(adr_type); 2532 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2533 2534 if (alias_type->adr_type() == TypeInstPtr::KLASS || 2535 alias_type->adr_type() == TypeAryPtr::RANGE) { 2536 return false; // not supported 2537 } 2538 2539 bool mismatched = false; 2540 BasicType bt = alias_type->basic_type(); 2541 if (bt != T_ILLEGAL) { 2542 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access"); 2543 if (bt == T_BYTE && adr_type->isa_aryptr()) { 2544 // Alias type doesn't differentiate between byte[] and boolean[]). 2545 // Use address type to get the element type. 2546 bt = adr_type->is_aryptr()->elem()->array_element_basic_type(); 2547 } 2548 if (is_reference_type(bt, true)) { 2549 // accessing an array field with getReference is not a mismatch 2550 bt = T_OBJECT; 2551 } 2552 if ((bt == T_OBJECT) != (type == T_OBJECT)) { 2553 // Don't intrinsify mismatched object accesses 2554 return false; 2555 } 2556 mismatched = (bt != type); 2557 } else if (alias_type->adr_type()->isa_oopptr()) { 2558 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched 2559 } 2560 2561 old_state.discard(); 2562 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched"); 2563 2564 if (mismatched) { 2565 decorators |= C2_MISMATCHED; 2566 } 2567 2568 // First guess at the value type. 2569 const Type *value_type = Type::get_const_basic_type(type); 2570 2571 // Figure out the memory ordering. 2572 decorators |= mo_decorator_for_access_kind(kind); 2573 2574 if (!is_store && type == T_OBJECT) { 2575 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2576 if (tjp != nullptr) { 2577 value_type = tjp; 2578 } 2579 } 2580 2581 receiver = null_check(receiver); 2582 if (stopped()) { 2583 return true; 2584 } 2585 // Heap pointers get a null-check from the interpreter, 2586 // as a courtesy. However, this is not guaranteed by Unsafe, 2587 // and it is not possible to fully distinguish unintended nulls 2588 // from intended ones in this API. 2589 2590 if (!is_store) { 2591 Node* p = nullptr; 2592 // Try to constant fold a load from a constant field 2593 ciField* field = alias_type->field(); 2594 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) { 2595 // final or stable field 2596 p = make_constant_from_field(field, heap_base_oop); 2597 } 2598 2599 if (p == nullptr) { // Could not constant fold the load 2600 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators); 2601 // Normalize the value returned by getBoolean in the following cases 2602 if (type == T_BOOLEAN && 2603 (mismatched || 2604 heap_base_oop == top() || // - heap_base_oop is null or 2605 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null 2606 // and the unsafe access is made to large offset 2607 // (i.e., larger than the maximum offset necessary for any 2608 // field access) 2609 ) { 2610 IdealKit ideal = IdealKit(this); 2611 #define __ ideal. 2612 IdealVariable normalized_result(ideal); 2613 __ declarations_done(); 2614 __ set(normalized_result, p); 2615 __ if_then(p, BoolTest::ne, ideal.ConI(0)); 2616 __ set(normalized_result, ideal.ConI(1)); 2617 ideal.end_if(); 2618 final_sync(ideal); 2619 p = __ value(normalized_result); 2620 #undef __ 2621 } 2622 } 2623 if (type == T_ADDRESS) { 2624 p = gvn().transform(new CastP2XNode(nullptr, p)); 2625 p = ConvX2UL(p); 2626 } 2627 // The load node has the control of the preceding MemBarCPUOrder. All 2628 // following nodes will have the control of the MemBarCPUOrder inserted at 2629 // the end of this method. So, pushing the load onto the stack at a later 2630 // point is fine. 2631 set_result(p); 2632 } else { 2633 if (bt == T_ADDRESS) { 2634 // Repackage the long as a pointer. 2635 val = ConvL2X(val); 2636 val = gvn().transform(new CastX2PNode(val)); 2637 } 2638 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators); 2639 } 2640 2641 return true; 2642 } 2643 2644 //----------------------------inline_unsafe_load_store---------------------------- 2645 // This method serves a couple of different customers (depending on LoadStoreKind): 2646 // 2647 // LS_cmp_swap: 2648 // 2649 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x); 2650 // boolean compareAndSetInt( Object o, long offset, int expected, int x); 2651 // boolean compareAndSetLong( Object o, long offset, long expected, long x); 2652 // 2653 // LS_cmp_swap_weak: 2654 // 2655 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x); 2656 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x); 2657 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x); 2658 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x); 2659 // 2660 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x); 2661 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x); 2662 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x); 2663 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x); 2664 // 2665 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x); 2666 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x); 2667 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x); 2668 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x); 2669 // 2670 // LS_cmp_exchange: 2671 // 2672 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x); 2673 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x); 2674 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x); 2675 // 2676 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2677 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2678 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2679 // 2680 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2681 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2682 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2683 // 2684 // LS_get_add: 2685 // 2686 // int getAndAddInt( Object o, long offset, int delta) 2687 // long getAndAddLong(Object o, long offset, long delta) 2688 // 2689 // LS_get_set: 2690 // 2691 // int getAndSet(Object o, long offset, int newValue) 2692 // long getAndSet(Object o, long offset, long newValue) 2693 // Object getAndSet(Object o, long offset, Object newValue) 2694 // 2695 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2696 // This basic scheme here is the same as inline_unsafe_access, but 2697 // differs in enough details that combining them would make the code 2698 // overly confusing. (This is a true fact! I originally combined 2699 // them, but even I was confused by it!) As much code/comments as 2700 // possible are retained from inline_unsafe_access though to make 2701 // the correspondences clearer. - dl 2702 2703 if (callee()->is_static()) return false; // caller must have the capability! 2704 2705 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2706 decorators |= mo_decorator_for_access_kind(access_kind); 2707 2708 #ifndef PRODUCT 2709 BasicType rtype; 2710 { 2711 ResourceMark rm; 2712 // Check the signatures. 2713 ciSignature* sig = callee()->signature(); 2714 rtype = sig->return_type()->basic_type(); 2715 switch(kind) { 2716 case LS_get_add: 2717 case LS_get_set: { 2718 // Check the signatures. 2719 #ifdef ASSERT 2720 assert(rtype == type, "get and set must return the expected type"); 2721 assert(sig->count() == 3, "get and set has 3 arguments"); 2722 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2723 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2724 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2725 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation"); 2726 #endif // ASSERT 2727 break; 2728 } 2729 case LS_cmp_swap: 2730 case LS_cmp_swap_weak: { 2731 // Check the signatures. 2732 #ifdef ASSERT 2733 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2734 assert(sig->count() == 4, "CAS has 4 arguments"); 2735 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2736 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2737 #endif // ASSERT 2738 break; 2739 } 2740 case LS_cmp_exchange: { 2741 // Check the signatures. 2742 #ifdef ASSERT 2743 assert(rtype == type, "CAS must return the expected type"); 2744 assert(sig->count() == 4, "CAS has 4 arguments"); 2745 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2746 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2747 #endif // ASSERT 2748 break; 2749 } 2750 default: 2751 ShouldNotReachHere(); 2752 } 2753 } 2754 #endif //PRODUCT 2755 2756 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2757 2758 // Get arguments: 2759 Node* receiver = nullptr; 2760 Node* base = nullptr; 2761 Node* offset = nullptr; 2762 Node* oldval = nullptr; 2763 Node* newval = nullptr; 2764 switch(kind) { 2765 case LS_cmp_swap: 2766 case LS_cmp_swap_weak: 2767 case LS_cmp_exchange: { 2768 const bool two_slot_type = type2size[type] == 2; 2769 receiver = argument(0); // type: oop 2770 base = argument(1); // type: oop 2771 offset = argument(2); // type: long 2772 oldval = argument(4); // type: oop, int, or long 2773 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2774 break; 2775 } 2776 case LS_get_add: 2777 case LS_get_set: { 2778 receiver = argument(0); // type: oop 2779 base = argument(1); // type: oop 2780 offset = argument(2); // type: long 2781 oldval = nullptr; 2782 newval = argument(4); // type: oop, int, or long 2783 break; 2784 } 2785 default: 2786 ShouldNotReachHere(); 2787 } 2788 2789 // Build field offset expression. 2790 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2791 // to be plain byte offsets, which are also the same as those accepted 2792 // by oopDesc::field_addr. 2793 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2794 // 32-bit machines ignore the high half of long offsets 2795 offset = ConvL2X(offset); 2796 // Save state and restore on bailout 2797 SavedState old_state(this); 2798 Node* adr = make_unsafe_address(base, offset,type, false); 2799 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2800 2801 Compile::AliasType* alias_type = C->alias_type(adr_type); 2802 BasicType bt = alias_type->basic_type(); 2803 if (bt != T_ILLEGAL && 2804 (is_reference_type(bt) != (type == T_OBJECT))) { 2805 // Don't intrinsify mismatched object accesses. 2806 return false; 2807 } 2808 2809 old_state.discard(); 2810 2811 // For CAS, unlike inline_unsafe_access, there seems no point in 2812 // trying to refine types. Just use the coarse types here. 2813 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2814 const Type *value_type = Type::get_const_basic_type(type); 2815 2816 switch (kind) { 2817 case LS_get_set: 2818 case LS_cmp_exchange: { 2819 if (type == T_OBJECT) { 2820 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2821 if (tjp != nullptr) { 2822 value_type = tjp; 2823 } 2824 } 2825 break; 2826 } 2827 case LS_cmp_swap: 2828 case LS_cmp_swap_weak: 2829 case LS_get_add: 2830 break; 2831 default: 2832 ShouldNotReachHere(); 2833 } 2834 2835 // Null check receiver. 2836 receiver = null_check(receiver); 2837 if (stopped()) { 2838 return true; 2839 } 2840 2841 int alias_idx = C->get_alias_index(adr_type); 2842 2843 if (is_reference_type(type)) { 2844 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF; 2845 2846 // Transformation of a value which could be null pointer (CastPP #null) 2847 // could be delayed during Parse (for example, in adjust_map_after_if()). 2848 // Execute transformation here to avoid barrier generation in such case. 2849 if (_gvn.type(newval) == TypePtr::NULL_PTR) 2850 newval = _gvn.makecon(TypePtr::NULL_PTR); 2851 2852 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) { 2853 // Refine the value to a null constant, when it is known to be null 2854 oldval = _gvn.makecon(TypePtr::NULL_PTR); 2855 } 2856 } 2857 2858 Node* result = nullptr; 2859 switch (kind) { 2860 case LS_cmp_exchange: { 2861 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx, 2862 oldval, newval, value_type, type, decorators); 2863 break; 2864 } 2865 case LS_cmp_swap_weak: 2866 decorators |= C2_WEAK_CMPXCHG; 2867 case LS_cmp_swap: { 2868 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx, 2869 oldval, newval, value_type, type, decorators); 2870 break; 2871 } 2872 case LS_get_set: { 2873 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx, 2874 newval, value_type, type, decorators); 2875 break; 2876 } 2877 case LS_get_add: { 2878 result = access_atomic_add_at(base, adr, adr_type, alias_idx, 2879 newval, value_type, type, decorators); 2880 break; 2881 } 2882 default: 2883 ShouldNotReachHere(); 2884 } 2885 2886 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 2887 set_result(result); 2888 return true; 2889 } 2890 2891 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 2892 // Regardless of form, don't allow previous ld/st to move down, 2893 // then issue acquire, release, or volatile mem_bar. 2894 insert_mem_bar(Op_MemBarCPUOrder); 2895 switch(id) { 2896 case vmIntrinsics::_loadFence: 2897 insert_mem_bar(Op_LoadFence); 2898 return true; 2899 case vmIntrinsics::_storeFence: 2900 insert_mem_bar(Op_StoreFence); 2901 return true; 2902 case vmIntrinsics::_storeStoreFence: 2903 insert_mem_bar(Op_StoreStoreFence); 2904 return true; 2905 case vmIntrinsics::_fullFence: 2906 insert_mem_bar(Op_MemBarVolatile); 2907 return true; 2908 default: 2909 fatal_unexpected_iid(id); 2910 return false; 2911 } 2912 } 2913 2914 bool LibraryCallKit::inline_onspinwait() { 2915 insert_mem_bar(Op_OnSpinWait); 2916 return true; 2917 } 2918 2919 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 2920 if (!kls->is_Con()) { 2921 return true; 2922 } 2923 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr(); 2924 if (klsptr == nullptr) { 2925 return true; 2926 } 2927 ciInstanceKlass* ik = klsptr->instance_klass(); 2928 // don't need a guard for a klass that is already initialized 2929 return !ik->is_initialized(); 2930 } 2931 2932 //----------------------------inline_unsafe_writeback0------------------------- 2933 // public native void Unsafe.writeback0(long address) 2934 bool LibraryCallKit::inline_unsafe_writeback0() { 2935 if (!Matcher::has_match_rule(Op_CacheWB)) { 2936 return false; 2937 } 2938 #ifndef PRODUCT 2939 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync"); 2940 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync"); 2941 ciSignature* sig = callee()->signature(); 2942 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!"); 2943 #endif 2944 null_check_receiver(); // null-check, then ignore 2945 Node *addr = argument(1); 2946 addr = new CastX2PNode(addr); 2947 addr = _gvn.transform(addr); 2948 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr); 2949 flush = _gvn.transform(flush); 2950 set_memory(flush, TypeRawPtr::BOTTOM); 2951 return true; 2952 } 2953 2954 //----------------------------inline_unsafe_writeback0------------------------- 2955 // public native void Unsafe.writeback0(long address) 2956 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) { 2957 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) { 2958 return false; 2959 } 2960 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) { 2961 return false; 2962 } 2963 #ifndef PRODUCT 2964 assert(Matcher::has_match_rule(Op_CacheWB), 2965 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB" 2966 : "found match rule for CacheWBPostSync but not CacheWB")); 2967 2968 #endif 2969 null_check_receiver(); // null-check, then ignore 2970 Node *sync; 2971 if (is_pre) { 2972 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2973 } else { 2974 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2975 } 2976 sync = _gvn.transform(sync); 2977 set_memory(sync, TypeRawPtr::BOTTOM); 2978 return true; 2979 } 2980 2981 //----------------------------inline_unsafe_allocate--------------------------- 2982 // public native Object Unsafe.allocateInstance(Class<?> cls); 2983 bool LibraryCallKit::inline_unsafe_allocate() { 2984 2985 #if INCLUDE_JVMTI 2986 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 2987 return false; 2988 } 2989 #endif //INCLUDE_JVMTI 2990 2991 if (callee()->is_static()) return false; // caller must have the capability! 2992 2993 null_check_receiver(); // null-check, then ignore 2994 Node* cls = null_check(argument(1)); 2995 if (stopped()) return true; 2996 2997 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0); 2998 kls = null_check(kls); 2999 if (stopped()) return true; // argument was like int.class 3000 3001 #if INCLUDE_JVMTI 3002 // Don't try to access new allocated obj in the intrinsic. 3003 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled. 3004 // Deoptimize and allocate in interpreter instead. 3005 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc)); 3006 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered); 3007 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0))); 3008 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq)); 3009 { 3010 BuildCutout unless(this, tst, PROB_MAX); 3011 uncommon_trap(Deoptimization::Reason_intrinsic, 3012 Deoptimization::Action_make_not_entrant); 3013 } 3014 if (stopped()) { 3015 return true; 3016 } 3017 #endif //INCLUDE_JVMTI 3018 3019 Node* test = nullptr; 3020 if (LibraryCallKit::klass_needs_init_guard(kls)) { 3021 // Note: The argument might still be an illegal value like 3022 // Serializable.class or Object[].class. The runtime will handle it. 3023 // But we must make an explicit check for initialization. 3024 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 3025 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 3026 // can generate code to load it as unsigned byte. 3027 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire); 3028 Node* bits = intcon(InstanceKlass::fully_initialized); 3029 test = _gvn.transform(new SubINode(inst, bits)); 3030 // The 'test' is non-zero if we need to take a slow path. 3031 } 3032 3033 Node* obj = new_instance(kls, test); 3034 set_result(obj); 3035 return true; 3036 } 3037 3038 //------------------------inline_native_time_funcs-------------- 3039 // inline code for System.currentTimeMillis() and System.nanoTime() 3040 // these have the same type and signature 3041 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 3042 const TypeFunc* tf = OptoRuntime::void_long_Type(); 3043 const TypePtr* no_memory_effects = nullptr; 3044 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 3045 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 3046 #ifdef ASSERT 3047 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 3048 assert(value_top == top(), "second value must be top"); 3049 #endif 3050 set_result(value); 3051 return true; 3052 } 3053 3054 3055 #if INCLUDE_JVMTI 3056 3057 // When notifications are disabled then just update the VTMS transition bit and return. 3058 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol. 3059 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) { 3060 if (!DoJVMTIVirtualThreadTransitions) { 3061 return true; 3062 } 3063 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument 3064 IdealKit ideal(this); 3065 3066 Node* ONE = ideal.ConI(1); 3067 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1))); 3068 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events)); 3069 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw); 3070 3071 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); { 3072 sync_kit(ideal); 3073 // if notifyJvmti enabled then make a call to the given SharedRuntime function 3074 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type(); 3075 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide); 3076 ideal.sync_kit(this); 3077 } ideal.else_(); { 3078 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object 3079 Node* thread = ideal.thread(); 3080 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset())); 3081 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset()); 3082 3083 sync_kit(ideal); 3084 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3085 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3086 3087 ideal.sync_kit(this); 3088 } ideal.end_if(); 3089 final_sync(ideal); 3090 3091 return true; 3092 } 3093 3094 // Always update the is_disable_suspend bit. 3095 bool LibraryCallKit::inline_native_notify_jvmti_sync() { 3096 if (!DoJVMTIVirtualThreadTransitions) { 3097 return true; 3098 } 3099 IdealKit ideal(this); 3100 3101 { 3102 // unconditionally update the is_disable_suspend bit in current JavaThread 3103 Node* thread = ideal.thread(); 3104 Node* arg = _gvn.transform(argument(0)); // argument for notification 3105 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset())); 3106 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr(); 3107 3108 sync_kit(ideal); 3109 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED); 3110 ideal.sync_kit(this); 3111 } 3112 final_sync(ideal); 3113 3114 return true; 3115 } 3116 3117 #endif // INCLUDE_JVMTI 3118 3119 #ifdef JFR_HAVE_INTRINSICS 3120 3121 /** 3122 * if oop->klass != null 3123 * // normal class 3124 * epoch = _epoch_state ? 2 : 1 3125 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch { 3126 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts 3127 * } 3128 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path 3129 * else 3130 * // primitive class 3131 * if oop->array_klass != null 3132 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path 3133 * else 3134 * id = LAST_TYPE_ID + 1 // void class path 3135 * if (!signaled) 3136 * signaled = true 3137 */ 3138 bool LibraryCallKit::inline_native_classID() { 3139 Node* cls = argument(0); 3140 3141 IdealKit ideal(this); 3142 #define __ ideal. 3143 IdealVariable result(ideal); __ declarations_done(); 3144 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), 3145 basic_plus_adr(cls, java_lang_Class::klass_offset()), 3146 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 3147 3148 3149 __ if_then(kls, BoolTest::ne, null()); { 3150 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET)); 3151 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw); 3152 3153 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address())); 3154 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw); 3155 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch)); 3156 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT))); 3157 mask = _gvn.transform(new OrLNode(mask, epoch)); 3158 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask)); 3159 3160 float unlikely = PROB_UNLIKELY(0.999); 3161 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); { 3162 sync_kit(ideal); 3163 make_runtime_call(RC_LEAF, 3164 OptoRuntime::class_id_load_barrier_Type(), 3165 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier), 3166 "class id load barrier", 3167 TypePtr::BOTTOM, 3168 kls); 3169 ideal.sync_kit(this); 3170 } __ end_if(); 3171 3172 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)))); 3173 } __ else_(); { 3174 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), 3175 basic_plus_adr(cls, java_lang_Class::array_klass_offset()), 3176 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 3177 __ if_then(array_kls, BoolTest::ne, null()); { 3178 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET)); 3179 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw); 3180 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))); 3181 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1)))); 3182 } __ else_(); { 3183 // void class case 3184 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1))); 3185 } __ end_if(); 3186 3187 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address())); 3188 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire); 3189 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); { 3190 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true); 3191 } __ end_if(); 3192 } __ end_if(); 3193 3194 final_sync(ideal); 3195 set_result(ideal.value(result)); 3196 #undef __ 3197 return true; 3198 } 3199 3200 //------------------------inline_native_jvm_commit------------------ 3201 bool LibraryCallKit::inline_native_jvm_commit() { 3202 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3203 3204 // Save input memory and i_o state. 3205 Node* input_memory_state = reset_memory(); 3206 set_all_memory(input_memory_state); 3207 Node* input_io_state = i_o(); 3208 3209 // TLS. 3210 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3211 // Jfr java buffer. 3212 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))))); 3213 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered)); 3214 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))))); 3215 3216 // Load the current value of the notified field in the JfrThreadLocal. 3217 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR)); 3218 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered); 3219 3220 // Test for notification. 3221 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1))); 3222 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq)); 3223 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN); 3224 3225 // True branch, is notified. 3226 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified)); 3227 set_control(is_notified); 3228 3229 // Reset notified state. 3230 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered); 3231 Node* notified_reset_memory = reset_memory(); 3232 3233 // 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. 3234 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered)); 3235 // Convert the machine-word to a long. 3236 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X)); 3237 3238 // False branch, not notified. 3239 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified)); 3240 set_control(not_notified); 3241 set_all_memory(input_memory_state); 3242 3243 // Arg is the next position as a long. 3244 Node* arg = argument(0); 3245 // Convert long to machine-word. 3246 Node* next_pos_X = _gvn.transform(ConvL2X(arg)); 3247 3248 // Store the next_position to the underlying jfr java buffer. 3249 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release); 3250 3251 Node* commit_memory = reset_memory(); 3252 set_all_memory(commit_memory); 3253 3254 // 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. 3255 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))))); 3256 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered); 3257 Node* lease_constant = _gvn.transform(_gvn.intcon(4)); 3258 3259 // And flags with lease constant. 3260 Node* lease = _gvn.transform(new AndINode(flags, lease_constant)); 3261 3262 // Branch on lease to conditionalize returning the leased java buffer. 3263 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant)); 3264 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq)); 3265 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN); 3266 3267 // False branch, not a lease. 3268 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease)); 3269 3270 // True branch, is lease. 3271 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease)); 3272 set_control(is_lease); 3273 3274 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop. 3275 Node* call_return_lease = make_runtime_call(RC_NO_LEAF, 3276 OptoRuntime::void_void_Type(), 3277 SharedRuntime::jfr_return_lease(), 3278 "return_lease", TypePtr::BOTTOM); 3279 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control)); 3280 3281 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT); 3282 record_for_igvn(lease_compare_rgn); 3283 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3284 record_for_igvn(lease_compare_mem); 3285 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO); 3286 record_for_igvn(lease_compare_io); 3287 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG); 3288 record_for_igvn(lease_result_value); 3289 3290 // Update control and phi nodes. 3291 lease_compare_rgn->init_req(_true_path, call_return_lease_control); 3292 lease_compare_rgn->init_req(_false_path, not_lease); 3293 3294 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3295 lease_compare_mem->init_req(_false_path, commit_memory); 3296 3297 lease_compare_io->init_req(_true_path, i_o()); 3298 lease_compare_io->init_req(_false_path, input_io_state); 3299 3300 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L. 3301 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position. 3302 3303 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3304 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3305 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO); 3306 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG); 3307 3308 // Update control and phi nodes. 3309 result_rgn->init_req(_true_path, is_notified); 3310 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn)); 3311 3312 result_mem->init_req(_true_path, notified_reset_memory); 3313 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem)); 3314 3315 result_io->init_req(_true_path, input_io_state); 3316 result_io->init_req(_false_path, _gvn.transform(lease_compare_io)); 3317 3318 result_value->init_req(_true_path, current_pos); 3319 result_value->init_req(_false_path, _gvn.transform(lease_result_value)); 3320 3321 // Set output state. 3322 set_control(_gvn.transform(result_rgn)); 3323 set_all_memory(_gvn.transform(result_mem)); 3324 set_i_o(_gvn.transform(result_io)); 3325 set_result(result_rgn, result_value); 3326 return true; 3327 } 3328 3329 /* 3330 * The intrinsic is a model of this pseudo-code: 3331 * 3332 * JfrThreadLocal* const tl = Thread::jfr_thread_local() 3333 * jobject h_event_writer = tl->java_event_writer(); 3334 * if (h_event_writer == nullptr) { 3335 * return nullptr; 3336 * } 3337 * oop threadObj = Thread::threadObj(); 3338 * oop vthread = java_lang_Thread::vthread(threadObj); 3339 * traceid tid; 3340 * bool pinVirtualThread; 3341 * bool excluded; 3342 * if (vthread != threadObj) { // i.e. current thread is virtual 3343 * tid = java_lang_Thread::tid(vthread); 3344 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread); 3345 * pinVirtualThread = VMContinuations; 3346 * excluded = vthread_epoch_raw & excluded_mask; 3347 * if (!excluded) { 3348 * traceid current_epoch = JfrTraceIdEpoch::current_generation(); 3349 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask; 3350 * if (vthread_epoch != current_epoch) { 3351 * write_checkpoint(); 3352 * } 3353 * } 3354 * } else { 3355 * tid = java_lang_Thread::tid(threadObj); 3356 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj); 3357 * pinVirtualThread = false; 3358 * excluded = thread_epoch_raw & excluded_mask; 3359 * } 3360 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer); 3361 * traceid tid_in_event_writer = getField(event_writer, "threadID"); 3362 * if (tid_in_event_writer != tid) { 3363 * setField(event_writer, "pinVirtualThread", pinVirtualThread); 3364 * setField(event_writer, "excluded", excluded); 3365 * setField(event_writer, "threadID", tid); 3366 * } 3367 * return event_writer 3368 */ 3369 bool LibraryCallKit::inline_native_getEventWriter() { 3370 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3371 3372 // Save input memory and i_o state. 3373 Node* input_memory_state = reset_memory(); 3374 set_all_memory(input_memory_state); 3375 Node* input_io_state = i_o(); 3376 3377 // The most significant bit of the u2 is used to denote thread exclusion 3378 Node* excluded_shift = _gvn.intcon(15); 3379 Node* excluded_mask = _gvn.intcon(1 << 15); 3380 // The epoch generation is the range [1-32767] 3381 Node* epoch_mask = _gvn.intcon(32767); 3382 3383 // TLS 3384 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3385 3386 // Load the address of java event writer jobject handle from the jfr_thread_local structure. 3387 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)); 3388 3389 // Load the eventwriter jobject handle. 3390 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered); 3391 3392 // Null check the jobject handle. 3393 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null())); 3394 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne)); 3395 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN); 3396 3397 // False path, jobj is null. 3398 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null)); 3399 3400 // True path, jobj is not null. 3401 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null)); 3402 3403 set_control(jobj_is_not_null); 3404 3405 // Load the threadObj for the CarrierThread. 3406 Node* threadObj = generate_current_thread(tls_ptr); 3407 3408 // Load the vthread. 3409 Node* vthread = generate_virtual_thread(tls_ptr); 3410 3411 // If vthread != threadObj, this is a virtual thread. 3412 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj)); 3413 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne)); 3414 IfNode* iff_vthread_not_equal_threadObj = 3415 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN); 3416 3417 // False branch, fallback to threadObj. 3418 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj)); 3419 set_control(vthread_equal_threadObj); 3420 3421 // Load the tid field from the vthread object. 3422 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J"); 3423 3424 // Load the raw epoch value from the threadObj. 3425 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset()); 3426 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, 3427 _gvn.type(threadObj_epoch_offset)->isa_ptr(), 3428 TypeInt::CHAR, T_CHAR, 3429 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3430 3431 // Mask off the excluded information from the epoch. 3432 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask)); 3433 3434 // True branch, this is a virtual thread. 3435 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj)); 3436 set_control(vthread_not_equal_threadObj); 3437 3438 // Load the tid field from the vthread object. 3439 Node* vthread_tid = load_field_from_object(vthread, "tid", "J"); 3440 3441 // Continuation support determines if a virtual thread should be pinned. 3442 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations)); 3443 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered); 3444 3445 // Load the raw epoch value from the vthread. 3446 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset()); 3447 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(), 3448 TypeInt::CHAR, T_CHAR, 3449 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3450 3451 // Mask off the excluded information from the epoch. 3452 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask))); 3453 3454 // Branch on excluded to conditionalize updating the epoch for the virtual thread. 3455 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask))); 3456 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne)); 3457 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN); 3458 3459 // False branch, vthread is excluded, no need to write epoch info. 3460 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded)); 3461 3462 // True branch, vthread is included, update epoch info. 3463 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded)); 3464 set_control(included); 3465 3466 // Get epoch value. 3467 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask))); 3468 3469 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR. 3470 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address())); 3471 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered); 3472 3473 // Compare the epoch in the vthread to the current epoch generation. 3474 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch)); 3475 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne)); 3476 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN); 3477 3478 // False path, epoch is equal, checkpoint information is valid. 3479 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal)); 3480 3481 // True path, epoch is not equal, write a checkpoint for the vthread. 3482 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal)); 3483 3484 set_control(epoch_is_not_equal); 3485 3486 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch. 3487 // The call also updates the native thread local thread id and the vthread with the current epoch. 3488 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF, 3489 OptoRuntime::jfr_write_checkpoint_Type(), 3490 SharedRuntime::jfr_write_checkpoint(), 3491 "write_checkpoint", TypePtr::BOTTOM); 3492 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control)); 3493 3494 // vthread epoch != current epoch 3495 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT); 3496 record_for_igvn(epoch_compare_rgn); 3497 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3498 record_for_igvn(epoch_compare_mem); 3499 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO); 3500 record_for_igvn(epoch_compare_io); 3501 3502 // Update control and phi nodes. 3503 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control); 3504 epoch_compare_rgn->init_req(_false_path, epoch_is_equal); 3505 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3506 epoch_compare_mem->init_req(_false_path, input_memory_state); 3507 epoch_compare_io->init_req(_true_path, i_o()); 3508 epoch_compare_io->init_req(_false_path, input_io_state); 3509 3510 // excluded != true 3511 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT); 3512 record_for_igvn(exclude_compare_rgn); 3513 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3514 record_for_igvn(exclude_compare_mem); 3515 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO); 3516 record_for_igvn(exclude_compare_io); 3517 3518 // Update control and phi nodes. 3519 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn)); 3520 exclude_compare_rgn->init_req(_false_path, excluded); 3521 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem)); 3522 exclude_compare_mem->init_req(_false_path, input_memory_state); 3523 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io)); 3524 exclude_compare_io->init_req(_false_path, input_io_state); 3525 3526 // vthread != threadObj 3527 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT); 3528 record_for_igvn(vthread_compare_rgn); 3529 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3530 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO); 3531 record_for_igvn(vthread_compare_io); 3532 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG); 3533 record_for_igvn(tid); 3534 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR); 3535 record_for_igvn(exclusion); 3536 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL); 3537 record_for_igvn(pinVirtualThread); 3538 3539 // Update control and phi nodes. 3540 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn)); 3541 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj); 3542 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem)); 3543 vthread_compare_mem->init_req(_false_path, input_memory_state); 3544 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io)); 3545 vthread_compare_io->init_req(_false_path, input_io_state); 3546 tid->init_req(_true_path, _gvn.transform(vthread_tid)); 3547 tid->init_req(_false_path, _gvn.transform(thread_obj_tid)); 3548 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded)); 3549 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded)); 3550 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support)); 3551 pinVirtualThread->init_req(_false_path, _gvn.intcon(0)); 3552 3553 // Update branch state. 3554 set_control(_gvn.transform(vthread_compare_rgn)); 3555 set_all_memory(_gvn.transform(vthread_compare_mem)); 3556 set_i_o(_gvn.transform(vthread_compare_io)); 3557 3558 // Load the event writer oop by dereferencing the jobject handle. 3559 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter")); 3560 assert(klass_EventWriter->is_loaded(), "invariant"); 3561 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass(); 3562 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter); 3563 const TypeOopPtr* const xtype = aklass->as_instance_type(); 3564 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global))); 3565 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD); 3566 3567 // Load the current thread id from the event writer object. 3568 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J"); 3569 // Get the field offset to, conditionally, store an updated tid value later. 3570 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false); 3571 // Get the field offset to, conditionally, store an updated exclusion value later. 3572 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false); 3573 // Get the field offset to, conditionally, store an updated pinVirtualThread value later. 3574 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false); 3575 3576 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT); 3577 record_for_igvn(event_writer_tid_compare_rgn); 3578 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3579 record_for_igvn(event_writer_tid_compare_mem); 3580 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO); 3581 record_for_igvn(event_writer_tid_compare_io); 3582 3583 // Compare the current tid from the thread object to what is currently stored in the event writer object. 3584 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid))); 3585 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne)); 3586 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN); 3587 3588 // False path, tids are the same. 3589 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal)); 3590 3591 // True path, tid is not equal, need to update the tid in the event writer. 3592 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal)); 3593 record_for_igvn(tid_is_not_equal); 3594 3595 // Store the pin state to the event writer. 3596 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered); 3597 3598 // Store the exclusion state to the event writer. 3599 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift)); 3600 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered); 3601 3602 // Store the tid to the event writer. 3603 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered); 3604 3605 // Update control and phi nodes. 3606 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal); 3607 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal); 3608 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3609 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem)); 3610 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o())); 3611 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io)); 3612 3613 // Result of top level CFG, Memory, IO and Value. 3614 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3615 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3616 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO); 3617 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM); 3618 3619 // Result control. 3620 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn)); 3621 result_rgn->init_req(_false_path, jobj_is_null); 3622 3623 // Result memory. 3624 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem)); 3625 result_mem->init_req(_false_path, _gvn.transform(input_memory_state)); 3626 3627 // Result IO. 3628 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io)); 3629 result_io->init_req(_false_path, _gvn.transform(input_io_state)); 3630 3631 // Result value. 3632 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop 3633 result_value->init_req(_false_path, null()); // return null 3634 3635 // Set output state. 3636 set_control(_gvn.transform(result_rgn)); 3637 set_all_memory(_gvn.transform(result_mem)); 3638 set_i_o(_gvn.transform(result_io)); 3639 set_result(result_rgn, result_value); 3640 return true; 3641 } 3642 3643 /* 3644 * The intrinsic is a model of this pseudo-code: 3645 * 3646 * JfrThreadLocal* const tl = thread->jfr_thread_local(); 3647 * if (carrierThread != thread) { // is virtual thread 3648 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread); 3649 * bool excluded = vthread_epoch_raw & excluded_mask; 3650 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread)); 3651 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded); 3652 * if (!excluded) { 3653 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask; 3654 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch); 3655 * } 3656 * AtomicAccess::release_store(&tl->_vthread, true); 3657 * return; 3658 * } 3659 * AtomicAccess::release_store(&tl->_vthread, false); 3660 */ 3661 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) { 3662 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3663 3664 Node* input_memory_state = reset_memory(); 3665 set_all_memory(input_memory_state); 3666 3667 // The most significant bit of the u2 is used to denote thread exclusion 3668 Node* excluded_mask = _gvn.intcon(1 << 15); 3669 // The epoch generation is the range [1-32767] 3670 Node* epoch_mask = _gvn.intcon(32767); 3671 3672 Node* const carrierThread = generate_current_thread(jt); 3673 // If thread != carrierThread, this is a virtual thread. 3674 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread)); 3675 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne)); 3676 IfNode* iff_thread_not_equal_carrierThread = 3677 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN); 3678 3679 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR)); 3680 3681 // False branch, is carrierThread. 3682 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread)); 3683 // Store release 3684 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true); 3685 3686 set_all_memory(input_memory_state); 3687 3688 // True branch, is virtual thread. 3689 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread)); 3690 set_control(thread_not_equal_carrierThread); 3691 3692 // Load the raw epoch value from the vthread. 3693 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset()); 3694 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR, 3695 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 3696 3697 // Mask off the excluded information from the epoch. 3698 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask))); 3699 3700 // Load the tid field from the thread. 3701 Node* tid = load_field_from_object(thread, "tid", "J"); 3702 3703 // Store the vthread tid to the jfr thread local. 3704 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR)); 3705 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true); 3706 3707 // Branch is_excluded to conditionalize updating the epoch . 3708 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask))); 3709 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq)); 3710 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN); 3711 3712 // True branch, vthread is excluded, no need to write epoch info. 3713 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded)); 3714 set_control(excluded); 3715 Node* vthread_is_excluded = _gvn.intcon(1); 3716 3717 // False branch, vthread is included, update epoch info. 3718 Node* included = _gvn.transform(new IfFalseNode(iff_excluded)); 3719 set_control(included); 3720 Node* vthread_is_included = _gvn.intcon(0); 3721 3722 // Get epoch value. 3723 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask))); 3724 3725 // Store the vthread epoch to the jfr thread local. 3726 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR)); 3727 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true); 3728 3729 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT); 3730 record_for_igvn(excluded_rgn); 3731 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM); 3732 record_for_igvn(excluded_mem); 3733 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL); 3734 record_for_igvn(exclusion); 3735 3736 // Merge the excluded control and memory. 3737 excluded_rgn->init_req(_true_path, excluded); 3738 excluded_rgn->init_req(_false_path, included); 3739 excluded_mem->init_req(_true_path, tid_memory); 3740 excluded_mem->init_req(_false_path, included_memory); 3741 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded)); 3742 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included)); 3743 3744 // Set intermediate state. 3745 set_control(_gvn.transform(excluded_rgn)); 3746 set_all_memory(excluded_mem); 3747 3748 // Store the vthread exclusion state to the jfr thread local. 3749 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR)); 3750 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true); 3751 3752 // Store release 3753 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true); 3754 3755 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT); 3756 record_for_igvn(thread_compare_rgn); 3757 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM); 3758 record_for_igvn(thread_compare_mem); 3759 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL); 3760 record_for_igvn(vthread); 3761 3762 // Merge the thread_compare control and memory. 3763 thread_compare_rgn->init_req(_true_path, control()); 3764 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread); 3765 thread_compare_mem->init_req(_true_path, vthread_true_memory); 3766 thread_compare_mem->init_req(_false_path, vthread_false_memory); 3767 3768 // Set output state. 3769 set_control(_gvn.transform(thread_compare_rgn)); 3770 set_all_memory(_gvn.transform(thread_compare_mem)); 3771 } 3772 3773 #endif // JFR_HAVE_INTRINSICS 3774 3775 //------------------------inline_native_currentCarrierThread------------------ 3776 bool LibraryCallKit::inline_native_currentCarrierThread() { 3777 Node* junk = nullptr; 3778 set_result(generate_current_thread(junk)); 3779 return true; 3780 } 3781 3782 //------------------------inline_native_currentThread------------------ 3783 bool LibraryCallKit::inline_native_currentThread() { 3784 Node* junk = nullptr; 3785 set_result(generate_virtual_thread(junk)); 3786 return true; 3787 } 3788 3789 //------------------------inline_native_setVthread------------------ 3790 bool LibraryCallKit::inline_native_setCurrentThread() { 3791 assert(C->method()->changes_current_thread(), 3792 "method changes current Thread but is not annotated ChangesCurrentThread"); 3793 Node* arr = argument(1); 3794 Node* thread = _gvn.transform(new ThreadLocalNode()); 3795 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset())); 3796 Node* thread_obj_handle 3797 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered); 3798 thread_obj_handle = _gvn.transform(thread_obj_handle); 3799 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr(); 3800 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED); 3801 3802 // Change the _monitor_owner_id of the JavaThread 3803 Node* tid = load_field_from_object(arr, "tid", "J"); 3804 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset())); 3805 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true); 3806 3807 JFR_ONLY(extend_setCurrentThread(thread, arr);) 3808 return true; 3809 } 3810 3811 const Type* LibraryCallKit::scopedValueCache_type() { 3812 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass()); 3813 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass()); 3814 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS); 3815 3816 // Because we create the scopedValue cache lazily we have to make the 3817 // type of the result BotPTR. 3818 bool xk = etype->klass_is_exact(); 3819 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0); 3820 return objects_type; 3821 } 3822 3823 Node* LibraryCallKit::scopedValueCache_helper() { 3824 Node* thread = _gvn.transform(new ThreadLocalNode()); 3825 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset())); 3826 // We cannot use immutable_memory() because we might flip onto a 3827 // different carrier thread, at which point we'll need to use that 3828 // carrier thread's cache. 3829 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(), 3830 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)); 3831 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered); 3832 } 3833 3834 //------------------------inline_native_scopedValueCache------------------ 3835 bool LibraryCallKit::inline_native_scopedValueCache() { 3836 Node* cache_obj_handle = scopedValueCache_helper(); 3837 const Type* objects_type = scopedValueCache_type(); 3838 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE)); 3839 3840 return true; 3841 } 3842 3843 //------------------------inline_native_setScopedValueCache------------------ 3844 bool LibraryCallKit::inline_native_setScopedValueCache() { 3845 Node* arr = argument(0); 3846 Node* cache_obj_handle = scopedValueCache_helper(); 3847 const Type* objects_type = scopedValueCache_type(); 3848 3849 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr(); 3850 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED); 3851 3852 return true; 3853 } 3854 3855 //------------------------inline_native_Continuation_pin and unpin----------- 3856 3857 // Shared implementation routine for both pin and unpin. 3858 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) { 3859 enum { _true_path = 1, _false_path = 2, PATH_LIMIT }; 3860 3861 // Save input memory. 3862 Node* input_memory_state = reset_memory(); 3863 set_all_memory(input_memory_state); 3864 3865 // TLS 3866 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3867 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset())); 3868 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered); 3869 3870 // Null check the last continuation object. 3871 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null())); 3872 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne)); 3873 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN); 3874 3875 // False path, last continuation is null. 3876 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null)); 3877 3878 // True path, last continuation is not null. 3879 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null)); 3880 3881 set_control(continuation_is_not_null); 3882 3883 // Load the pin count from the last continuation. 3884 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset())); 3885 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered); 3886 3887 // The loaded pin count is compared against a context specific rhs for over/underflow detection. 3888 Node* pin_count_rhs; 3889 if (unpin) { 3890 pin_count_rhs = _gvn.intcon(0); 3891 } else { 3892 pin_count_rhs = _gvn.intcon(UINT32_MAX); 3893 } 3894 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs)); 3895 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq)); 3896 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN); 3897 3898 // True branch, pin count over/underflow. 3899 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow)); 3900 { 3901 // Trap (but not deoptimize (Action_none)) and continue in the interpreter 3902 // which will throw IllegalStateException for pin count over/underflow. 3903 // No memory changed so far - we can use memory create by reset_memory() 3904 // at the beginning of this intrinsic. No need to call reset_memory() again. 3905 PreserveJVMState pjvms(this); 3906 set_control(pin_count_over_underflow); 3907 uncommon_trap(Deoptimization::Reason_intrinsic, 3908 Deoptimization::Action_none); 3909 assert(stopped(), "invariant"); 3910 } 3911 3912 // False branch, no pin count over/underflow. Increment or decrement pin count and store back. 3913 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow)); 3914 set_control(valid_pin_count); 3915 3916 Node* next_pin_count; 3917 if (unpin) { 3918 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1))); 3919 } else { 3920 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1))); 3921 } 3922 3923 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered); 3924 3925 // Result of top level CFG and Memory. 3926 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3927 record_for_igvn(result_rgn); 3928 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM); 3929 record_for_igvn(result_mem); 3930 3931 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count)); 3932 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null)); 3933 result_mem->init_req(_true_path, _gvn.transform(reset_memory())); 3934 result_mem->init_req(_false_path, _gvn.transform(input_memory_state)); 3935 3936 // Set output state. 3937 set_control(_gvn.transform(result_rgn)); 3938 set_all_memory(_gvn.transform(result_mem)); 3939 3940 return true; 3941 } 3942 3943 //---------------------------load_mirror_from_klass---------------------------- 3944 // Given a klass oop, load its java mirror (a java.lang.Class oop). 3945 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) { 3946 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset())); 3947 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3948 // mirror = ((OopHandle)mirror)->resolve(); 3949 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE); 3950 } 3951 3952 //-----------------------load_klass_from_mirror_common------------------------- 3953 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3954 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3955 // and branch to the given path on the region. 3956 // If never_see_null, take an uncommon trap on null, so we can optimistically 3957 // compile for the non-null case. 3958 // If the region is null, force never_see_null = true. 3959 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3960 bool never_see_null, 3961 RegionNode* region, 3962 int null_path, 3963 int offset) { 3964 if (region == nullptr) never_see_null = true; 3965 Node* p = basic_plus_adr(mirror, offset); 3966 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL; 3967 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3968 Node* null_ctl = top(); 3969 kls = null_check_oop(kls, &null_ctl, never_see_null); 3970 if (region != nullptr) { 3971 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3972 region->init_req(null_path, null_ctl); 3973 } else { 3974 assert(null_ctl == top(), "no loose ends"); 3975 } 3976 return kls; 3977 } 3978 3979 //--------------------(inline_native_Class_query helpers)--------------------- 3980 // Use this for JVM_ACC_INTERFACE. 3981 // Fall through if (mods & mask) == bits, take the guard otherwise. 3982 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region, 3983 ByteSize offset, const Type* type, BasicType bt) { 3984 // Branch around if the given klass has the given modifier bit set. 3985 // Like generate_guard, adds a new path onto the region. 3986 Node* modp = basic_plus_adr(kls, in_bytes(offset)); 3987 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered); 3988 Node* mask = intcon(modifier_mask); 3989 Node* bits = intcon(modifier_bits); 3990 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3991 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3992 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3993 return generate_fair_guard(bol, region); 3994 } 3995 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3996 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region, 3997 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR); 3998 } 3999 4000 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast. 4001 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 4002 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region, 4003 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN); 4004 } 4005 4006 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) { 4007 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region); 4008 } 4009 4010 //-------------------------inline_native_Class_query------------------- 4011 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 4012 const Type* return_type = TypeInt::BOOL; 4013 Node* prim_return_value = top(); // what happens if it's a primitive class? 4014 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 4015 bool expect_prim = false; // most of these guys expect to work on refs 4016 4017 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 4018 4019 Node* mirror = argument(0); 4020 Node* obj = top(); 4021 4022 switch (id) { 4023 case vmIntrinsics::_isInstance: 4024 // nothing is an instance of a primitive type 4025 prim_return_value = intcon(0); 4026 obj = argument(1); 4027 break; 4028 case vmIntrinsics::_isHidden: 4029 prim_return_value = intcon(0); 4030 break; 4031 case vmIntrinsics::_getSuperclass: 4032 prim_return_value = null(); 4033 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 4034 break; 4035 default: 4036 fatal_unexpected_iid(id); 4037 break; 4038 } 4039 4040 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 4041 if (mirror_con == nullptr) return false; // cannot happen? 4042 4043 #ifndef PRODUCT 4044 if (C->print_intrinsics() || C->print_inlining()) { 4045 ciType* k = mirror_con->java_mirror_type(); 4046 if (k) { 4047 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 4048 k->print_name(); 4049 tty->cr(); 4050 } 4051 } 4052 #endif 4053 4054 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 4055 RegionNode* region = new RegionNode(PATH_LIMIT); 4056 record_for_igvn(region); 4057 PhiNode* phi = new PhiNode(region, return_type); 4058 4059 // The mirror will never be null of Reflection.getClassAccessFlags, however 4060 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 4061 // if it is. See bug 4774291. 4062 4063 // For Reflection.getClassAccessFlags(), the null check occurs in 4064 // the wrong place; see inline_unsafe_access(), above, for a similar 4065 // situation. 4066 mirror = null_check(mirror); 4067 // If mirror or obj is dead, only null-path is taken. 4068 if (stopped()) return true; 4069 4070 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 4071 4072 // Now load the mirror's klass metaobject, and null-check it. 4073 // Side-effects region with the control path if the klass is null. 4074 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 4075 // If kls is null, we have a primitive mirror. 4076 phi->init_req(_prim_path, prim_return_value); 4077 if (stopped()) { set_result(region, phi); return true; } 4078 bool safe_for_replace = (region->in(_prim_path) == top()); 4079 4080 Node* p; // handy temp 4081 Node* null_ctl; 4082 4083 // Now that we have the non-null klass, we can perform the real query. 4084 // For constant classes, the query will constant-fold in LoadNode::Value. 4085 Node* query_value = top(); 4086 switch (id) { 4087 case vmIntrinsics::_isInstance: 4088 // nothing is an instance of a primitive type 4089 query_value = gen_instanceof(obj, kls, safe_for_replace); 4090 break; 4091 4092 case vmIntrinsics::_isHidden: 4093 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.) 4094 if (generate_hidden_class_guard(kls, region) != nullptr) 4095 // A guard was added. If the guard is taken, it was an hidden class. 4096 phi->add_req(intcon(1)); 4097 // If we fall through, it's a plain class. 4098 query_value = intcon(0); 4099 break; 4100 4101 4102 case vmIntrinsics::_getSuperclass: 4103 // The rules here are somewhat unfortunate, but we can still do better 4104 // with random logic than with a JNI call. 4105 // Interfaces store null or Object as _super, but must report null. 4106 // Arrays store an intermediate super as _super, but must report Object. 4107 // Other types can report the actual _super. 4108 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 4109 if (generate_interface_guard(kls, region) != nullptr) 4110 // A guard was added. If the guard is taken, it was an interface. 4111 phi->add_req(null()); 4112 if (generate_array_guard(kls, region) != nullptr) 4113 // A guard was added. If the guard is taken, it was an array. 4114 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 4115 // If we fall through, it's a plain class. Get its _super. 4116 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 4117 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL)); 4118 null_ctl = top(); 4119 kls = null_check_oop(kls, &null_ctl); 4120 if (null_ctl != top()) { 4121 // If the guard is taken, Object.superClass is null (both klass and mirror). 4122 region->add_req(null_ctl); 4123 phi ->add_req(null()); 4124 } 4125 if (!stopped()) { 4126 query_value = load_mirror_from_klass(kls); 4127 } 4128 break; 4129 4130 default: 4131 fatal_unexpected_iid(id); 4132 break; 4133 } 4134 4135 // Fall-through is the normal case of a query to a real class. 4136 phi->init_req(1, query_value); 4137 region->init_req(1, control()); 4138 4139 C->set_has_split_ifs(true); // Has chance for split-if optimization 4140 set_result(region, phi); 4141 return true; 4142 } 4143 4144 //-------------------------inline_Class_cast------------------- 4145 bool LibraryCallKit::inline_Class_cast() { 4146 Node* mirror = argument(0); // Class 4147 Node* obj = argument(1); 4148 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 4149 if (mirror_con == nullptr) { 4150 return false; // dead path (mirror->is_top()). 4151 } 4152 if (obj == nullptr || obj->is_top()) { 4153 return false; // dead path 4154 } 4155 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr(); 4156 4157 // First, see if Class.cast() can be folded statically. 4158 // java_mirror_type() returns non-null for compile-time Class constants. 4159 ciType* tm = mirror_con->java_mirror_type(); 4160 if (tm != nullptr && tm->is_klass() && 4161 tp != nullptr) { 4162 if (!tp->is_loaded()) { 4163 // Don't use intrinsic when class is not loaded. 4164 return false; 4165 } else { 4166 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type()); 4167 if (static_res == Compile::SSC_always_true) { 4168 // isInstance() is true - fold the code. 4169 set_result(obj); 4170 return true; 4171 } else if (static_res == Compile::SSC_always_false) { 4172 // Don't use intrinsic, have to throw ClassCastException. 4173 // If the reference is null, the non-intrinsic bytecode will 4174 // be optimized appropriately. 4175 return false; 4176 } 4177 } 4178 } 4179 4180 // Bailout intrinsic and do normal inlining if exception path is frequent. 4181 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 4182 return false; 4183 } 4184 4185 // Generate dynamic checks. 4186 // Class.cast() is java implementation of _checkcast bytecode. 4187 // Do checkcast (Parse::do_checkcast()) optimizations here. 4188 4189 mirror = null_check(mirror); 4190 // If mirror is dead, only null-path is taken. 4191 if (stopped()) { 4192 return true; 4193 } 4194 4195 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive). 4196 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT }; 4197 RegionNode* region = new RegionNode(PATH_LIMIT); 4198 record_for_igvn(region); 4199 4200 // Now load the mirror's klass metaobject, and null-check it. 4201 // If kls is null, we have a primitive mirror and 4202 // nothing is an instance of a primitive type. 4203 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 4204 4205 Node* res = top(); 4206 if (!stopped()) { 4207 Node* bad_type_ctrl = top(); 4208 // Do checkcast optimizations. 4209 res = gen_checkcast(obj, kls, &bad_type_ctrl); 4210 region->init_req(_bad_type_path, bad_type_ctrl); 4211 } 4212 if (region->in(_prim_path) != top() || 4213 region->in(_bad_type_path) != top()) { 4214 // Let Interpreter throw ClassCastException. 4215 PreserveJVMState pjvms(this); 4216 set_control(_gvn.transform(region)); 4217 uncommon_trap(Deoptimization::Reason_intrinsic, 4218 Deoptimization::Action_maybe_recompile); 4219 } 4220 if (!stopped()) { 4221 set_result(res); 4222 } 4223 return true; 4224 } 4225 4226 4227 //--------------------------inline_native_subtype_check------------------------ 4228 // This intrinsic takes the JNI calls out of the heart of 4229 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 4230 bool LibraryCallKit::inline_native_subtype_check() { 4231 // Pull both arguments off the stack. 4232 Node* args[2]; // two java.lang.Class mirrors: superc, subc 4233 args[0] = argument(0); 4234 args[1] = argument(1); 4235 Node* klasses[2]; // corresponding Klasses: superk, subk 4236 klasses[0] = klasses[1] = top(); 4237 4238 enum { 4239 // A full decision tree on {superc is prim, subc is prim}: 4240 _prim_0_path = 1, // {P,N} => false 4241 // {P,P} & superc!=subc => false 4242 _prim_same_path, // {P,P} & superc==subc => true 4243 _prim_1_path, // {N,P} => false 4244 _ref_subtype_path, // {N,N} & subtype check wins => true 4245 _both_ref_path, // {N,N} & subtype check loses => false 4246 PATH_LIMIT 4247 }; 4248 4249 RegionNode* region = new RegionNode(PATH_LIMIT); 4250 Node* phi = new PhiNode(region, TypeInt::BOOL); 4251 record_for_igvn(region); 4252 4253 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 4254 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL; 4255 int class_klass_offset = java_lang_Class::klass_offset(); 4256 4257 // First null-check both mirrors and load each mirror's klass metaobject. 4258 int which_arg; 4259 for (which_arg = 0; which_arg <= 1; which_arg++) { 4260 Node* arg = args[which_arg]; 4261 arg = null_check(arg); 4262 if (stopped()) break; 4263 args[which_arg] = arg; 4264 4265 Node* p = basic_plus_adr(arg, class_klass_offset); 4266 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type); 4267 klasses[which_arg] = _gvn.transform(kls); 4268 } 4269 4270 // Having loaded both klasses, test each for null. 4271 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 4272 for (which_arg = 0; which_arg <= 1; which_arg++) { 4273 Node* kls = klasses[which_arg]; 4274 Node* null_ctl = top(); 4275 kls = null_check_oop(kls, &null_ctl, never_see_null); 4276 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path); 4277 region->init_req(prim_path, null_ctl); 4278 if (stopped()) break; 4279 klasses[which_arg] = kls; 4280 } 4281 4282 if (!stopped()) { 4283 // now we have two reference types, in klasses[0..1] 4284 Node* subk = klasses[1]; // the argument to isAssignableFrom 4285 Node* superk = klasses[0]; // the receiver 4286 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 4287 // now we have a successful reference subtype check 4288 region->set_req(_ref_subtype_path, control()); 4289 } 4290 4291 // If both operands are primitive (both klasses null), then 4292 // we must return true when they are identical primitives. 4293 // It is convenient to test this after the first null klass check. 4294 set_control(region->in(_prim_0_path)); // go back to first null check 4295 if (!stopped()) { 4296 // Since superc is primitive, make a guard for the superc==subc case. 4297 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 4298 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 4299 generate_guard(bol_eq, region, PROB_FAIR); 4300 if (region->req() == PATH_LIMIT+1) { 4301 // A guard was added. If the added guard is taken, superc==subc. 4302 region->swap_edges(PATH_LIMIT, _prim_same_path); 4303 region->del_req(PATH_LIMIT); 4304 } 4305 region->set_req(_prim_0_path, control()); // Not equal after all. 4306 } 4307 4308 // these are the only paths that produce 'true': 4309 phi->set_req(_prim_same_path, intcon(1)); 4310 phi->set_req(_ref_subtype_path, intcon(1)); 4311 4312 // pull together the cases: 4313 assert(region->req() == PATH_LIMIT, "sane region"); 4314 for (uint i = 1; i < region->req(); i++) { 4315 Node* ctl = region->in(i); 4316 if (ctl == nullptr || ctl == top()) { 4317 region->set_req(i, top()); 4318 phi ->set_req(i, top()); 4319 } else if (phi->in(i) == nullptr) { 4320 phi->set_req(i, intcon(0)); // all other paths produce 'false' 4321 } 4322 } 4323 4324 set_control(_gvn.transform(region)); 4325 set_result(_gvn.transform(phi)); 4326 return true; 4327 } 4328 4329 //---------------------generate_array_guard_common------------------------ 4330 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, 4331 bool obj_array, bool not_array, Node** obj) { 4332 4333 if (stopped()) { 4334 return nullptr; 4335 } 4336 4337 // If obj_array/non_array==false/false: 4338 // Branch around if the given klass is in fact an array (either obj or prim). 4339 // If obj_array/non_array==false/true: 4340 // Branch around if the given klass is not an array klass of any kind. 4341 // If obj_array/non_array==true/true: 4342 // Branch around if the kls is not an oop array (kls is int[], String, etc.) 4343 // If obj_array/non_array==true/false: 4344 // Branch around if the kls is an oop array (Object[] or subtype) 4345 // 4346 // Like generate_guard, adds a new path onto the region. 4347 jint layout_con = 0; 4348 Node* layout_val = get_layout_helper(kls, layout_con); 4349 if (layout_val == nullptr) { 4350 bool query = (obj_array 4351 ? Klass::layout_helper_is_objArray(layout_con) 4352 : Klass::layout_helper_is_array(layout_con)); 4353 if (query == not_array) { 4354 return nullptr; // never a branch 4355 } else { // always a branch 4356 Node* always_branch = control(); 4357 if (region != nullptr) 4358 region->add_req(always_branch); 4359 set_control(top()); 4360 return always_branch; 4361 } 4362 } 4363 // Now test the correct condition. 4364 jint nval = (obj_array 4365 ? (jint)(Klass::_lh_array_tag_type_value 4366 << Klass::_lh_array_tag_shift) 4367 : Klass::_lh_neutral_value); 4368 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 4369 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array 4370 // invert the test if we are looking for a non-array 4371 if (not_array) btest = BoolTest(btest).negate(); 4372 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 4373 Node* ctrl = generate_fair_guard(bol, region); 4374 Node* is_array_ctrl = not_array ? control() : ctrl; 4375 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) { 4376 // Keep track of the fact that 'obj' is an array to prevent 4377 // array specific accesses from floating above the guard. 4378 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM)); 4379 } 4380 return ctrl; 4381 } 4382 4383 4384 //-----------------------inline_native_newArray-------------------------- 4385 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length); 4386 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size); 4387 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) { 4388 Node* mirror; 4389 Node* count_val; 4390 if (uninitialized) { 4391 null_check_receiver(); 4392 mirror = argument(1); 4393 count_val = argument(2); 4394 } else { 4395 mirror = argument(0); 4396 count_val = argument(1); 4397 } 4398 4399 mirror = null_check(mirror); 4400 // If mirror or obj is dead, only null-path is taken. 4401 if (stopped()) return true; 4402 4403 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 4404 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4405 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4406 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4407 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4408 4409 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 4410 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 4411 result_reg, _slow_path); 4412 Node* normal_ctl = control(); 4413 Node* no_array_ctl = result_reg->in(_slow_path); 4414 4415 // Generate code for the slow case. We make a call to newArray(). 4416 set_control(no_array_ctl); 4417 if (!stopped()) { 4418 // Either the input type is void.class, or else the 4419 // array klass has not yet been cached. Either the 4420 // ensuing call will throw an exception, or else it 4421 // will cache the array klass for next time. 4422 PreserveJVMState pjvms(this); 4423 CallJavaNode* slow_call = nullptr; 4424 if (uninitialized) { 4425 // Generate optimized virtual call (holder class 'Unsafe' is final) 4426 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true); 4427 } else { 4428 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true); 4429 } 4430 Node* slow_result = set_results_for_java_call(slow_call); 4431 // this->control() comes from set_results_for_java_call 4432 result_reg->set_req(_slow_path, control()); 4433 result_val->set_req(_slow_path, slow_result); 4434 result_io ->set_req(_slow_path, i_o()); 4435 result_mem->set_req(_slow_path, reset_memory()); 4436 } 4437 4438 set_control(normal_ctl); 4439 if (!stopped()) { 4440 // Normal case: The array type has been cached in the java.lang.Class. 4441 // The following call works fine even if the array type is polymorphic. 4442 // It could be a dynamic mix of int[], boolean[], Object[], etc. 4443 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push 4444 result_reg->init_req(_normal_path, control()); 4445 result_val->init_req(_normal_path, obj); 4446 result_io ->init_req(_normal_path, i_o()); 4447 result_mem->init_req(_normal_path, reset_memory()); 4448 4449 if (uninitialized) { 4450 // Mark the allocation so that zeroing is skipped 4451 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj); 4452 alloc->maybe_set_complete(&_gvn); 4453 } 4454 } 4455 4456 // Return the combined state. 4457 set_i_o( _gvn.transform(result_io) ); 4458 set_all_memory( _gvn.transform(result_mem)); 4459 4460 C->set_has_split_ifs(true); // Has chance for split-if optimization 4461 set_result(result_reg, result_val); 4462 return true; 4463 } 4464 4465 //----------------------inline_native_getLength-------------------------- 4466 // public static native int java.lang.reflect.Array.getLength(Object array); 4467 bool LibraryCallKit::inline_native_getLength() { 4468 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 4469 4470 Node* array = null_check(argument(0)); 4471 // If array is dead, only null-path is taken. 4472 if (stopped()) return true; 4473 4474 // Deoptimize if it is a non-array. 4475 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array); 4476 4477 if (non_array != nullptr) { 4478 PreserveJVMState pjvms(this); 4479 set_control(non_array); 4480 uncommon_trap(Deoptimization::Reason_intrinsic, 4481 Deoptimization::Action_maybe_recompile); 4482 } 4483 4484 // If control is dead, only non-array-path is taken. 4485 if (stopped()) return true; 4486 4487 // The works fine even if the array type is polymorphic. 4488 // It could be a dynamic mix of int[], boolean[], Object[], etc. 4489 Node* result = load_array_length(array); 4490 4491 C->set_has_split_ifs(true); // Has chance for split-if optimization 4492 set_result(result); 4493 return true; 4494 } 4495 4496 //------------------------inline_array_copyOf---------------------------- 4497 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 4498 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 4499 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 4500 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 4501 4502 // Get the arguments. 4503 Node* original = argument(0); 4504 Node* start = is_copyOfRange? argument(1): intcon(0); 4505 Node* end = is_copyOfRange? argument(2): argument(1); 4506 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 4507 4508 Node* newcopy = nullptr; 4509 4510 // Set the original stack and the reexecute bit for the interpreter to reexecute 4511 // the bytecode that invokes Arrays.copyOf if deoptimization happens. 4512 { PreserveReexecuteState preexecs(this); 4513 jvms()->set_should_reexecute(true); 4514 4515 array_type_mirror = null_check(array_type_mirror); 4516 original = null_check(original); 4517 4518 // Check if a null path was taken unconditionally. 4519 if (stopped()) return true; 4520 4521 Node* orig_length = load_array_length(original); 4522 4523 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0); 4524 klass_node = null_check(klass_node); 4525 4526 RegionNode* bailout = new RegionNode(1); 4527 record_for_igvn(bailout); 4528 4529 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc. 4530 // Bail out if that is so. 4531 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout); 4532 if (not_objArray != nullptr) { 4533 // Improve the klass node's type from the new optimistic assumption: 4534 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 4535 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/); 4536 Node* cast = new CastPPNode(control(), klass_node, akls); 4537 klass_node = _gvn.transform(cast); 4538 } 4539 4540 // Bail out if either start or end is negative. 4541 generate_negative_guard(start, bailout, &start); 4542 generate_negative_guard(end, bailout, &end); 4543 4544 Node* length = end; 4545 if (_gvn.type(start) != TypeInt::ZERO) { 4546 length = _gvn.transform(new SubINode(end, start)); 4547 } 4548 4549 // Bail out if length is negative (i.e., if start > end). 4550 // Without this the new_array would throw 4551 // NegativeArraySizeException but IllegalArgumentException is what 4552 // should be thrown 4553 generate_negative_guard(length, bailout, &length); 4554 4555 // Bail out if start is larger than the original length 4556 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 4557 generate_negative_guard(orig_tail, bailout, &orig_tail); 4558 4559 if (bailout->req() > 1) { 4560 PreserveJVMState pjvms(this); 4561 set_control(_gvn.transform(bailout)); 4562 uncommon_trap(Deoptimization::Reason_intrinsic, 4563 Deoptimization::Action_maybe_recompile); 4564 } 4565 4566 if (!stopped()) { 4567 // How many elements will we copy from the original? 4568 // The answer is MinI(orig_tail, length). 4569 Node* moved = _gvn.transform(new MinINode(orig_tail, length)); 4570 4571 // Generate a direct call to the right arraycopy function(s). 4572 // We know the copy is disjoint but we might not know if the 4573 // oop stores need checking. 4574 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 4575 // This will fail a store-check if x contains any non-nulls. 4576 4577 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 4578 // loads/stores but it is legal only if we're sure the 4579 // Arrays.copyOf would succeed. So we need all input arguments 4580 // to the copyOf to be validated, including that the copy to the 4581 // new array won't trigger an ArrayStoreException. That subtype 4582 // check can be optimized if we know something on the type of 4583 // the input array from type speculation. 4584 if (_gvn.type(klass_node)->singleton()) { 4585 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr(); 4586 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr(); 4587 4588 int test = C->static_subtype_check(superk, subk); 4589 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 4590 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 4591 if (t_original->speculative_type() != nullptr) { 4592 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 4593 } 4594 } 4595 } 4596 4597 bool validated = false; 4598 // Reason_class_check rather than Reason_intrinsic because we 4599 // want to intrinsify even if this traps. 4600 if (!too_many_traps(Deoptimization::Reason_class_check)) { 4601 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node); 4602 4603 if (not_subtype_ctrl != top()) { 4604 PreserveJVMState pjvms(this); 4605 set_control(not_subtype_ctrl); 4606 uncommon_trap(Deoptimization::Reason_class_check, 4607 Deoptimization::Action_make_not_entrant); 4608 assert(stopped(), "Should be stopped"); 4609 } 4610 validated = true; 4611 } 4612 4613 if (!stopped()) { 4614 newcopy = new_array(klass_node, length, 0); // no arguments to push 4615 4616 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true, 4617 load_object_klass(original), klass_node); 4618 if (!is_copyOfRange) { 4619 ac->set_copyof(validated); 4620 } else { 4621 ac->set_copyofrange(validated); 4622 } 4623 Node* n = _gvn.transform(ac); 4624 if (n == ac) { 4625 ac->connect_outputs(this); 4626 } else { 4627 assert(validated, "shouldn't transform if all arguments not validated"); 4628 set_all_memory(n); 4629 } 4630 } 4631 } 4632 } // original reexecute is set back here 4633 4634 C->set_has_split_ifs(true); // Has chance for split-if optimization 4635 if (!stopped()) { 4636 set_result(newcopy); 4637 } 4638 return true; 4639 } 4640 4641 4642 //----------------------generate_virtual_guard--------------------------- 4643 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 4644 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 4645 RegionNode* slow_region) { 4646 ciMethod* method = callee(); 4647 int vtable_index = method->vtable_index(); 4648 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4649 "bad index %d", vtable_index); 4650 // Get the Method* out of the appropriate vtable entry. 4651 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 4652 vtable_index*vtableEntry::size_in_bytes() + 4653 in_bytes(vtableEntry::method_offset()); 4654 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 4655 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 4656 4657 // Compare the target method with the expected method (e.g., Object.hashCode). 4658 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 4659 4660 Node* native_call = makecon(native_call_addr); 4661 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 4662 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 4663 4664 return generate_slow_guard(test_native, slow_region); 4665 } 4666 4667 //-----------------------generate_method_call---------------------------- 4668 // Use generate_method_call to make a slow-call to the real 4669 // method if the fast path fails. An alternative would be to 4670 // use a stub like OptoRuntime::slow_arraycopy_Java. 4671 // This only works for expanding the current library call, 4672 // not another intrinsic. (E.g., don't use this for making an 4673 // arraycopy call inside of the copyOf intrinsic.) 4674 CallJavaNode* 4675 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) { 4676 // When compiling the intrinsic method itself, do not use this technique. 4677 guarantee(callee() != C->method(), "cannot make slow-call to self"); 4678 4679 ciMethod* method = callee(); 4680 // ensure the JVMS we have will be correct for this call 4681 guarantee(method_id == method->intrinsic_id(), "must match"); 4682 4683 const TypeFunc* tf = TypeFunc::make(method); 4684 if (res_not_null) { 4685 assert(tf->return_type() == T_OBJECT, ""); 4686 const TypeTuple* range = tf->range(); 4687 const Type** fields = TypeTuple::fields(range->cnt()); 4688 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL); 4689 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields); 4690 tf = TypeFunc::make(tf->domain(), new_range); 4691 } 4692 CallJavaNode* slow_call; 4693 if (is_static) { 4694 assert(!is_virtual, ""); 4695 slow_call = new CallStaticJavaNode(C, tf, 4696 SharedRuntime::get_resolve_static_call_stub(), method); 4697 } else if (is_virtual) { 4698 assert(!gvn().type(argument(0))->maybe_null(), "should not be null"); 4699 int vtable_index = Method::invalid_vtable_index; 4700 if (UseInlineCaches) { 4701 // Suppress the vtable call 4702 } else { 4703 // hashCode and clone are not a miranda methods, 4704 // so the vtable index is fixed. 4705 // No need to use the linkResolver to get it. 4706 vtable_index = method->vtable_index(); 4707 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4708 "bad index %d", vtable_index); 4709 } 4710 slow_call = new CallDynamicJavaNode(tf, 4711 SharedRuntime::get_resolve_virtual_call_stub(), 4712 method, vtable_index); 4713 } else { // neither virtual nor static: opt_virtual 4714 assert(!gvn().type(argument(0))->maybe_null(), "should not be null"); 4715 slow_call = new CallStaticJavaNode(C, tf, 4716 SharedRuntime::get_resolve_opt_virtual_call_stub(), method); 4717 slow_call->set_optimized_virtual(true); 4718 } 4719 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) { 4720 // To be able to issue a direct call (optimized virtual or virtual) 4721 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information 4722 // about the method being invoked should be attached to the call site to 4723 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C). 4724 slow_call->set_override_symbolic_info(true); 4725 } 4726 set_arguments_for_java_call(slow_call); 4727 set_edges_for_java_call(slow_call); 4728 return slow_call; 4729 } 4730 4731 4732 /** 4733 * Build special case code for calls to hashCode on an object. This call may 4734 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 4735 * slightly different code. 4736 */ 4737 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 4738 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 4739 assert(!(is_virtual && is_static), "either virtual, special, or static"); 4740 4741 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 4742 4743 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4744 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 4745 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4746 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4747 Node* obj = nullptr; 4748 if (!is_static) { 4749 // Check for hashing null object 4750 obj = null_check_receiver(); 4751 if (stopped()) return true; // unconditionally null 4752 result_reg->init_req(_null_path, top()); 4753 result_val->init_req(_null_path, top()); 4754 } else { 4755 // Do a null check, and return zero if null. 4756 // System.identityHashCode(null) == 0 4757 obj = argument(0); 4758 Node* null_ctl = top(); 4759 obj = null_check_oop(obj, &null_ctl); 4760 result_reg->init_req(_null_path, null_ctl); 4761 result_val->init_req(_null_path, _gvn.intcon(0)); 4762 } 4763 4764 // Unconditionally null? Then return right away. 4765 if (stopped()) { 4766 set_control( result_reg->in(_null_path)); 4767 if (!stopped()) 4768 set_result(result_val->in(_null_path)); 4769 return true; 4770 } 4771 4772 // We only go to the fast case code if we pass a number of guards. The 4773 // paths which do not pass are accumulated in the slow_region. 4774 RegionNode* slow_region = new RegionNode(1); 4775 record_for_igvn(slow_region); 4776 4777 // If this is a virtual call, we generate a funny guard. We pull out 4778 // the vtable entry corresponding to hashCode() from the target object. 4779 // If the target method which we are calling happens to be the native 4780 // Object hashCode() method, we pass the guard. We do not need this 4781 // guard for non-virtual calls -- the caller is known to be the native 4782 // Object hashCode(). 4783 if (is_virtual) { 4784 // After null check, get the object's klass. 4785 Node* obj_klass = load_object_klass(obj); 4786 generate_virtual_guard(obj_klass, slow_region); 4787 } 4788 4789 // Get the header out of the object, use LoadMarkNode when available 4790 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 4791 // The control of the load must be null. Otherwise, the load can move before 4792 // the null check after castPP removal. 4793 Node* no_ctrl = nullptr; 4794 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 4795 4796 if (!UseObjectMonitorTable) { 4797 // Test the header to see if it is safe to read w.r.t. locking. 4798 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place); 4799 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 4800 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value); 4801 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val)); 4802 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq)); 4803 4804 generate_slow_guard(test_monitor, slow_region); 4805 } 4806 4807 // Get the hash value and check to see that it has been properly assigned. 4808 // We depend on hash_mask being at most 32 bits and avoid the use of 4809 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 4810 // vm: see markWord.hpp. 4811 Node *hash_mask = _gvn.intcon(markWord::hash_mask); 4812 Node *hash_shift = _gvn.intcon(markWord::hash_shift); 4813 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 4814 // This hack lets the hash bits live anywhere in the mark object now, as long 4815 // as the shift drops the relevant bits into the low 32 bits. Note that 4816 // Java spec says that HashCode is an int so there's no point in capturing 4817 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 4818 hshifted_header = ConvX2I(hshifted_header); 4819 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 4820 4821 Node *no_hash_val = _gvn.intcon(markWord::no_hash); 4822 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 4823 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 4824 4825 generate_slow_guard(test_assigned, slow_region); 4826 4827 Node* init_mem = reset_memory(); 4828 // fill in the rest of the null path: 4829 result_io ->init_req(_null_path, i_o()); 4830 result_mem->init_req(_null_path, init_mem); 4831 4832 result_val->init_req(_fast_path, hash_val); 4833 result_reg->init_req(_fast_path, control()); 4834 result_io ->init_req(_fast_path, i_o()); 4835 result_mem->init_req(_fast_path, init_mem); 4836 4837 // Generate code for the slow case. We make a call to hashCode(). 4838 set_control(_gvn.transform(slow_region)); 4839 if (!stopped()) { 4840 // No need for PreserveJVMState, because we're using up the present state. 4841 set_all_memory(init_mem); 4842 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 4843 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false); 4844 Node* slow_result = set_results_for_java_call(slow_call); 4845 // this->control() comes from set_results_for_java_call 4846 result_reg->init_req(_slow_path, control()); 4847 result_val->init_req(_slow_path, slow_result); 4848 result_io ->set_req(_slow_path, i_o()); 4849 result_mem ->set_req(_slow_path, reset_memory()); 4850 } 4851 4852 // Return the combined state. 4853 set_i_o( _gvn.transform(result_io) ); 4854 set_all_memory( _gvn.transform(result_mem)); 4855 4856 set_result(result_reg, result_val); 4857 return true; 4858 } 4859 4860 //---------------------------inline_native_getClass---------------------------- 4861 // public final native Class<?> java.lang.Object.getClass(); 4862 // 4863 // Build special case code for calls to getClass on an object. 4864 bool LibraryCallKit::inline_native_getClass() { 4865 Node* obj = null_check_receiver(); 4866 if (stopped()) return true; 4867 set_result(load_mirror_from_klass(load_object_klass(obj))); 4868 return true; 4869 } 4870 4871 //-----------------inline_native_Reflection_getCallerClass--------------------- 4872 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 4873 // 4874 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 4875 // 4876 // NOTE: This code must perform the same logic as JVM_GetCallerClass 4877 // in that it must skip particular security frames and checks for 4878 // caller sensitive methods. 4879 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 4880 #ifndef PRODUCT 4881 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4882 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 4883 } 4884 #endif 4885 4886 if (!jvms()->has_method()) { 4887 #ifndef PRODUCT 4888 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4889 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 4890 } 4891 #endif 4892 return false; 4893 } 4894 4895 // Walk back up the JVM state to find the caller at the required 4896 // depth. 4897 JVMState* caller_jvms = jvms(); 4898 4899 // Cf. JVM_GetCallerClass 4900 // NOTE: Start the loop at depth 1 because the current JVM state does 4901 // not include the Reflection.getCallerClass() frame. 4902 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) { 4903 ciMethod* m = caller_jvms->method(); 4904 switch (n) { 4905 case 0: 4906 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 4907 break; 4908 case 1: 4909 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 4910 if (!m->caller_sensitive()) { 4911 #ifndef PRODUCT 4912 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4913 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 4914 } 4915 #endif 4916 return false; // bail-out; let JVM_GetCallerClass do the work 4917 } 4918 break; 4919 default: 4920 if (!m->is_ignored_by_security_stack_walk()) { 4921 // We have reached the desired frame; return the holder class. 4922 // Acquire method holder as java.lang.Class and push as constant. 4923 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 4924 ciInstance* caller_mirror = caller_klass->java_mirror(); 4925 set_result(makecon(TypeInstPtr::make(caller_mirror))); 4926 4927 #ifndef PRODUCT 4928 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4929 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()); 4930 tty->print_cr(" JVM state at this point:"); 4931 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4932 ciMethod* m = jvms()->of_depth(i)->method(); 4933 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4934 } 4935 } 4936 #endif 4937 return true; 4938 } 4939 break; 4940 } 4941 } 4942 4943 #ifndef PRODUCT 4944 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4945 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 4946 tty->print_cr(" JVM state at this point:"); 4947 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4948 ciMethod* m = jvms()->of_depth(i)->method(); 4949 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4950 } 4951 } 4952 #endif 4953 4954 return false; // bail-out; let JVM_GetCallerClass do the work 4955 } 4956 4957 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 4958 Node* arg = argument(0); 4959 Node* result = nullptr; 4960 4961 switch (id) { 4962 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 4963 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 4964 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 4965 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 4966 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break; 4967 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break; 4968 4969 case vmIntrinsics::_doubleToLongBits: { 4970 // two paths (plus control) merge in a wood 4971 RegionNode *r = new RegionNode(3); 4972 Node *phi = new PhiNode(r, TypeLong::LONG); 4973 4974 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 4975 // Build the boolean node 4976 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4977 4978 // Branch either way. 4979 // NaN case is less traveled, which makes all the difference. 4980 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4981 Node *opt_isnan = _gvn.transform(ifisnan); 4982 assert( opt_isnan->is_If(), "Expect an IfNode"); 4983 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4984 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4985 4986 set_control(iftrue); 4987 4988 static const jlong nan_bits = CONST64(0x7ff8000000000000); 4989 Node *slow_result = longcon(nan_bits); // return NaN 4990 phi->init_req(1, _gvn.transform( slow_result )); 4991 r->init_req(1, iftrue); 4992 4993 // Else fall through 4994 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4995 set_control(iffalse); 4996 4997 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 4998 r->init_req(2, iffalse); 4999 5000 // Post merge 5001 set_control(_gvn.transform(r)); 5002 record_for_igvn(r); 5003 5004 C->set_has_split_ifs(true); // Has chance for split-if optimization 5005 result = phi; 5006 assert(result->bottom_type()->isa_long(), "must be"); 5007 break; 5008 } 5009 5010 case vmIntrinsics::_floatToIntBits: { 5011 // two paths (plus control) merge in a wood 5012 RegionNode *r = new RegionNode(3); 5013 Node *phi = new PhiNode(r, TypeInt::INT); 5014 5015 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 5016 // Build the boolean node 5017 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 5018 5019 // Branch either way. 5020 // NaN case is less traveled, which makes all the difference. 5021 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 5022 Node *opt_isnan = _gvn.transform(ifisnan); 5023 assert( opt_isnan->is_If(), "Expect an IfNode"); 5024 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 5025 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 5026 5027 set_control(iftrue); 5028 5029 static const jint nan_bits = 0x7fc00000; 5030 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 5031 phi->init_req(1, _gvn.transform( slow_result )); 5032 r->init_req(1, iftrue); 5033 5034 // Else fall through 5035 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 5036 set_control(iffalse); 5037 5038 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 5039 r->init_req(2, iffalse); 5040 5041 // Post merge 5042 set_control(_gvn.transform(r)); 5043 record_for_igvn(r); 5044 5045 C->set_has_split_ifs(true); // Has chance for split-if optimization 5046 result = phi; 5047 assert(result->bottom_type()->isa_int(), "must be"); 5048 break; 5049 } 5050 5051 default: 5052 fatal_unexpected_iid(id); 5053 break; 5054 } 5055 set_result(_gvn.transform(result)); 5056 return true; 5057 } 5058 5059 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) { 5060 Node* arg = argument(0); 5061 Node* result = nullptr; 5062 5063 switch (id) { 5064 case vmIntrinsics::_floatIsInfinite: 5065 result = new IsInfiniteFNode(arg); 5066 break; 5067 case vmIntrinsics::_floatIsFinite: 5068 result = new IsFiniteFNode(arg); 5069 break; 5070 case vmIntrinsics::_doubleIsInfinite: 5071 result = new IsInfiniteDNode(arg); 5072 break; 5073 case vmIntrinsics::_doubleIsFinite: 5074 result = new IsFiniteDNode(arg); 5075 break; 5076 default: 5077 fatal_unexpected_iid(id); 5078 break; 5079 } 5080 set_result(_gvn.transform(result)); 5081 return true; 5082 } 5083 5084 //----------------------inline_unsafe_copyMemory------------------------- 5085 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 5086 5087 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) { 5088 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr(); 5089 const Type* base_t = gvn.type(base); 5090 5091 bool in_native = (base_t == TypePtr::NULL_PTR); 5092 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t); 5093 bool is_mixed = !in_heap && !in_native; 5094 5095 if (is_mixed) { 5096 return true; // mixed accesses can touch both on-heap and off-heap memory 5097 } 5098 if (in_heap) { 5099 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM); 5100 if (!is_prim_array) { 5101 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array, 5102 // there's not enough type information available to determine proper memory slice for it. 5103 return true; 5104 } 5105 } 5106 return false; 5107 } 5108 5109 bool LibraryCallKit::inline_unsafe_copyMemory() { 5110 if (callee()->is_static()) return false; // caller must have the capability! 5111 null_check_receiver(); // null-check receiver 5112 if (stopped()) return true; 5113 5114 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 5115 5116 Node* src_base = argument(1); // type: oop 5117 Node* src_off = ConvL2X(argument(2)); // type: long 5118 Node* dst_base = argument(4); // type: oop 5119 Node* dst_off = ConvL2X(argument(5)); // type: long 5120 Node* size = ConvL2X(argument(7)); // type: long 5121 5122 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 5123 "fieldOffset must be byte-scaled"); 5124 5125 Node* src_addr = make_unsafe_address(src_base, src_off); 5126 Node* dst_addr = make_unsafe_address(dst_base, dst_off); 5127 5128 Node* thread = _gvn.transform(new ThreadLocalNode()); 5129 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 5130 BasicType doing_unsafe_access_bt = T_BYTE; 5131 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 5132 5133 // update volatile field 5134 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered); 5135 5136 int flags = RC_LEAF | RC_NO_FP; 5137 5138 const TypePtr* dst_type = TypePtr::BOTTOM; 5139 5140 // Adjust memory effects of the runtime call based on input values. 5141 if (!has_wide_mem(_gvn, src_addr, src_base) && 5142 !has_wide_mem(_gvn, dst_addr, dst_base)) { 5143 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory 5144 5145 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr(); 5146 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) { 5147 flags |= RC_NARROW_MEM; // narrow in memory 5148 } 5149 } 5150 5151 // Call it. Note that the length argument is not scaled. 5152 make_runtime_call(flags, 5153 OptoRuntime::fast_arraycopy_Type(), 5154 StubRoutines::unsafe_arraycopy(), 5155 "unsafe_arraycopy", 5156 dst_type, 5157 src_addr, dst_addr, size XTOP); 5158 5159 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered); 5160 5161 return true; 5162 } 5163 5164 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value); 5165 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value' 5166 bool LibraryCallKit::inline_unsafe_setMemory() { 5167 if (callee()->is_static()) return false; // caller must have the capability! 5168 null_check_receiver(); // null-check receiver 5169 if (stopped()) return true; 5170 5171 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 5172 5173 Node* dst_base = argument(1); // type: oop 5174 Node* dst_off = ConvL2X(argument(2)); // type: long 5175 Node* size = ConvL2X(argument(4)); // type: long 5176 Node* byte = argument(6); // type: byte 5177 5178 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 5179 "fieldOffset must be byte-scaled"); 5180 5181 Node* dst_addr = make_unsafe_address(dst_base, dst_off); 5182 5183 Node* thread = _gvn.transform(new ThreadLocalNode()); 5184 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 5185 BasicType doing_unsafe_access_bt = T_BYTE; 5186 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 5187 5188 // update volatile field 5189 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered); 5190 5191 int flags = RC_LEAF | RC_NO_FP; 5192 5193 const TypePtr* dst_type = TypePtr::BOTTOM; 5194 5195 // Adjust memory effects of the runtime call based on input values. 5196 if (!has_wide_mem(_gvn, dst_addr, dst_base)) { 5197 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory 5198 5199 flags |= RC_NARROW_MEM; // narrow in memory 5200 } 5201 5202 // Call it. Note that the length argument is not scaled. 5203 make_runtime_call(flags, 5204 OptoRuntime::unsafe_setmemory_Type(), 5205 StubRoutines::unsafe_setmemory(), 5206 "unsafe_setmemory", 5207 dst_type, 5208 dst_addr, size XTOP, byte); 5209 5210 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered); 5211 5212 return true; 5213 } 5214 5215 #undef XTOP 5216 5217 //------------------------clone_coping----------------------------------- 5218 // Helper function for inline_native_clone. 5219 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) { 5220 assert(obj_size != nullptr, ""); 5221 Node* raw_obj = alloc_obj->in(1); 5222 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 5223 5224 AllocateNode* alloc = nullptr; 5225 if (ReduceBulkZeroing && 5226 // If we are implementing an array clone without knowing its source type 5227 // (can happen when compiling the array-guarded branch of a reflective 5228 // Object.clone() invocation), initialize the array within the allocation. 5229 // This is needed because some GCs (e.g. ZGC) might fall back in this case 5230 // to a runtime clone call that assumes fully initialized source arrays. 5231 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) { 5232 // We will be completely responsible for initializing this object - 5233 // mark Initialize node as complete. 5234 alloc = AllocateNode::Ideal_allocation(alloc_obj); 5235 // The object was just allocated - there should be no any stores! 5236 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), ""); 5237 // Mark as complete_with_arraycopy so that on AllocateNode 5238 // expansion, we know this AllocateNode is initialized by an array 5239 // copy and a StoreStore barrier exists after the array copy. 5240 alloc->initialization()->set_complete_with_arraycopy(); 5241 } 5242 5243 Node* size = _gvn.transform(obj_size); 5244 access_clone(obj, alloc_obj, size, is_array); 5245 5246 // Do not let reads from the cloned object float above the arraycopy. 5247 if (alloc != nullptr) { 5248 // Do not let stores that initialize this object be reordered with 5249 // a subsequent store that would make this object accessible by 5250 // other threads. 5251 // Record what AllocateNode this StoreStore protects so that 5252 // escape analysis can go from the MemBarStoreStoreNode to the 5253 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 5254 // based on the escape status of the AllocateNode. 5255 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 5256 } else { 5257 insert_mem_bar(Op_MemBarCPUOrder); 5258 } 5259 } 5260 5261 //------------------------inline_native_clone---------------------------- 5262 // protected native Object java.lang.Object.clone(); 5263 // 5264 // Here are the simple edge cases: 5265 // null receiver => normal trap 5266 // virtual and clone was overridden => slow path to out-of-line clone 5267 // not cloneable or finalizer => slow path to out-of-line Object.clone 5268 // 5269 // The general case has two steps, allocation and copying. 5270 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 5271 // 5272 // Copying also has two cases, oop arrays and everything else. 5273 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 5274 // Everything else uses the tight inline loop supplied by CopyArrayNode. 5275 // 5276 // These steps fold up nicely if and when the cloned object's klass 5277 // can be sharply typed as an object array, a type array, or an instance. 5278 // 5279 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 5280 PhiNode* result_val; 5281 5282 // Set the reexecute bit for the interpreter to reexecute 5283 // the bytecode that invokes Object.clone if deoptimization happens. 5284 { PreserveReexecuteState preexecs(this); 5285 jvms()->set_should_reexecute(true); 5286 5287 Node* obj = null_check_receiver(); 5288 if (stopped()) return true; 5289 5290 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 5291 5292 // If we are going to clone an instance, we need its exact type to 5293 // know the number and types of fields to convert the clone to 5294 // loads/stores. Maybe a speculative type can help us. 5295 if (!obj_type->klass_is_exact() && 5296 obj_type->speculative_type() != nullptr && 5297 obj_type->speculative_type()->is_instance_klass()) { 5298 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 5299 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 5300 !spec_ik->has_injected_fields()) { 5301 if (!obj_type->isa_instptr() || 5302 obj_type->is_instptr()->instance_klass()->has_subklass()) { 5303 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 5304 } 5305 } 5306 } 5307 5308 // Conservatively insert a memory barrier on all memory slices. 5309 // Do not let writes into the original float below the clone. 5310 insert_mem_bar(Op_MemBarCPUOrder); 5311 5312 // paths into result_reg: 5313 enum { 5314 _slow_path = 1, // out-of-line call to clone method (virtual or not) 5315 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 5316 _array_path, // plain array allocation, plus arrayof_long_arraycopy 5317 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 5318 PATH_LIMIT 5319 }; 5320 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 5321 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 5322 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 5323 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 5324 record_for_igvn(result_reg); 5325 5326 Node* obj_klass = load_object_klass(obj); 5327 Node* array_obj = obj; 5328 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj); 5329 if (array_ctl != nullptr) { 5330 // It's an array. 5331 PreserveJVMState pjvms(this); 5332 set_control(array_ctl); 5333 Node* obj_length = load_array_length(array_obj); 5334 Node* array_size = nullptr; // Size of the array without object alignment padding. 5335 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true); 5336 5337 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 5338 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) { 5339 // If it is an oop array, it requires very special treatment, 5340 // because gc barriers are required when accessing the array. 5341 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr); 5342 if (is_obja != nullptr) { 5343 PreserveJVMState pjvms2(this); 5344 set_control(is_obja); 5345 // Generate a direct call to the right arraycopy function(s). 5346 // Clones are always tightly coupled. 5347 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false); 5348 ac->set_clone_oop_array(); 5349 Node* n = _gvn.transform(ac); 5350 assert(n == ac, "cannot disappear"); 5351 ac->connect_outputs(this, /*deoptimize_on_exception=*/true); 5352 5353 result_reg->init_req(_objArray_path, control()); 5354 result_val->init_req(_objArray_path, alloc_obj); 5355 result_i_o ->set_req(_objArray_path, i_o()); 5356 result_mem ->set_req(_objArray_path, reset_memory()); 5357 } 5358 } 5359 // Otherwise, there are no barriers to worry about. 5360 // (We can dispense with card marks if we know the allocation 5361 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 5362 // causes the non-eden paths to take compensating steps to 5363 // simulate a fresh allocation, so that no further 5364 // card marks are required in compiled code to initialize 5365 // the object.) 5366 5367 if (!stopped()) { 5368 copy_to_clone(array_obj, alloc_obj, array_size, true); 5369 5370 // Present the results of the copy. 5371 result_reg->init_req(_array_path, control()); 5372 result_val->init_req(_array_path, alloc_obj); 5373 result_i_o ->set_req(_array_path, i_o()); 5374 result_mem ->set_req(_array_path, reset_memory()); 5375 } 5376 } 5377 5378 // We only go to the instance fast case code if we pass a number of guards. 5379 // The paths which do not pass are accumulated in the slow_region. 5380 RegionNode* slow_region = new RegionNode(1); 5381 record_for_igvn(slow_region); 5382 if (!stopped()) { 5383 // It's an instance (we did array above). Make the slow-path tests. 5384 // If this is a virtual call, we generate a funny guard. We grab 5385 // the vtable entry corresponding to clone() from the target object. 5386 // If the target method which we are calling happens to be the 5387 // Object clone() method, we pass the guard. We do not need this 5388 // guard for non-virtual calls; the caller is known to be the native 5389 // Object clone(). 5390 if (is_virtual) { 5391 generate_virtual_guard(obj_klass, slow_region); 5392 } 5393 5394 // The object must be easily cloneable and must not have a finalizer. 5395 // Both of these conditions may be checked in a single test. 5396 // We could optimize the test further, but we don't care. 5397 generate_misc_flags_guard(obj_klass, 5398 // Test both conditions: 5399 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer, 5400 // Must be cloneable but not finalizer: 5401 KlassFlags::_misc_is_cloneable_fast, 5402 slow_region); 5403 } 5404 5405 if (!stopped()) { 5406 // It's an instance, and it passed the slow-path tests. 5407 PreserveJVMState pjvms(this); 5408 Node* obj_size = nullptr; // Total object size, including object alignment padding. 5409 // Need to deoptimize on exception from allocation since Object.clone intrinsic 5410 // is reexecuted if deoptimization occurs and there could be problems when merging 5411 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 5412 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true); 5413 5414 copy_to_clone(obj, alloc_obj, obj_size, false); 5415 5416 // Present the results of the slow call. 5417 result_reg->init_req(_instance_path, control()); 5418 result_val->init_req(_instance_path, alloc_obj); 5419 result_i_o ->set_req(_instance_path, i_o()); 5420 result_mem ->set_req(_instance_path, reset_memory()); 5421 } 5422 5423 // Generate code for the slow case. We make a call to clone(). 5424 set_control(_gvn.transform(slow_region)); 5425 if (!stopped()) { 5426 PreserveJVMState pjvms(this); 5427 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true); 5428 // We need to deoptimize on exception (see comment above) 5429 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true); 5430 // this->control() comes from set_results_for_java_call 5431 result_reg->init_req(_slow_path, control()); 5432 result_val->init_req(_slow_path, slow_result); 5433 result_i_o ->set_req(_slow_path, i_o()); 5434 result_mem ->set_req(_slow_path, reset_memory()); 5435 } 5436 5437 // Return the combined state. 5438 set_control( _gvn.transform(result_reg)); 5439 set_i_o( _gvn.transform(result_i_o)); 5440 set_all_memory( _gvn.transform(result_mem)); 5441 } // original reexecute is set back here 5442 5443 set_result(_gvn.transform(result_val)); 5444 return true; 5445 } 5446 5447 // If we have a tightly coupled allocation, the arraycopy may take care 5448 // of the array initialization. If one of the guards we insert between 5449 // the allocation and the arraycopy causes a deoptimization, an 5450 // uninitialized array will escape the compiled method. To prevent that 5451 // we set the JVM state for uncommon traps between the allocation and 5452 // the arraycopy to the state before the allocation so, in case of 5453 // deoptimization, we'll reexecute the allocation and the 5454 // initialization. 5455 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 5456 if (alloc != nullptr) { 5457 ciMethod* trap_method = alloc->jvms()->method(); 5458 int trap_bci = alloc->jvms()->bci(); 5459 5460 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5461 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 5462 // Make sure there's no store between the allocation and the 5463 // arraycopy otherwise visible side effects could be rexecuted 5464 // in case of deoptimization and cause incorrect execution. 5465 bool no_interfering_store = true; 5466 Node* mem = alloc->in(TypeFunc::Memory); 5467 if (mem->is_MergeMem()) { 5468 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 5469 Node* n = mms.memory(); 5470 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 5471 assert(n->is_Store(), "what else?"); 5472 no_interfering_store = false; 5473 break; 5474 } 5475 } 5476 } else { 5477 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 5478 Node* n = mms.memory(); 5479 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 5480 assert(n->is_Store(), "what else?"); 5481 no_interfering_store = false; 5482 break; 5483 } 5484 } 5485 } 5486 5487 if (no_interfering_store) { 5488 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc); 5489 5490 JVMState* saved_jvms = jvms(); 5491 saved_reexecute_sp = _reexecute_sp; 5492 5493 set_jvms(sfpt->jvms()); 5494 _reexecute_sp = jvms()->sp(); 5495 5496 return saved_jvms; 5497 } 5498 } 5499 } 5500 return nullptr; 5501 } 5502 5503 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack 5504 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter. 5505 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const { 5506 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 5507 uint size = alloc->req(); 5508 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 5509 old_jvms->set_map(sfpt); 5510 for (uint i = 0; i < size; i++) { 5511 sfpt->init_req(i, alloc->in(i)); 5512 } 5513 // re-push array length for deoptimization 5514 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength)); 5515 old_jvms->set_sp(old_jvms->sp()+1); 5516 old_jvms->set_monoff(old_jvms->monoff()+1); 5517 old_jvms->set_scloff(old_jvms->scloff()+1); 5518 old_jvms->set_endoff(old_jvms->endoff()+1); 5519 old_jvms->set_should_reexecute(true); 5520 5521 sfpt->set_i_o(map()->i_o()); 5522 sfpt->set_memory(map()->memory()); 5523 sfpt->set_control(map()->control()); 5524 return sfpt; 5525 } 5526 5527 // In case of a deoptimization, we restart execution at the 5528 // allocation, allocating a new array. We would leave an uninitialized 5529 // array in the heap that GCs wouldn't expect. Move the allocation 5530 // after the traps so we don't allocate the array if we 5531 // deoptimize. This is possible because tightly_coupled_allocation() 5532 // guarantees there's no observer of the allocated array at this point 5533 // and the control flow is simple enough. 5534 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards, 5535 int saved_reexecute_sp, uint new_idx) { 5536 if (saved_jvms_before_guards != nullptr && !stopped()) { 5537 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards); 5538 5539 assert(alloc != nullptr, "only with a tightly coupled allocation"); 5540 // restore JVM state to the state at the arraycopy 5541 saved_jvms_before_guards->map()->set_control(map()->control()); 5542 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?"); 5543 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?"); 5544 // If we've improved the types of some nodes (null check) while 5545 // emitting the guards, propagate them to the current state 5546 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx); 5547 set_jvms(saved_jvms_before_guards); 5548 _reexecute_sp = saved_reexecute_sp; 5549 5550 // Remove the allocation from above the guards 5551 CallProjections callprojs; 5552 alloc->extract_projections(&callprojs, true); 5553 InitializeNode* init = alloc->initialization(); 5554 Node* alloc_mem = alloc->in(TypeFunc::Memory); 5555 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 5556 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 5557 5558 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below 5559 // the allocation (i.e. is only valid if the allocation succeeds): 5560 // 1) replace CastIINode with AllocateArrayNode's length here 5561 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method 5562 // 5563 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate 5564 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy) 5565 Node* init_control = init->proj_out(TypeFunc::Control); 5566 Node* alloc_length = alloc->Ideal_length(); 5567 #ifdef ASSERT 5568 Node* prev_cast = nullptr; 5569 #endif 5570 for (uint i = 0; i < init_control->outcnt(); i++) { 5571 Node* init_out = init_control->raw_out(i); 5572 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) { 5573 #ifdef ASSERT 5574 if (prev_cast == nullptr) { 5575 prev_cast = init_out; 5576 } else { 5577 if (prev_cast->cmp(*init_out) == false) { 5578 prev_cast->dump(); 5579 init_out->dump(); 5580 assert(false, "not equal CastIINode"); 5581 } 5582 } 5583 #endif 5584 C->gvn_replace_by(init_out, alloc_length); 5585 } 5586 } 5587 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 5588 5589 // move the allocation here (after the guards) 5590 _gvn.hash_delete(alloc); 5591 alloc->set_req(TypeFunc::Control, control()); 5592 alloc->set_req(TypeFunc::I_O, i_o()); 5593 Node *mem = reset_memory(); 5594 set_all_memory(mem); 5595 alloc->set_req(TypeFunc::Memory, mem); 5596 set_control(init->proj_out_or_null(TypeFunc::Control)); 5597 set_i_o(callprojs.fallthrough_ioproj); 5598 5599 // Update memory as done in GraphKit::set_output_for_allocation() 5600 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 5601 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 5602 if (ary_type->isa_aryptr() && length_type != nullptr) { 5603 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 5604 } 5605 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 5606 int elemidx = C->get_alias_index(telemref); 5607 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw); 5608 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx); 5609 5610 Node* allocx = _gvn.transform(alloc); 5611 assert(allocx == alloc, "where has the allocation gone?"); 5612 assert(dest->is_CheckCastPP(), "not an allocation result?"); 5613 5614 _gvn.hash_delete(dest); 5615 dest->set_req(0, control()); 5616 Node* destx = _gvn.transform(dest); 5617 assert(destx == dest, "where has the allocation result gone?"); 5618 5619 array_ideal_length(alloc, ary_type, true); 5620 } 5621 } 5622 5623 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(), 5624 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary 5625 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array 5626 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter, 5627 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in 5628 // the interpreter similar to what we are doing for the newly emitted guards for the array copy. 5629 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc, 5630 JVMState* saved_jvms_before_guards) { 5631 if (saved_jvms_before_guards->map()->control()->is_IfProj()) { 5632 // There is at least one unrelated uncommon trap which needs to be replaced. 5633 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc); 5634 5635 JVMState* saved_jvms = jvms(); 5636 const int saved_reexecute_sp = _reexecute_sp; 5637 set_jvms(sfpt->jvms()); 5638 _reexecute_sp = jvms()->sp(); 5639 5640 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards); 5641 5642 // Restore state 5643 set_jvms(saved_jvms); 5644 _reexecute_sp = saved_reexecute_sp; 5645 } 5646 } 5647 5648 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon 5649 // traps will have the state of the array allocation. Let the old uncommon trap nodes die. 5650 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) { 5651 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards 5652 while (if_proj->is_IfProj()) { 5653 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj); 5654 if (uncommon_trap != nullptr) { 5655 create_new_uncommon_trap(uncommon_trap); 5656 } 5657 assert(if_proj->in(0)->is_If(), "must be If"); 5658 if_proj = if_proj->in(0)->in(0); 5659 } 5660 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(), 5661 "must have reached control projection of init node"); 5662 } 5663 5664 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) { 5665 const int trap_request = uncommon_trap_call->uncommon_trap_request(); 5666 assert(trap_request != 0, "no valid UCT trap request"); 5667 PreserveJVMState pjvms(this); 5668 set_control(uncommon_trap_call->in(0)); 5669 uncommon_trap(Deoptimization::trap_request_reason(trap_request), 5670 Deoptimization::trap_request_action(trap_request)); 5671 assert(stopped(), "Should be stopped"); 5672 _gvn.hash_delete(uncommon_trap_call); 5673 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it 5674 } 5675 5676 // Common checks for array sorting intrinsics arguments. 5677 // Returns `true` if checks passed. 5678 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) { 5679 // check address of the class 5680 if (elementType == nullptr || elementType->is_top()) { 5681 return false; // dead path 5682 } 5683 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr(); 5684 if (elem_klass == nullptr) { 5685 return false; // dead path 5686 } 5687 // java_mirror_type() returns non-null for compile-time Class constants only 5688 ciType* elem_type = elem_klass->java_mirror_type(); 5689 if (elem_type == nullptr) { 5690 return false; 5691 } 5692 bt = elem_type->basic_type(); 5693 // Disable the intrinsic if the CPU does not support SIMD sort 5694 if (!Matcher::supports_simd_sort(bt)) { 5695 return false; 5696 } 5697 // check address of the array 5698 if (obj == nullptr || obj->is_top()) { 5699 return false; // dead path 5700 } 5701 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr(); 5702 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) { 5703 return false; // failed input validation 5704 } 5705 return true; 5706 } 5707 5708 //------------------------------inline_array_partition----------------------- 5709 bool LibraryCallKit::inline_array_partition() { 5710 address stubAddr = StubRoutines::select_array_partition_function(); 5711 if (stubAddr == nullptr) { 5712 return false; // Intrinsic's stub is not implemented on this platform 5713 } 5714 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)"); 5715 5716 // no receiver because it is a static method 5717 Node* elementType = argument(0); 5718 Node* obj = argument(1); 5719 Node* offset = argument(2); // long 5720 Node* fromIndex = argument(4); 5721 Node* toIndex = argument(5); 5722 Node* indexPivot1 = argument(6); 5723 Node* indexPivot2 = argument(7); 5724 // PartitionOperation: argument(8) is ignored 5725 5726 Node* pivotIndices = nullptr; 5727 BasicType bt = T_ILLEGAL; 5728 5729 if (!check_array_sort_arguments(elementType, obj, bt)) { 5730 return false; 5731 } 5732 null_check(obj); 5733 // If obj is dead, only null-path is taken. 5734 if (stopped()) { 5735 return true; 5736 } 5737 // Set the original stack and the reexecute bit for the interpreter to reexecute 5738 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens. 5739 { PreserveReexecuteState preexecs(this); 5740 jvms()->set_should_reexecute(true); 5741 5742 Node* obj_adr = make_unsafe_address(obj, offset); 5743 5744 // create the pivotIndices array of type int and size = 2 5745 Node* size = intcon(2); 5746 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT))); 5747 pivotIndices = new_array(klass_node, size, 0); // no arguments to push 5748 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices); 5749 guarantee(alloc != nullptr, "created above"); 5750 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT)); 5751 5752 // pass the basic type enum to the stub 5753 Node* elemType = intcon(bt); 5754 5755 // Call the stub 5756 const char *stubName = "array_partition_stub"; 5757 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(), 5758 stubAddr, stubName, TypePtr::BOTTOM, 5759 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr, 5760 indexPivot1, indexPivot2); 5761 5762 } // original reexecute is set back here 5763 5764 if (!stopped()) { 5765 set_result(pivotIndices); 5766 } 5767 5768 return true; 5769 } 5770 5771 5772 //------------------------------inline_array_sort----------------------- 5773 bool LibraryCallKit::inline_array_sort() { 5774 address stubAddr = StubRoutines::select_arraysort_function(); 5775 if (stubAddr == nullptr) { 5776 return false; // Intrinsic's stub is not implemented on this platform 5777 } 5778 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)"); 5779 5780 // no receiver because it is a static method 5781 Node* elementType = argument(0); 5782 Node* obj = argument(1); 5783 Node* offset = argument(2); // long 5784 Node* fromIndex = argument(4); 5785 Node* toIndex = argument(5); 5786 // SortOperation: argument(6) is ignored 5787 5788 BasicType bt = T_ILLEGAL; 5789 5790 if (!check_array_sort_arguments(elementType, obj, bt)) { 5791 return false; 5792 } 5793 null_check(obj); 5794 // If obj is dead, only null-path is taken. 5795 if (stopped()) { 5796 return true; 5797 } 5798 Node* obj_adr = make_unsafe_address(obj, offset); 5799 5800 // pass the basic type enum to the stub 5801 Node* elemType = intcon(bt); 5802 5803 // Call the stub. 5804 const char *stubName = "arraysort_stub"; 5805 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(), 5806 stubAddr, stubName, TypePtr::BOTTOM, 5807 obj_adr, elemType, fromIndex, toIndex); 5808 5809 return true; 5810 } 5811 5812 5813 //------------------------------inline_arraycopy----------------------- 5814 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 5815 // Object dest, int destPos, 5816 // int length); 5817 bool LibraryCallKit::inline_arraycopy() { 5818 // Get the arguments. 5819 Node* src = argument(0); // type: oop 5820 Node* src_offset = argument(1); // type: int 5821 Node* dest = argument(2); // type: oop 5822 Node* dest_offset = argument(3); // type: int 5823 Node* length = argument(4); // type: int 5824 5825 uint new_idx = C->unique(); 5826 5827 // Check for allocation before we add nodes that would confuse 5828 // tightly_coupled_allocation() 5829 AllocateArrayNode* alloc = tightly_coupled_allocation(dest); 5830 5831 int saved_reexecute_sp = -1; 5832 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 5833 // See arraycopy_restore_alloc_state() comment 5834 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards 5835 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation 5836 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards 5837 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr); 5838 5839 // The following tests must be performed 5840 // (1) src and dest are arrays. 5841 // (2) src and dest arrays must have elements of the same BasicType 5842 // (3) src and dest must not be null. 5843 // (4) src_offset must not be negative. 5844 // (5) dest_offset must not be negative. 5845 // (6) length must not be negative. 5846 // (7) src_offset + length must not exceed length of src. 5847 // (8) dest_offset + length must not exceed length of dest. 5848 // (9) each element of an oop array must be assignable 5849 5850 // (3) src and dest must not be null. 5851 // always do this here because we need the JVM state for uncommon traps 5852 Node* null_ctl = top(); 5853 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 5854 assert(null_ctl->is_top(), "no null control here"); 5855 dest = null_check(dest, T_ARRAY); 5856 5857 if (!can_emit_guards) { 5858 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any 5859 // guards but the arraycopy node could still take advantage of a 5860 // tightly allocated allocation. tightly_coupled_allocation() is 5861 // called again to make sure it takes the null check above into 5862 // account: the null check is mandatory and if it caused an 5863 // uncommon trap to be emitted then the allocation can't be 5864 // considered tightly coupled in this context. 5865 alloc = tightly_coupled_allocation(dest); 5866 } 5867 5868 bool validated = false; 5869 5870 const Type* src_type = _gvn.type(src); 5871 const Type* dest_type = _gvn.type(dest); 5872 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5873 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5874 5875 // Do we have the type of src? 5876 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM); 5877 // Do we have the type of dest? 5878 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM); 5879 // Is the type for src from speculation? 5880 bool src_spec = false; 5881 // Is the type for dest from speculation? 5882 bool dest_spec = false; 5883 5884 if ((!has_src || !has_dest) && can_emit_guards) { 5885 // We don't have sufficient type information, let's see if 5886 // speculative types can help. We need to have types for both src 5887 // and dest so that it pays off. 5888 5889 // Do we already have or could we have type information for src 5890 bool could_have_src = has_src; 5891 // Do we already have or could we have type information for dest 5892 bool could_have_dest = has_dest; 5893 5894 ciKlass* src_k = nullptr; 5895 if (!has_src) { 5896 src_k = src_type->speculative_type_not_null(); 5897 if (src_k != nullptr && src_k->is_array_klass()) { 5898 could_have_src = true; 5899 } 5900 } 5901 5902 ciKlass* dest_k = nullptr; 5903 if (!has_dest) { 5904 dest_k = dest_type->speculative_type_not_null(); 5905 if (dest_k != nullptr && dest_k->is_array_klass()) { 5906 could_have_dest = true; 5907 } 5908 } 5909 5910 if (could_have_src && could_have_dest) { 5911 // This is going to pay off so emit the required guards 5912 if (!has_src) { 5913 src = maybe_cast_profiled_obj(src, src_k, true); 5914 src_type = _gvn.type(src); 5915 top_src = src_type->isa_aryptr(); 5916 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM); 5917 src_spec = true; 5918 } 5919 if (!has_dest) { 5920 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5921 dest_type = _gvn.type(dest); 5922 top_dest = dest_type->isa_aryptr(); 5923 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM); 5924 dest_spec = true; 5925 } 5926 } 5927 } 5928 5929 if (has_src && has_dest && can_emit_guards) { 5930 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type(); 5931 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type(); 5932 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT; 5933 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT; 5934 5935 if (src_elem == dest_elem && src_elem == T_OBJECT) { 5936 // If both arrays are object arrays then having the exact types 5937 // for both will remove the need for a subtype check at runtime 5938 // before the call and may make it possible to pick a faster copy 5939 // routine (without a subtype check on every element) 5940 // Do we have the exact type of src? 5941 bool could_have_src = src_spec; 5942 // Do we have the exact type of dest? 5943 bool could_have_dest = dest_spec; 5944 ciKlass* src_k = nullptr; 5945 ciKlass* dest_k = nullptr; 5946 if (!src_spec) { 5947 src_k = src_type->speculative_type_not_null(); 5948 if (src_k != nullptr && src_k->is_array_klass()) { 5949 could_have_src = true; 5950 } 5951 } 5952 if (!dest_spec) { 5953 dest_k = dest_type->speculative_type_not_null(); 5954 if (dest_k != nullptr && dest_k->is_array_klass()) { 5955 could_have_dest = true; 5956 } 5957 } 5958 if (could_have_src && could_have_dest) { 5959 // If we can have both exact types, emit the missing guards 5960 if (could_have_src && !src_spec) { 5961 src = maybe_cast_profiled_obj(src, src_k, true); 5962 } 5963 if (could_have_dest && !dest_spec) { 5964 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5965 } 5966 } 5967 } 5968 } 5969 5970 ciMethod* trap_method = method(); 5971 int trap_bci = bci(); 5972 if (saved_jvms_before_guards != nullptr) { 5973 trap_method = alloc->jvms()->method(); 5974 trap_bci = alloc->jvms()->bci(); 5975 } 5976 5977 bool negative_length_guard_generated = false; 5978 5979 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5980 can_emit_guards && 5981 !src->is_top() && !dest->is_top()) { 5982 // validate arguments: enables transformation the ArrayCopyNode 5983 validated = true; 5984 5985 RegionNode* slow_region = new RegionNode(1); 5986 record_for_igvn(slow_region); 5987 5988 // (1) src and dest are arrays. 5989 generate_non_array_guard(load_object_klass(src), slow_region, &src); 5990 generate_non_array_guard(load_object_klass(dest), slow_region, &dest); 5991 5992 // (2) src and dest arrays must have elements of the same BasicType 5993 // done at macro expansion or at Ideal transformation time 5994 5995 // (4) src_offset must not be negative. 5996 generate_negative_guard(src_offset, slow_region); 5997 5998 // (5) dest_offset must not be negative. 5999 generate_negative_guard(dest_offset, slow_region); 6000 6001 // (7) src_offset + length must not exceed length of src. 6002 generate_limit_guard(src_offset, length, 6003 load_array_length(src), 6004 slow_region); 6005 6006 // (8) dest_offset + length must not exceed length of dest. 6007 generate_limit_guard(dest_offset, length, 6008 load_array_length(dest), 6009 slow_region); 6010 6011 // (6) length must not be negative. 6012 // This is also checked in generate_arraycopy() during macro expansion, but 6013 // we also have to check it here for the case where the ArrayCopyNode will 6014 // be eliminated by Escape Analysis. 6015 if (EliminateAllocations) { 6016 generate_negative_guard(length, slow_region); 6017 negative_length_guard_generated = true; 6018 } 6019 6020 // (9) each element of an oop array must be assignable 6021 Node* dest_klass = load_object_klass(dest); 6022 if (src != dest) { 6023 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass); 6024 6025 if (not_subtype_ctrl != top()) { 6026 PreserveJVMState pjvms(this); 6027 set_control(not_subtype_ctrl); 6028 uncommon_trap(Deoptimization::Reason_intrinsic, 6029 Deoptimization::Action_make_not_entrant); 6030 assert(stopped(), "Should be stopped"); 6031 } 6032 } 6033 { 6034 PreserveJVMState pjvms(this); 6035 set_control(_gvn.transform(slow_region)); 6036 uncommon_trap(Deoptimization::Reason_intrinsic, 6037 Deoptimization::Action_make_not_entrant); 6038 assert(stopped(), "Should be stopped"); 6039 } 6040 6041 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr(); 6042 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type(); 6043 src = _gvn.transform(new CheckCastPPNode(control(), src, toop)); 6044 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx); 6045 } 6046 6047 if (stopped()) { 6048 return true; 6049 } 6050 6051 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated, 6052 // Create LoadRange and LoadKlass nodes for use during macro expansion here 6053 // so the compiler has a chance to eliminate them: during macro expansion, 6054 // we have to set their control (CastPP nodes are eliminated). 6055 load_object_klass(src), load_object_klass(dest), 6056 load_array_length(src), load_array_length(dest)); 6057 6058 ac->set_arraycopy(validated); 6059 6060 Node* n = _gvn.transform(ac); 6061 if (n == ac) { 6062 ac->connect_outputs(this); 6063 } else { 6064 assert(validated, "shouldn't transform if all arguments not validated"); 6065 set_all_memory(n); 6066 } 6067 clear_upper_avx(); 6068 6069 6070 return true; 6071 } 6072 6073 6074 // Helper function which determines if an arraycopy immediately follows 6075 // an allocation, with no intervening tests or other escapes for the object. 6076 AllocateArrayNode* 6077 LibraryCallKit::tightly_coupled_allocation(Node* ptr) { 6078 if (stopped()) return nullptr; // no fast path 6079 if (!C->do_aliasing()) return nullptr; // no MergeMems around 6080 6081 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr); 6082 if (alloc == nullptr) return nullptr; 6083 6084 Node* rawmem = memory(Compile::AliasIdxRaw); 6085 // Is the allocation's memory state untouched? 6086 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 6087 // Bail out if there have been raw-memory effects since the allocation. 6088 // (Example: There might have been a call or safepoint.) 6089 return nullptr; 6090 } 6091 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 6092 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 6093 return nullptr; 6094 } 6095 6096 // There must be no unexpected observers of this allocation. 6097 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 6098 Node* obs = ptr->fast_out(i); 6099 if (obs != this->map()) { 6100 return nullptr; 6101 } 6102 } 6103 6104 // This arraycopy must unconditionally follow the allocation of the ptr. 6105 Node* alloc_ctl = ptr->in(0); 6106 Node* ctl = control(); 6107 while (ctl != alloc_ctl) { 6108 // There may be guards which feed into the slow_region. 6109 // Any other control flow means that we might not get a chance 6110 // to finish initializing the allocated object. 6111 // Various low-level checks bottom out in uncommon traps. These 6112 // are considered safe since we've already checked above that 6113 // there is no unexpected observer of this allocation. 6114 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) { 6115 assert(ctl->in(0)->is_If(), "must be If"); 6116 ctl = ctl->in(0)->in(0); 6117 } else { 6118 return nullptr; 6119 } 6120 } 6121 6122 // If we get this far, we have an allocation which immediately 6123 // precedes the arraycopy, and we can take over zeroing the new object. 6124 // The arraycopy will finish the initialization, and provide 6125 // a new control state to which we will anchor the destination pointer. 6126 6127 return alloc; 6128 } 6129 6130 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) { 6131 if (node->is_IfProj()) { 6132 Node* other_proj = node->as_IfProj()->other_if_proj(); 6133 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) { 6134 Node* obs = other_proj->fast_out(j); 6135 if (obs->in(0) == other_proj && obs->is_CallStaticJava() && 6136 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) { 6137 return obs->as_CallStaticJava(); 6138 } 6139 } 6140 } 6141 return nullptr; 6142 } 6143 6144 //-------------inline_encodeISOArray----------------------------------- 6145 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len) 6146 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len) 6147 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len) 6148 // encode char[] to byte[] in ISO_8859_1 or ASCII 6149 bool LibraryCallKit::inline_encodeISOArray(bool ascii) { 6150 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 6151 // no receiver since it is static method 6152 Node *src = argument(0); 6153 Node *src_offset = argument(1); 6154 Node *dst = argument(2); 6155 Node *dst_offset = argument(3); 6156 Node *length = argument(4); 6157 6158 // Cast source & target arrays to not-null 6159 if (VerifyIntrinsicChecks) { 6160 src = must_be_not_null(src, true); 6161 dst = must_be_not_null(dst, true); 6162 if (stopped()) { 6163 return true; 6164 } 6165 } 6166 6167 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6168 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr(); 6169 if (src_type == nullptr || src_type->elem() == Type::BOTTOM || 6170 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) { 6171 // failed array check 6172 return false; 6173 } 6174 6175 // Figure out the size and type of the elements we will be copying. 6176 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6177 BasicType dst_elem = dst_type->elem()->array_element_basic_type(); 6178 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 6179 return false; 6180 } 6181 6182 // Check source & target bounds 6183 if (VerifyIntrinsicChecks) { 6184 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true); 6185 generate_string_range_check(dst, dst_offset, length, false, true); 6186 if (stopped()) { 6187 return true; 6188 } 6189 } 6190 6191 Node* src_start = array_element_address(src, src_offset, T_CHAR); 6192 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 6193 // 'src_start' points to src array + scaled offset 6194 // 'dst_start' points to dst array + scaled offset 6195 6196 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 6197 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii); 6198 enc = _gvn.transform(enc); 6199 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 6200 set_memory(res_mem, mtype); 6201 set_result(enc); 6202 clear_upper_avx(); 6203 6204 return true; 6205 } 6206 6207 //-------------inline_multiplyToLen----------------------------------- 6208 bool LibraryCallKit::inline_multiplyToLen() { 6209 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 6210 6211 address stubAddr = StubRoutines::multiplyToLen(); 6212 if (stubAddr == nullptr) { 6213 return false; // Intrinsic's stub is not implemented on this platform 6214 } 6215 const char* stubName = "multiplyToLen"; 6216 6217 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 6218 6219 // no receiver because it is a static method 6220 Node* x = argument(0); 6221 Node* xlen = argument(1); 6222 Node* y = argument(2); 6223 Node* ylen = argument(3); 6224 Node* z = argument(4); 6225 6226 x = must_be_not_null(x, true); 6227 y = must_be_not_null(y, true); 6228 6229 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr(); 6230 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr(); 6231 if (x_type == nullptr || x_type->elem() == Type::BOTTOM || 6232 y_type == nullptr || y_type->elem() == Type::BOTTOM) { 6233 // failed array check 6234 return false; 6235 } 6236 6237 BasicType x_elem = x_type->elem()->array_element_basic_type(); 6238 BasicType y_elem = y_type->elem()->array_element_basic_type(); 6239 if (x_elem != T_INT || y_elem != T_INT) { 6240 return false; 6241 } 6242 6243 Node* x_start = array_element_address(x, intcon(0), x_elem); 6244 Node* y_start = array_element_address(y, intcon(0), y_elem); 6245 // 'x_start' points to x array + scaled xlen 6246 // 'y_start' points to y array + scaled ylen 6247 6248 Node* z_start = array_element_address(z, intcon(0), T_INT); 6249 6250 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6251 OptoRuntime::multiplyToLen_Type(), 6252 stubAddr, stubName, TypePtr::BOTTOM, 6253 x_start, xlen, y_start, ylen, z_start); 6254 6255 C->set_has_split_ifs(true); // Has chance for split-if optimization 6256 set_result(z); 6257 return true; 6258 } 6259 6260 //-------------inline_squareToLen------------------------------------ 6261 bool LibraryCallKit::inline_squareToLen() { 6262 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 6263 6264 address stubAddr = StubRoutines::squareToLen(); 6265 if (stubAddr == nullptr) { 6266 return false; // Intrinsic's stub is not implemented on this platform 6267 } 6268 const char* stubName = "squareToLen"; 6269 6270 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 6271 6272 Node* x = argument(0); 6273 Node* len = argument(1); 6274 Node* z = argument(2); 6275 Node* zlen = argument(3); 6276 6277 x = must_be_not_null(x, true); 6278 z = must_be_not_null(z, true); 6279 6280 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr(); 6281 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr(); 6282 if (x_type == nullptr || x_type->elem() == Type::BOTTOM || 6283 z_type == nullptr || z_type->elem() == Type::BOTTOM) { 6284 // failed array check 6285 return false; 6286 } 6287 6288 BasicType x_elem = x_type->elem()->array_element_basic_type(); 6289 BasicType z_elem = z_type->elem()->array_element_basic_type(); 6290 if (x_elem != T_INT || z_elem != T_INT) { 6291 return false; 6292 } 6293 6294 6295 Node* x_start = array_element_address(x, intcon(0), x_elem); 6296 Node* z_start = array_element_address(z, intcon(0), z_elem); 6297 6298 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6299 OptoRuntime::squareToLen_Type(), 6300 stubAddr, stubName, TypePtr::BOTTOM, 6301 x_start, len, z_start, zlen); 6302 6303 set_result(z); 6304 return true; 6305 } 6306 6307 //-------------inline_mulAdd------------------------------------------ 6308 bool LibraryCallKit::inline_mulAdd() { 6309 assert(UseMulAddIntrinsic, "not implemented on this platform"); 6310 6311 address stubAddr = StubRoutines::mulAdd(); 6312 if (stubAddr == nullptr) { 6313 return false; // Intrinsic's stub is not implemented on this platform 6314 } 6315 const char* stubName = "mulAdd"; 6316 6317 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 6318 6319 Node* out = argument(0); 6320 Node* in = argument(1); 6321 Node* offset = argument(2); 6322 Node* len = argument(3); 6323 Node* k = argument(4); 6324 6325 in = must_be_not_null(in, true); 6326 out = must_be_not_null(out, true); 6327 6328 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr(); 6329 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr(); 6330 if (out_type == nullptr || out_type->elem() == Type::BOTTOM || 6331 in_type == nullptr || in_type->elem() == Type::BOTTOM) { 6332 // failed array check 6333 return false; 6334 } 6335 6336 BasicType out_elem = out_type->elem()->array_element_basic_type(); 6337 BasicType in_elem = in_type->elem()->array_element_basic_type(); 6338 if (out_elem != T_INT || in_elem != T_INT) { 6339 return false; 6340 } 6341 6342 Node* outlen = load_array_length(out); 6343 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 6344 Node* out_start = array_element_address(out, intcon(0), out_elem); 6345 Node* in_start = array_element_address(in, intcon(0), in_elem); 6346 6347 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6348 OptoRuntime::mulAdd_Type(), 6349 stubAddr, stubName, TypePtr::BOTTOM, 6350 out_start,in_start, new_offset, len, k); 6351 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6352 set_result(result); 6353 return true; 6354 } 6355 6356 //-------------inline_montgomeryMultiply----------------------------------- 6357 bool LibraryCallKit::inline_montgomeryMultiply() { 6358 address stubAddr = StubRoutines::montgomeryMultiply(); 6359 if (stubAddr == nullptr) { 6360 return false; // Intrinsic's stub is not implemented on this platform 6361 } 6362 6363 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 6364 const char* stubName = "montgomery_multiply"; 6365 6366 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 6367 6368 Node* a = argument(0); 6369 Node* b = argument(1); 6370 Node* n = argument(2); 6371 Node* len = argument(3); 6372 Node* inv = argument(4); 6373 Node* m = argument(6); 6374 6375 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr(); 6376 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr(); 6377 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr(); 6378 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr(); 6379 if (a_type == nullptr || a_type->elem() == Type::BOTTOM || 6380 b_type == nullptr || b_type->elem() == Type::BOTTOM || 6381 n_type == nullptr || n_type->elem() == Type::BOTTOM || 6382 m_type == nullptr || m_type->elem() == Type::BOTTOM) { 6383 // failed array check 6384 return false; 6385 } 6386 6387 BasicType a_elem = a_type->elem()->array_element_basic_type(); 6388 BasicType b_elem = b_type->elem()->array_element_basic_type(); 6389 BasicType n_elem = n_type->elem()->array_element_basic_type(); 6390 BasicType m_elem = m_type->elem()->array_element_basic_type(); 6391 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 6392 return false; 6393 } 6394 6395 // Make the call 6396 { 6397 Node* a_start = array_element_address(a, intcon(0), a_elem); 6398 Node* b_start = array_element_address(b, intcon(0), b_elem); 6399 Node* n_start = array_element_address(n, intcon(0), n_elem); 6400 Node* m_start = array_element_address(m, intcon(0), m_elem); 6401 6402 Node* call = make_runtime_call(RC_LEAF, 6403 OptoRuntime::montgomeryMultiply_Type(), 6404 stubAddr, stubName, TypePtr::BOTTOM, 6405 a_start, b_start, n_start, len, inv, top(), 6406 m_start); 6407 set_result(m); 6408 } 6409 6410 return true; 6411 } 6412 6413 bool LibraryCallKit::inline_montgomerySquare() { 6414 address stubAddr = StubRoutines::montgomerySquare(); 6415 if (stubAddr == nullptr) { 6416 return false; // Intrinsic's stub is not implemented on this platform 6417 } 6418 6419 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 6420 const char* stubName = "montgomery_square"; 6421 6422 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 6423 6424 Node* a = argument(0); 6425 Node* n = argument(1); 6426 Node* len = argument(2); 6427 Node* inv = argument(3); 6428 Node* m = argument(5); 6429 6430 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr(); 6431 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr(); 6432 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr(); 6433 if (a_type == nullptr || a_type->elem() == Type::BOTTOM || 6434 n_type == nullptr || n_type->elem() == Type::BOTTOM || 6435 m_type == nullptr || m_type->elem() == Type::BOTTOM) { 6436 // failed array check 6437 return false; 6438 } 6439 6440 BasicType a_elem = a_type->elem()->array_element_basic_type(); 6441 BasicType n_elem = n_type->elem()->array_element_basic_type(); 6442 BasicType m_elem = m_type->elem()->array_element_basic_type(); 6443 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 6444 return false; 6445 } 6446 6447 // Make the call 6448 { 6449 Node* a_start = array_element_address(a, intcon(0), a_elem); 6450 Node* n_start = array_element_address(n, intcon(0), n_elem); 6451 Node* m_start = array_element_address(m, intcon(0), m_elem); 6452 6453 Node* call = make_runtime_call(RC_LEAF, 6454 OptoRuntime::montgomerySquare_Type(), 6455 stubAddr, stubName, TypePtr::BOTTOM, 6456 a_start, n_start, len, inv, top(), 6457 m_start); 6458 set_result(m); 6459 } 6460 6461 return true; 6462 } 6463 6464 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) { 6465 address stubAddr = nullptr; 6466 const char* stubName = nullptr; 6467 6468 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift(); 6469 if (stubAddr == nullptr) { 6470 return false; // Intrinsic's stub is not implemented on this platform 6471 } 6472 6473 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker"; 6474 6475 assert(callee()->signature()->size() == 5, "expected 5 arguments"); 6476 6477 Node* newArr = argument(0); 6478 Node* oldArr = argument(1); 6479 Node* newIdx = argument(2); 6480 Node* shiftCount = argument(3); 6481 Node* numIter = argument(4); 6482 6483 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr(); 6484 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr(); 6485 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM || 6486 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) { 6487 return false; 6488 } 6489 6490 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type(); 6491 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type(); 6492 if (newArr_elem != T_INT || oldArr_elem != T_INT) { 6493 return false; 6494 } 6495 6496 // Make the call 6497 { 6498 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem); 6499 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem); 6500 6501 Node* call = make_runtime_call(RC_LEAF, 6502 OptoRuntime::bigIntegerShift_Type(), 6503 stubAddr, 6504 stubName, 6505 TypePtr::BOTTOM, 6506 newArr_start, 6507 oldArr_start, 6508 newIdx, 6509 shiftCount, 6510 numIter); 6511 } 6512 6513 return true; 6514 } 6515 6516 //-------------inline_vectorizedMismatch------------------------------ 6517 bool LibraryCallKit::inline_vectorizedMismatch() { 6518 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform"); 6519 6520 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 6521 Node* obja = argument(0); // Object 6522 Node* aoffset = argument(1); // long 6523 Node* objb = argument(3); // Object 6524 Node* boffset = argument(4); // long 6525 Node* length = argument(6); // int 6526 Node* scale = argument(7); // int 6527 6528 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr(); 6529 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr(); 6530 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM || 6531 objb_t == nullptr || objb_t->elem() == Type::BOTTOM || 6532 scale == top()) { 6533 return false; // failed input validation 6534 } 6535 6536 Node* obja_adr = make_unsafe_address(obja, aoffset); 6537 Node* objb_adr = make_unsafe_address(objb, boffset); 6538 6539 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size. 6540 // 6541 // inline_limit = ArrayOperationPartialInlineSize / element_size; 6542 // if (length <= inline_limit) { 6543 // inline_path: 6544 // vmask = VectorMaskGen length 6545 // vload1 = LoadVectorMasked obja, vmask 6546 // vload2 = LoadVectorMasked objb, vmask 6547 // result1 = VectorCmpMasked vload1, vload2, vmask 6548 // } else { 6549 // call_stub_path: 6550 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale) 6551 // } 6552 // exit_block: 6553 // return Phi(result1, result2); 6554 // 6555 enum { inline_path = 1, // input is small enough to process it all at once 6556 stub_path = 2, // input is too large; call into the VM 6557 PATH_LIMIT = 3 6558 }; 6559 6560 Node* exit_block = new RegionNode(PATH_LIMIT); 6561 Node* result_phi = new PhiNode(exit_block, TypeInt::INT); 6562 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM); 6563 6564 Node* call_stub_path = control(); 6565 6566 BasicType elem_bt = T_ILLEGAL; 6567 6568 const TypeInt* scale_t = _gvn.type(scale)->is_int(); 6569 if (scale_t->is_con()) { 6570 switch (scale_t->get_con()) { 6571 case 0: elem_bt = T_BYTE; break; 6572 case 1: elem_bt = T_SHORT; break; 6573 case 2: elem_bt = T_INT; break; 6574 case 3: elem_bt = T_LONG; break; 6575 6576 default: elem_bt = T_ILLEGAL; break; // not supported 6577 } 6578 } 6579 6580 int inline_limit = 0; 6581 bool do_partial_inline = false; 6582 6583 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) { 6584 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt); 6585 do_partial_inline = inline_limit >= 16; 6586 } 6587 6588 if (do_partial_inline) { 6589 assert(elem_bt != T_ILLEGAL, "sanity"); 6590 6591 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) && 6592 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) && 6593 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) { 6594 6595 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit); 6596 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit))); 6597 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt)); 6598 6599 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN); 6600 6601 if (!stopped()) { 6602 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin))); 6603 6604 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr(); 6605 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr(); 6606 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t)); 6607 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t)); 6608 6609 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt)); 6610 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask)); 6611 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask)); 6612 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT)); 6613 6614 exit_block->init_req(inline_path, control()); 6615 memory_phi->init_req(inline_path, map()->memory()); 6616 result_phi->init_req(inline_path, result); 6617 6618 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size())); 6619 clear_upper_avx(); 6620 } 6621 } 6622 } 6623 6624 if (call_stub_path != nullptr) { 6625 set_control(call_stub_path); 6626 6627 Node* call = make_runtime_call(RC_LEAF, 6628 OptoRuntime::vectorizedMismatch_Type(), 6629 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM, 6630 obja_adr, objb_adr, length, scale); 6631 6632 exit_block->init_req(stub_path, control()); 6633 memory_phi->init_req(stub_path, map()->memory()); 6634 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms))); 6635 } 6636 6637 exit_block = _gvn.transform(exit_block); 6638 memory_phi = _gvn.transform(memory_phi); 6639 result_phi = _gvn.transform(result_phi); 6640 6641 record_for_igvn(exit_block); 6642 record_for_igvn(memory_phi); 6643 record_for_igvn(result_phi); 6644 6645 set_control(exit_block); 6646 set_all_memory(memory_phi); 6647 set_result(result_phi); 6648 6649 return true; 6650 } 6651 6652 //------------------------------inline_vectorizedHashcode---------------------------- 6653 bool LibraryCallKit::inline_vectorizedHashCode() { 6654 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform"); 6655 6656 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters"); 6657 Node* array = argument(0); 6658 Node* offset = argument(1); 6659 Node* length = argument(2); 6660 Node* initialValue = argument(3); 6661 Node* basic_type = argument(4); 6662 6663 if (basic_type == top()) { 6664 return false; // failed input validation 6665 } 6666 6667 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int(); 6668 if (!basic_type_t->is_con()) { 6669 return false; // Only intrinsify if mode argument is constant 6670 } 6671 6672 array = must_be_not_null(array, true); 6673 6674 BasicType bt = (BasicType)basic_type_t->get_con(); 6675 6676 // Resolve address of first element 6677 Node* array_start = array_element_address(array, offset, bt); 6678 6679 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)), 6680 array_start, length, initialValue, basic_type))); 6681 clear_upper_avx(); 6682 6683 return true; 6684 } 6685 6686 /** 6687 * Calculate CRC32 for byte. 6688 * int java.util.zip.CRC32.update(int crc, int b) 6689 */ 6690 bool LibraryCallKit::inline_updateCRC32() { 6691 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support"); 6692 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 6693 // no receiver since it is static method 6694 Node* crc = argument(0); // type: int 6695 Node* b = argument(1); // type: int 6696 6697 /* 6698 * int c = ~ crc; 6699 * b = timesXtoThe32[(b ^ c) & 0xFF]; 6700 * b = b ^ (c >>> 8); 6701 * crc = ~b; 6702 */ 6703 6704 Node* M1 = intcon(-1); 6705 crc = _gvn.transform(new XorINode(crc, M1)); 6706 Node* result = _gvn.transform(new XorINode(crc, b)); 6707 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 6708 6709 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 6710 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 6711 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 6712 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 6713 6714 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 6715 result = _gvn.transform(new XorINode(crc, result)); 6716 result = _gvn.transform(new XorINode(result, M1)); 6717 set_result(result); 6718 return true; 6719 } 6720 6721 /** 6722 * Calculate CRC32 for byte[] array. 6723 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 6724 */ 6725 bool LibraryCallKit::inline_updateBytesCRC32() { 6726 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support"); 6727 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6728 // no receiver since it is static method 6729 Node* crc = argument(0); // type: int 6730 Node* src = argument(1); // type: oop 6731 Node* offset = argument(2); // type: int 6732 Node* length = argument(3); // type: int 6733 6734 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6735 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6736 // failed array check 6737 return false; 6738 } 6739 6740 // Figure out the size and type of the elements we will be copying. 6741 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6742 if (src_elem != T_BYTE) { 6743 return false; 6744 } 6745 6746 // 'src_start' points to src array + scaled offset 6747 src = must_be_not_null(src, true); 6748 Node* src_start = array_element_address(src, offset, src_elem); 6749 6750 // We assume that range check is done by caller. 6751 // TODO: generate range check (offset+length < src.length) in debug VM. 6752 6753 // Call the stub. 6754 address stubAddr = StubRoutines::updateBytesCRC32(); 6755 const char *stubName = "updateBytesCRC32"; 6756 6757 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 6758 stubAddr, stubName, TypePtr::BOTTOM, 6759 crc, src_start, length); 6760 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6761 set_result(result); 6762 return true; 6763 } 6764 6765 /** 6766 * Calculate CRC32 for ByteBuffer. 6767 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 6768 */ 6769 bool LibraryCallKit::inline_updateByteBufferCRC32() { 6770 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support"); 6771 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 6772 // no receiver since it is static method 6773 Node* crc = argument(0); // type: int 6774 Node* src = argument(1); // type: long 6775 Node* offset = argument(3); // type: int 6776 Node* length = argument(4); // type: int 6777 6778 src = ConvL2X(src); // adjust Java long to machine word 6779 Node* base = _gvn.transform(new CastX2PNode(src)); 6780 offset = ConvI2X(offset); 6781 6782 // 'src_start' points to src array + scaled offset 6783 Node* src_start = basic_plus_adr(top(), base, offset); 6784 6785 // Call the stub. 6786 address stubAddr = StubRoutines::updateBytesCRC32(); 6787 const char *stubName = "updateBytesCRC32"; 6788 6789 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 6790 stubAddr, stubName, TypePtr::BOTTOM, 6791 crc, src_start, length); 6792 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6793 set_result(result); 6794 return true; 6795 } 6796 6797 //------------------------------get_table_from_crc32c_class----------------------- 6798 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 6799 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class); 6800 assert (table != nullptr, "wrong version of java.util.zip.CRC32C"); 6801 6802 return table; 6803 } 6804 6805 //------------------------------inline_updateBytesCRC32C----------------------- 6806 // 6807 // Calculate CRC32C for byte[] array. 6808 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 6809 // 6810 bool LibraryCallKit::inline_updateBytesCRC32C() { 6811 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6812 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6813 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6814 // no receiver since it is a static method 6815 Node* crc = argument(0); // type: int 6816 Node* src = argument(1); // type: oop 6817 Node* offset = argument(2); // type: int 6818 Node* end = argument(3); // type: int 6819 6820 Node* length = _gvn.transform(new SubINode(end, offset)); 6821 6822 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6823 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6824 // failed array check 6825 return false; 6826 } 6827 6828 // Figure out the size and type of the elements we will be copying. 6829 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6830 if (src_elem != T_BYTE) { 6831 return false; 6832 } 6833 6834 // 'src_start' points to src array + scaled offset 6835 src = must_be_not_null(src, true); 6836 Node* src_start = array_element_address(src, offset, src_elem); 6837 6838 // static final int[] byteTable in class CRC32C 6839 Node* table = get_table_from_crc32c_class(callee()->holder()); 6840 table = must_be_not_null(table, true); 6841 Node* table_start = array_element_address(table, intcon(0), T_INT); 6842 6843 // We assume that range check is done by caller. 6844 // TODO: generate range check (offset+length < src.length) in debug VM. 6845 6846 // Call the stub. 6847 address stubAddr = StubRoutines::updateBytesCRC32C(); 6848 const char *stubName = "updateBytesCRC32C"; 6849 6850 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 6851 stubAddr, stubName, TypePtr::BOTTOM, 6852 crc, src_start, length, table_start); 6853 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6854 set_result(result); 6855 return true; 6856 } 6857 6858 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 6859 // 6860 // Calculate CRC32C for DirectByteBuffer. 6861 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 6862 // 6863 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 6864 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6865 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 6866 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6867 // no receiver since it is a static method 6868 Node* crc = argument(0); // type: int 6869 Node* src = argument(1); // type: long 6870 Node* offset = argument(3); // type: int 6871 Node* end = argument(4); // type: int 6872 6873 Node* length = _gvn.transform(new SubINode(end, offset)); 6874 6875 src = ConvL2X(src); // adjust Java long to machine word 6876 Node* base = _gvn.transform(new CastX2PNode(src)); 6877 offset = ConvI2X(offset); 6878 6879 // 'src_start' points to src array + scaled offset 6880 Node* src_start = basic_plus_adr(top(), base, offset); 6881 6882 // static final int[] byteTable in class CRC32C 6883 Node* table = get_table_from_crc32c_class(callee()->holder()); 6884 table = must_be_not_null(table, true); 6885 Node* table_start = array_element_address(table, intcon(0), T_INT); 6886 6887 // Call the stub. 6888 address stubAddr = StubRoutines::updateBytesCRC32C(); 6889 const char *stubName = "updateBytesCRC32C"; 6890 6891 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 6892 stubAddr, stubName, TypePtr::BOTTOM, 6893 crc, src_start, length, table_start); 6894 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6895 set_result(result); 6896 return true; 6897 } 6898 6899 //------------------------------inline_updateBytesAdler32---------------------- 6900 // 6901 // Calculate Adler32 checksum for byte[] array. 6902 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 6903 // 6904 bool LibraryCallKit::inline_updateBytesAdler32() { 6905 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one 6906 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6907 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 6908 // no receiver since it is static method 6909 Node* crc = argument(0); // type: int 6910 Node* src = argument(1); // type: oop 6911 Node* offset = argument(2); // type: int 6912 Node* length = argument(3); // type: int 6913 6914 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 6915 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 6916 // failed array check 6917 return false; 6918 } 6919 6920 // Figure out the size and type of the elements we will be copying. 6921 BasicType src_elem = src_type->elem()->array_element_basic_type(); 6922 if (src_elem != T_BYTE) { 6923 return false; 6924 } 6925 6926 // 'src_start' points to src array + scaled offset 6927 Node* src_start = array_element_address(src, offset, src_elem); 6928 6929 // We assume that range check is done by caller. 6930 // TODO: generate range check (offset+length < src.length) in debug VM. 6931 6932 // Call the stub. 6933 address stubAddr = StubRoutines::updateBytesAdler32(); 6934 const char *stubName = "updateBytesAdler32"; 6935 6936 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6937 stubAddr, stubName, TypePtr::BOTTOM, 6938 crc, src_start, length); 6939 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6940 set_result(result); 6941 return true; 6942 } 6943 6944 //------------------------------inline_updateByteBufferAdler32--------------- 6945 // 6946 // Calculate Adler32 checksum for DirectByteBuffer. 6947 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 6948 // 6949 bool LibraryCallKit::inline_updateByteBufferAdler32() { 6950 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one 6951 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 6952 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 6953 // no receiver since it is static method 6954 Node* crc = argument(0); // type: int 6955 Node* src = argument(1); // type: long 6956 Node* offset = argument(3); // type: int 6957 Node* length = argument(4); // type: int 6958 6959 src = ConvL2X(src); // adjust Java long to machine word 6960 Node* base = _gvn.transform(new CastX2PNode(src)); 6961 offset = ConvI2X(offset); 6962 6963 // 'src_start' points to src array + scaled offset 6964 Node* src_start = basic_plus_adr(top(), base, offset); 6965 6966 // Call the stub. 6967 address stubAddr = StubRoutines::updateBytesAdler32(); 6968 const char *stubName = "updateBytesAdler32"; 6969 6970 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6971 stubAddr, stubName, TypePtr::BOTTOM, 6972 crc, src_start, length); 6973 6974 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6975 set_result(result); 6976 return true; 6977 } 6978 6979 //----------------------------inline_reference_get0---------------------------- 6980 // public T java.lang.ref.Reference.get(); 6981 bool LibraryCallKit::inline_reference_get0() { 6982 const int referent_offset = java_lang_ref_Reference::referent_offset(); 6983 6984 // Get the argument: 6985 Node* reference_obj = null_check_receiver(); 6986 if (stopped()) return true; 6987 6988 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF; 6989 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;", 6990 decorators, /*is_static*/ false, nullptr); 6991 if (result == nullptr) return false; 6992 6993 // Add memory barrier to prevent commoning reads from this field 6994 // across safepoint since GC can change its value. 6995 insert_mem_bar(Op_MemBarCPUOrder); 6996 6997 set_result(result); 6998 return true; 6999 } 7000 7001 //----------------------------inline_reference_refersTo0---------------------------- 7002 // bool java.lang.ref.Reference.refersTo0(); 7003 // bool java.lang.ref.PhantomReference.refersTo0(); 7004 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) { 7005 // Get arguments: 7006 Node* reference_obj = null_check_receiver(); 7007 Node* other_obj = argument(1); 7008 if (stopped()) return true; 7009 7010 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE; 7011 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF); 7012 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;", 7013 decorators, /*is_static*/ false, nullptr); 7014 if (referent == nullptr) return false; 7015 7016 // Add memory barrier to prevent commoning reads from this field 7017 // across safepoint since GC can change its value. 7018 insert_mem_bar(Op_MemBarCPUOrder); 7019 7020 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj)); 7021 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 7022 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN); 7023 7024 RegionNode* region = new RegionNode(3); 7025 PhiNode* phi = new PhiNode(region, TypeInt::BOOL); 7026 7027 Node* if_true = _gvn.transform(new IfTrueNode(if_node)); 7028 region->init_req(1, if_true); 7029 phi->init_req(1, intcon(1)); 7030 7031 Node* if_false = _gvn.transform(new IfFalseNode(if_node)); 7032 region->init_req(2, if_false); 7033 phi->init_req(2, intcon(0)); 7034 7035 set_control(_gvn.transform(region)); 7036 record_for_igvn(region); 7037 set_result(_gvn.transform(phi)); 7038 return true; 7039 } 7040 7041 //----------------------------inline_reference_clear0---------------------------- 7042 // void java.lang.ref.Reference.clear0(); 7043 // void java.lang.ref.PhantomReference.clear0(); 7044 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) { 7045 // This matches the implementation in JVM_ReferenceClear, see the comments there. 7046 7047 // Get arguments 7048 Node* reference_obj = null_check_receiver(); 7049 if (stopped()) return true; 7050 7051 // Common access parameters 7052 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE; 7053 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF); 7054 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset()); 7055 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr(); 7056 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass()); 7057 7058 Node* referent = access_load_at(reference_obj, 7059 referent_field_addr, 7060 referent_field_addr_type, 7061 val_type, 7062 T_OBJECT, 7063 decorators); 7064 7065 IdealKit ideal(this); 7066 #define __ ideal. 7067 __ if_then(referent, BoolTest::ne, null()); 7068 sync_kit(ideal); 7069 access_store_at(reference_obj, 7070 referent_field_addr, 7071 referent_field_addr_type, 7072 null(), 7073 val_type, 7074 T_OBJECT, 7075 decorators); 7076 __ sync_kit(this); 7077 __ end_if(); 7078 final_sync(ideal); 7079 #undef __ 7080 7081 return true; 7082 } 7083 7084 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString, 7085 DecoratorSet decorators, bool is_static, 7086 ciInstanceKlass* fromKls) { 7087 if (fromKls == nullptr) { 7088 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 7089 assert(tinst != nullptr, "obj is null"); 7090 assert(tinst->is_loaded(), "obj is not loaded"); 7091 fromKls = tinst->instance_klass(); 7092 } else { 7093 assert(is_static, "only for static field access"); 7094 } 7095 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 7096 ciSymbol::make(fieldTypeString), 7097 is_static); 7098 7099 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName); 7100 if (field == nullptr) return (Node *) nullptr; 7101 7102 if (is_static) { 7103 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 7104 fromObj = makecon(tip); 7105 } 7106 7107 // Next code copied from Parse::do_get_xxx(): 7108 7109 // Compute address and memory type. 7110 int offset = field->offset_in_bytes(); 7111 bool is_vol = field->is_volatile(); 7112 ciType* field_klass = field->type(); 7113 assert(field_klass->is_loaded(), "should be loaded"); 7114 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 7115 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 7116 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()), 7117 "slice of address and input slice don't match"); 7118 BasicType bt = field->layout_type(); 7119 7120 // Build the resultant type of the load 7121 const Type *type; 7122 if (bt == T_OBJECT) { 7123 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 7124 } else { 7125 type = Type::get_const_basic_type(bt); 7126 } 7127 7128 if (is_vol) { 7129 decorators |= MO_SEQ_CST; 7130 } 7131 7132 return access_load_at(fromObj, adr, adr_type, type, bt, decorators); 7133 } 7134 7135 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 7136 bool is_exact /* true */, bool is_static /* false */, 7137 ciInstanceKlass * fromKls /* nullptr */) { 7138 if (fromKls == nullptr) { 7139 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 7140 assert(tinst != nullptr, "obj is null"); 7141 assert(tinst->is_loaded(), "obj is not loaded"); 7142 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 7143 fromKls = tinst->instance_klass(); 7144 } 7145 else { 7146 assert(is_static, "only for static field access"); 7147 } 7148 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 7149 ciSymbol::make(fieldTypeString), 7150 is_static); 7151 7152 assert(field != nullptr, "undefined field"); 7153 assert(!field->is_volatile(), "not defined for volatile fields"); 7154 7155 if (is_static) { 7156 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 7157 fromObj = makecon(tip); 7158 } 7159 7160 // Next code copied from Parse::do_get_xxx(): 7161 7162 // Compute address and memory type. 7163 int offset = field->offset_in_bytes(); 7164 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 7165 7166 return adr; 7167 } 7168 7169 //------------------------------inline_aescrypt_Block----------------------- 7170 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 7171 address stubAddr = nullptr; 7172 const char *stubName; 7173 assert(UseAES, "need AES instruction support"); 7174 7175 switch(id) { 7176 case vmIntrinsics::_aescrypt_encryptBlock: 7177 stubAddr = StubRoutines::aescrypt_encryptBlock(); 7178 stubName = "aescrypt_encryptBlock"; 7179 break; 7180 case vmIntrinsics::_aescrypt_decryptBlock: 7181 stubAddr = StubRoutines::aescrypt_decryptBlock(); 7182 stubName = "aescrypt_decryptBlock"; 7183 break; 7184 default: 7185 break; 7186 } 7187 if (stubAddr == nullptr) return false; 7188 7189 Node* aescrypt_object = argument(0); 7190 Node* src = argument(1); 7191 Node* src_offset = argument(2); 7192 Node* dest = argument(3); 7193 Node* dest_offset = argument(4); 7194 7195 src = must_be_not_null(src, true); 7196 dest = must_be_not_null(dest, true); 7197 7198 // (1) src and dest are arrays. 7199 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7200 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7201 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7202 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7203 7204 // for the quick and dirty code we will skip all the checks. 7205 // we are just trying to get the call to be generated. 7206 Node* src_start = src; 7207 Node* dest_start = dest; 7208 if (src_offset != nullptr || dest_offset != nullptr) { 7209 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7210 src_start = array_element_address(src, src_offset, T_BYTE); 7211 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7212 } 7213 7214 // now need to get the start of its expanded key array 7215 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7216 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7217 if (k_start == nullptr) return false; 7218 7219 // Call the stub. 7220 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 7221 stubAddr, stubName, TypePtr::BOTTOM, 7222 src_start, dest_start, k_start); 7223 7224 return true; 7225 } 7226 7227 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 7228 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 7229 address stubAddr = nullptr; 7230 const char *stubName = nullptr; 7231 7232 assert(UseAES, "need AES instruction support"); 7233 7234 switch(id) { 7235 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 7236 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 7237 stubName = "cipherBlockChaining_encryptAESCrypt"; 7238 break; 7239 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 7240 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 7241 stubName = "cipherBlockChaining_decryptAESCrypt"; 7242 break; 7243 default: 7244 break; 7245 } 7246 if (stubAddr == nullptr) return false; 7247 7248 Node* cipherBlockChaining_object = argument(0); 7249 Node* src = argument(1); 7250 Node* src_offset = argument(2); 7251 Node* len = argument(3); 7252 Node* dest = argument(4); 7253 Node* dest_offset = argument(5); 7254 7255 src = must_be_not_null(src, false); 7256 dest = must_be_not_null(dest, false); 7257 7258 // (1) src and dest are arrays. 7259 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7260 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7261 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7262 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7263 7264 // checks are the responsibility of the caller 7265 Node* src_start = src; 7266 Node* dest_start = dest; 7267 if (src_offset != nullptr || dest_offset != nullptr) { 7268 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7269 src_start = array_element_address(src, src_offset, T_BYTE); 7270 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7271 } 7272 7273 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7274 // (because of the predicated logic executed earlier). 7275 // so we cast it here safely. 7276 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7277 7278 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7279 if (embeddedCipherObj == nullptr) return false; 7280 7281 // cast it to what we know it will be at runtime 7282 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 7283 assert(tinst != nullptr, "CBC obj is null"); 7284 assert(tinst->is_loaded(), "CBC obj is not loaded"); 7285 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7286 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7287 7288 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7289 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7290 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7291 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7292 aescrypt_object = _gvn.transform(aescrypt_object); 7293 7294 // we need to get the start of the aescrypt_object's expanded key array 7295 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7296 if (k_start == nullptr) return false; 7297 7298 // similarly, get the start address of the r vector 7299 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B"); 7300 if (objRvec == nullptr) return false; 7301 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 7302 7303 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7304 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 7305 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 7306 stubAddr, stubName, TypePtr::BOTTOM, 7307 src_start, dest_start, k_start, r_start, len); 7308 7309 // return cipher length (int) 7310 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 7311 set_result(retvalue); 7312 return true; 7313 } 7314 7315 //------------------------------inline_electronicCodeBook_AESCrypt----------------------- 7316 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) { 7317 address stubAddr = nullptr; 7318 const char *stubName = nullptr; 7319 7320 assert(UseAES, "need AES instruction support"); 7321 7322 switch (id) { 7323 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 7324 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt(); 7325 stubName = "electronicCodeBook_encryptAESCrypt"; 7326 break; 7327 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 7328 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt(); 7329 stubName = "electronicCodeBook_decryptAESCrypt"; 7330 break; 7331 default: 7332 break; 7333 } 7334 7335 if (stubAddr == nullptr) return false; 7336 7337 Node* electronicCodeBook_object = argument(0); 7338 Node* src = argument(1); 7339 Node* src_offset = argument(2); 7340 Node* len = argument(3); 7341 Node* dest = argument(4); 7342 Node* dest_offset = argument(5); 7343 7344 // (1) src and dest are arrays. 7345 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7346 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7347 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7348 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7349 7350 // checks are the responsibility of the caller 7351 Node* src_start = src; 7352 Node* dest_start = dest; 7353 if (src_offset != nullptr || dest_offset != nullptr) { 7354 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7355 src_start = array_element_address(src, src_offset, T_BYTE); 7356 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7357 } 7358 7359 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7360 // (because of the predicated logic executed earlier). 7361 // so we cast it here safely. 7362 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7363 7364 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7365 if (embeddedCipherObj == nullptr) return false; 7366 7367 // cast it to what we know it will be at runtime 7368 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr(); 7369 assert(tinst != nullptr, "ECB obj is null"); 7370 assert(tinst->is_loaded(), "ECB obj is not loaded"); 7371 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7372 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7373 7374 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7375 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7376 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7377 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7378 aescrypt_object = _gvn.transform(aescrypt_object); 7379 7380 // we need to get the start of the aescrypt_object's expanded key array 7381 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7382 if (k_start == nullptr) return false; 7383 7384 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7385 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP, 7386 OptoRuntime::electronicCodeBook_aescrypt_Type(), 7387 stubAddr, stubName, TypePtr::BOTTOM, 7388 src_start, dest_start, k_start, len); 7389 7390 // return cipher length (int) 7391 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms)); 7392 set_result(retvalue); 7393 return true; 7394 } 7395 7396 //------------------------------inline_counterMode_AESCrypt----------------------- 7397 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 7398 assert(UseAES, "need AES instruction support"); 7399 if (!UseAESCTRIntrinsics) return false; 7400 7401 address stubAddr = nullptr; 7402 const char *stubName = nullptr; 7403 if (id == vmIntrinsics::_counterMode_AESCrypt) { 7404 stubAddr = StubRoutines::counterMode_AESCrypt(); 7405 stubName = "counterMode_AESCrypt"; 7406 } 7407 if (stubAddr == nullptr) return false; 7408 7409 Node* counterMode_object = argument(0); 7410 Node* src = argument(1); 7411 Node* src_offset = argument(2); 7412 Node* len = argument(3); 7413 Node* dest = argument(4); 7414 Node* dest_offset = argument(5); 7415 7416 // (1) src and dest are arrays. 7417 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 7418 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr(); 7419 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM && 7420 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange"); 7421 7422 // checks are the responsibility of the caller 7423 Node* src_start = src; 7424 Node* dest_start = dest; 7425 if (src_offset != nullptr || dest_offset != nullptr) { 7426 assert(src_offset != nullptr && dest_offset != nullptr, ""); 7427 src_start = array_element_address(src, src_offset, T_BYTE); 7428 dest_start = array_element_address(dest, dest_offset, T_BYTE); 7429 } 7430 7431 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 7432 // (because of the predicated logic executed earlier). 7433 // so we cast it here safely. 7434 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 7435 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7436 if (embeddedCipherObj == nullptr) return false; 7437 // cast it to what we know it will be at runtime 7438 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 7439 assert(tinst != nullptr, "CTR obj is null"); 7440 assert(tinst->is_loaded(), "CTR obj is not loaded"); 7441 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7442 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 7443 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7444 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 7445 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 7446 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 7447 aescrypt_object = _gvn.transform(aescrypt_object); 7448 // we need to get the start of the aescrypt_object's expanded key array 7449 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 7450 if (k_start == nullptr) return false; 7451 // similarly, get the start address of the r vector 7452 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B"); 7453 if (obj_counter == nullptr) return false; 7454 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 7455 7456 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B"); 7457 if (saved_encCounter == nullptr) return false; 7458 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 7459 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 7460 7461 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 7462 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 7463 OptoRuntime::counterMode_aescrypt_Type(), 7464 stubAddr, stubName, TypePtr::BOTTOM, 7465 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 7466 7467 // return cipher length (int) 7468 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 7469 set_result(retvalue); 7470 return true; 7471 } 7472 7473 //------------------------------get_key_start_from_aescrypt_object----------------------- 7474 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 7475 #if defined(PPC64) || defined(S390) || defined(RISCV64) 7476 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys. 7477 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns. 7478 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption. 7479 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]). 7480 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I"); 7481 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt"); 7482 if (objSessionK == nullptr) { 7483 return (Node *) nullptr; 7484 } 7485 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true); 7486 #else 7487 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I"); 7488 #endif // PPC64 7489 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt"); 7490 if (objAESCryptKey == nullptr) return (Node *) nullptr; 7491 7492 // now have the array, need to get the start address of the K array 7493 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 7494 return k_start; 7495 } 7496 7497 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 7498 // Return node representing slow path of predicate check. 7499 // the pseudo code we want to emulate with this predicate is: 7500 // for encryption: 7501 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7502 // for decryption: 7503 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7504 // note cipher==plain is more conservative than the original java code but that's OK 7505 // 7506 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 7507 // The receiver was checked for null already. 7508 Node* objCBC = argument(0); 7509 7510 Node* src = argument(1); 7511 Node* dest = argument(4); 7512 7513 // Load embeddedCipher field of CipherBlockChaining object. 7514 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7515 7516 // get AESCrypt klass for instanceOf check 7517 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7518 // will have same classloader as CipherBlockChaining object 7519 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 7520 assert(tinst != nullptr, "CBCobj is null"); 7521 assert(tinst->is_loaded(), "CBCobj is not loaded"); 7522 7523 // we want to do an instanceof comparison against the AESCrypt class 7524 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7525 if (!klass_AESCrypt->is_loaded()) { 7526 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7527 Node* ctrl = control(); 7528 set_control(top()); // no regular fast path 7529 return ctrl; 7530 } 7531 7532 src = must_be_not_null(src, true); 7533 dest = must_be_not_null(dest, true); 7534 7535 // Resolve oops to stable for CmpP below. 7536 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7537 7538 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7539 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7540 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7541 7542 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7543 7544 // for encryption, we are done 7545 if (!decrypting) 7546 return instof_false; // even if it is null 7547 7548 // for decryption, we need to add a further check to avoid 7549 // taking the intrinsic path when cipher and plain are the same 7550 // see the original java code for why. 7551 RegionNode* region = new RegionNode(3); 7552 region->init_req(1, instof_false); 7553 7554 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 7555 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 7556 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN); 7557 region->init_req(2, src_dest_conjoint); 7558 7559 record_for_igvn(region); 7560 return _gvn.transform(region); 7561 } 7562 7563 //----------------------------inline_electronicCodeBook_AESCrypt_predicate---------------------------- 7564 // Return node representing slow path of predicate check. 7565 // the pseudo code we want to emulate with this predicate is: 7566 // for encryption: 7567 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7568 // for decryption: 7569 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7570 // note cipher==plain is more conservative than the original java code but that's OK 7571 // 7572 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) { 7573 // The receiver was checked for null already. 7574 Node* objECB = argument(0); 7575 7576 // Load embeddedCipher field of ElectronicCodeBook object. 7577 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7578 7579 // get AESCrypt klass for instanceOf check 7580 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7581 // will have same classloader as ElectronicCodeBook object 7582 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr(); 7583 assert(tinst != nullptr, "ECBobj is null"); 7584 assert(tinst->is_loaded(), "ECBobj is not loaded"); 7585 7586 // we want to do an instanceof comparison against the AESCrypt class 7587 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7588 if (!klass_AESCrypt->is_loaded()) { 7589 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7590 Node* ctrl = control(); 7591 set_control(top()); // no regular fast path 7592 return ctrl; 7593 } 7594 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7595 7596 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7597 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7598 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7599 7600 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7601 7602 // for encryption, we are done 7603 if (!decrypting) 7604 return instof_false; // even if it is null 7605 7606 // for decryption, we need to add a further check to avoid 7607 // taking the intrinsic path when cipher and plain are the same 7608 // see the original java code for why. 7609 RegionNode* region = new RegionNode(3); 7610 region->init_req(1, instof_false); 7611 Node* src = argument(1); 7612 Node* dest = argument(4); 7613 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 7614 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 7615 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN); 7616 region->init_req(2, src_dest_conjoint); 7617 7618 record_for_igvn(region); 7619 return _gvn.transform(region); 7620 } 7621 7622 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 7623 // Return node representing slow path of predicate check. 7624 // the pseudo code we want to emulate with this predicate is: 7625 // for encryption: 7626 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 7627 // for decryption: 7628 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 7629 // note cipher==plain is more conservative than the original java code but that's OK 7630 // 7631 7632 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 7633 // The receiver was checked for null already. 7634 Node* objCTR = argument(0); 7635 7636 // Load embeddedCipher field of CipherBlockChaining object. 7637 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 7638 7639 // get AESCrypt klass for instanceOf check 7640 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 7641 // will have same classloader as CipherBlockChaining object 7642 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 7643 assert(tinst != nullptr, "CTRobj is null"); 7644 assert(tinst->is_loaded(), "CTRobj is not loaded"); 7645 7646 // we want to do an instanceof comparison against the AESCrypt class 7647 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 7648 if (!klass_AESCrypt->is_loaded()) { 7649 // if AESCrypt is not even loaded, we never take the intrinsic fast path 7650 Node* ctrl = control(); 7651 set_control(top()); // no regular fast path 7652 return ctrl; 7653 } 7654 7655 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 7656 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 7657 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 7658 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7659 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 7660 7661 return instof_false; // even if it is null 7662 } 7663 7664 //------------------------------inline_ghash_processBlocks 7665 bool LibraryCallKit::inline_ghash_processBlocks() { 7666 address stubAddr; 7667 const char *stubName; 7668 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 7669 7670 stubAddr = StubRoutines::ghash_processBlocks(); 7671 stubName = "ghash_processBlocks"; 7672 7673 Node* data = argument(0); 7674 Node* offset = argument(1); 7675 Node* len = argument(2); 7676 Node* state = argument(3); 7677 Node* subkeyH = argument(4); 7678 7679 state = must_be_not_null(state, true); 7680 subkeyH = must_be_not_null(subkeyH, true); 7681 data = must_be_not_null(data, true); 7682 7683 Node* state_start = array_element_address(state, intcon(0), T_LONG); 7684 assert(state_start, "state is null"); 7685 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 7686 assert(subkeyH_start, "subkeyH is null"); 7687 Node* data_start = array_element_address(data, offset, T_BYTE); 7688 assert(data_start, "data is null"); 7689 7690 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 7691 OptoRuntime::ghash_processBlocks_Type(), 7692 stubAddr, stubName, TypePtr::BOTTOM, 7693 state_start, subkeyH_start, data_start, len); 7694 return true; 7695 } 7696 7697 //------------------------------inline_chacha20Block 7698 bool LibraryCallKit::inline_chacha20Block() { 7699 address stubAddr; 7700 const char *stubName; 7701 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support"); 7702 7703 stubAddr = StubRoutines::chacha20Block(); 7704 stubName = "chacha20Block"; 7705 7706 Node* state = argument(0); 7707 Node* result = argument(1); 7708 7709 state = must_be_not_null(state, true); 7710 result = must_be_not_null(result, true); 7711 7712 Node* state_start = array_element_address(state, intcon(0), T_INT); 7713 assert(state_start, "state is null"); 7714 Node* result_start = array_element_address(result, intcon(0), T_BYTE); 7715 assert(result_start, "result is null"); 7716 7717 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP, 7718 OptoRuntime::chacha20Block_Type(), 7719 stubAddr, stubName, TypePtr::BOTTOM, 7720 state_start, result_start); 7721 // return key stream length (int) 7722 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms)); 7723 set_result(retvalue); 7724 return true; 7725 } 7726 7727 //------------------------------inline_kyberNtt 7728 bool LibraryCallKit::inline_kyberNtt() { 7729 address stubAddr; 7730 const char *stubName; 7731 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7732 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters"); 7733 7734 stubAddr = StubRoutines::kyberNtt(); 7735 stubName = "kyberNtt"; 7736 if (!stubAddr) return false; 7737 7738 Node* coeffs = argument(0); 7739 Node* ntt_zetas = argument(1); 7740 7741 coeffs = must_be_not_null(coeffs, true); 7742 ntt_zetas = must_be_not_null(ntt_zetas, true); 7743 7744 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT); 7745 assert(coeffs_start, "coeffs is null"); 7746 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT); 7747 assert(ntt_zetas_start, "ntt_zetas is null"); 7748 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 7749 OptoRuntime::kyberNtt_Type(), 7750 stubAddr, stubName, TypePtr::BOTTOM, 7751 coeffs_start, ntt_zetas_start); 7752 // return an int 7753 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms)); 7754 set_result(retvalue); 7755 return true; 7756 } 7757 7758 //------------------------------inline_kyberInverseNtt 7759 bool LibraryCallKit::inline_kyberInverseNtt() { 7760 address stubAddr; 7761 const char *stubName; 7762 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7763 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters"); 7764 7765 stubAddr = StubRoutines::kyberInverseNtt(); 7766 stubName = "kyberInverseNtt"; 7767 if (!stubAddr) return false; 7768 7769 Node* coeffs = argument(0); 7770 Node* zetas = argument(1); 7771 7772 coeffs = must_be_not_null(coeffs, true); 7773 zetas = must_be_not_null(zetas, true); 7774 7775 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT); 7776 assert(coeffs_start, "coeffs is null"); 7777 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT); 7778 assert(zetas_start, "inverseNtt_zetas is null"); 7779 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 7780 OptoRuntime::kyberInverseNtt_Type(), 7781 stubAddr, stubName, TypePtr::BOTTOM, 7782 coeffs_start, zetas_start); 7783 7784 // return an int 7785 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms)); 7786 set_result(retvalue); 7787 return true; 7788 } 7789 7790 //------------------------------inline_kyberNttMult 7791 bool LibraryCallKit::inline_kyberNttMult() { 7792 address stubAddr; 7793 const char *stubName; 7794 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7795 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters"); 7796 7797 stubAddr = StubRoutines::kyberNttMult(); 7798 stubName = "kyberNttMult"; 7799 if (!stubAddr) return false; 7800 7801 Node* result = argument(0); 7802 Node* ntta = argument(1); 7803 Node* nttb = argument(2); 7804 Node* zetas = argument(3); 7805 7806 result = must_be_not_null(result, true); 7807 ntta = must_be_not_null(ntta, true); 7808 nttb = must_be_not_null(nttb, true); 7809 zetas = must_be_not_null(zetas, true); 7810 7811 Node* result_start = array_element_address(result, intcon(0), T_SHORT); 7812 assert(result_start, "result is null"); 7813 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT); 7814 assert(ntta_start, "ntta is null"); 7815 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT); 7816 assert(nttb_start, "nttb is null"); 7817 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT); 7818 assert(zetas_start, "nttMult_zetas is null"); 7819 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP, 7820 OptoRuntime::kyberNttMult_Type(), 7821 stubAddr, stubName, TypePtr::BOTTOM, 7822 result_start, ntta_start, nttb_start, 7823 zetas_start); 7824 7825 // return an int 7826 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms)); 7827 set_result(retvalue); 7828 7829 return true; 7830 } 7831 7832 //------------------------------inline_kyberAddPoly_2 7833 bool LibraryCallKit::inline_kyberAddPoly_2() { 7834 address stubAddr; 7835 const char *stubName; 7836 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7837 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters"); 7838 7839 stubAddr = StubRoutines::kyberAddPoly_2(); 7840 stubName = "kyberAddPoly_2"; 7841 if (!stubAddr) return false; 7842 7843 Node* result = argument(0); 7844 Node* a = argument(1); 7845 Node* b = argument(2); 7846 7847 result = must_be_not_null(result, true); 7848 a = must_be_not_null(a, true); 7849 b = must_be_not_null(b, true); 7850 7851 Node* result_start = array_element_address(result, intcon(0), T_SHORT); 7852 assert(result_start, "result is null"); 7853 Node* a_start = array_element_address(a, intcon(0), T_SHORT); 7854 assert(a_start, "a is null"); 7855 Node* b_start = array_element_address(b, intcon(0), T_SHORT); 7856 assert(b_start, "b is null"); 7857 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP, 7858 OptoRuntime::kyberAddPoly_2_Type(), 7859 stubAddr, stubName, TypePtr::BOTTOM, 7860 result_start, a_start, b_start); 7861 // return an int 7862 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms)); 7863 set_result(retvalue); 7864 return true; 7865 } 7866 7867 //------------------------------inline_kyberAddPoly_3 7868 bool LibraryCallKit::inline_kyberAddPoly_3() { 7869 address stubAddr; 7870 const char *stubName; 7871 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7872 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters"); 7873 7874 stubAddr = StubRoutines::kyberAddPoly_3(); 7875 stubName = "kyberAddPoly_3"; 7876 if (!stubAddr) return false; 7877 7878 Node* result = argument(0); 7879 Node* a = argument(1); 7880 Node* b = argument(2); 7881 Node* c = argument(3); 7882 7883 result = must_be_not_null(result, true); 7884 a = must_be_not_null(a, true); 7885 b = must_be_not_null(b, true); 7886 c = must_be_not_null(c, true); 7887 7888 Node* result_start = array_element_address(result, intcon(0), T_SHORT); 7889 assert(result_start, "result is null"); 7890 Node* a_start = array_element_address(a, intcon(0), T_SHORT); 7891 assert(a_start, "a is null"); 7892 Node* b_start = array_element_address(b, intcon(0), T_SHORT); 7893 assert(b_start, "b is null"); 7894 Node* c_start = array_element_address(c, intcon(0), T_SHORT); 7895 assert(c_start, "c is null"); 7896 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP, 7897 OptoRuntime::kyberAddPoly_3_Type(), 7898 stubAddr, stubName, TypePtr::BOTTOM, 7899 result_start, a_start, b_start, c_start); 7900 // return an int 7901 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms)); 7902 set_result(retvalue); 7903 return true; 7904 } 7905 7906 //------------------------------inline_kyber12To16 7907 bool LibraryCallKit::inline_kyber12To16() { 7908 address stubAddr; 7909 const char *stubName; 7910 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7911 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters"); 7912 7913 stubAddr = StubRoutines::kyber12To16(); 7914 stubName = "kyber12To16"; 7915 if (!stubAddr) return false; 7916 7917 Node* condensed = argument(0); 7918 Node* condensedOffs = argument(1); 7919 Node* parsed = argument(2); 7920 Node* parsedLength = argument(3); 7921 7922 condensed = must_be_not_null(condensed, true); 7923 parsed = must_be_not_null(parsed, true); 7924 7925 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE); 7926 assert(condensed_start, "condensed is null"); 7927 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT); 7928 assert(parsed_start, "parsed is null"); 7929 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP, 7930 OptoRuntime::kyber12To16_Type(), 7931 stubAddr, stubName, TypePtr::BOTTOM, 7932 condensed_start, condensedOffs, parsed_start, parsedLength); 7933 // return an int 7934 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms)); 7935 set_result(retvalue); 7936 return true; 7937 7938 } 7939 7940 //------------------------------inline_kyberBarrettReduce 7941 bool LibraryCallKit::inline_kyberBarrettReduce() { 7942 address stubAddr; 7943 const char *stubName; 7944 assert(UseKyberIntrinsics, "need Kyber intrinsics support"); 7945 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters"); 7946 7947 stubAddr = StubRoutines::kyberBarrettReduce(); 7948 stubName = "kyberBarrettReduce"; 7949 if (!stubAddr) return false; 7950 7951 Node* coeffs = argument(0); 7952 7953 coeffs = must_be_not_null(coeffs, true); 7954 7955 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT); 7956 assert(coeffs_start, "coeffs is null"); 7957 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP, 7958 OptoRuntime::kyberBarrettReduce_Type(), 7959 stubAddr, stubName, TypePtr::BOTTOM, 7960 coeffs_start); 7961 // return an int 7962 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms)); 7963 set_result(retvalue); 7964 return true; 7965 } 7966 7967 //------------------------------inline_dilithiumAlmostNtt 7968 bool LibraryCallKit::inline_dilithiumAlmostNtt() { 7969 address stubAddr; 7970 const char *stubName; 7971 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 7972 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters"); 7973 7974 stubAddr = StubRoutines::dilithiumAlmostNtt(); 7975 stubName = "dilithiumAlmostNtt"; 7976 if (!stubAddr) return false; 7977 7978 Node* coeffs = argument(0); 7979 Node* ntt_zetas = argument(1); 7980 7981 coeffs = must_be_not_null(coeffs, true); 7982 ntt_zetas = must_be_not_null(ntt_zetas, true); 7983 7984 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT); 7985 assert(coeffs_start, "coeffs is null"); 7986 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT); 7987 assert(ntt_zetas_start, "ntt_zetas is null"); 7988 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 7989 OptoRuntime::dilithiumAlmostNtt_Type(), 7990 stubAddr, stubName, TypePtr::BOTTOM, 7991 coeffs_start, ntt_zetas_start); 7992 // return an int 7993 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms)); 7994 set_result(retvalue); 7995 return true; 7996 } 7997 7998 //------------------------------inline_dilithiumAlmostInverseNtt 7999 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() { 8000 address stubAddr; 8001 const char *stubName; 8002 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 8003 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters"); 8004 8005 stubAddr = StubRoutines::dilithiumAlmostInverseNtt(); 8006 stubName = "dilithiumAlmostInverseNtt"; 8007 if (!stubAddr) return false; 8008 8009 Node* coeffs = argument(0); 8010 Node* zetas = argument(1); 8011 8012 coeffs = must_be_not_null(coeffs, true); 8013 zetas = must_be_not_null(zetas, true); 8014 8015 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT); 8016 assert(coeffs_start, "coeffs is null"); 8017 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT); 8018 assert(zetas_start, "inverseNtt_zetas is null"); 8019 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP, 8020 OptoRuntime::dilithiumAlmostInverseNtt_Type(), 8021 stubAddr, stubName, TypePtr::BOTTOM, 8022 coeffs_start, zetas_start); 8023 // return an int 8024 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms)); 8025 set_result(retvalue); 8026 return true; 8027 } 8028 8029 //------------------------------inline_dilithiumNttMult 8030 bool LibraryCallKit::inline_dilithiumNttMult() { 8031 address stubAddr; 8032 const char *stubName; 8033 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 8034 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters"); 8035 8036 stubAddr = StubRoutines::dilithiumNttMult(); 8037 stubName = "dilithiumNttMult"; 8038 if (!stubAddr) return false; 8039 8040 Node* result = argument(0); 8041 Node* ntta = argument(1); 8042 Node* nttb = argument(2); 8043 Node* zetas = argument(3); 8044 8045 result = must_be_not_null(result, true); 8046 ntta = must_be_not_null(ntta, true); 8047 nttb = must_be_not_null(nttb, true); 8048 zetas = must_be_not_null(zetas, true); 8049 8050 Node* result_start = array_element_address(result, intcon(0), T_INT); 8051 assert(result_start, "result is null"); 8052 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT); 8053 assert(ntta_start, "ntta is null"); 8054 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT); 8055 assert(nttb_start, "nttb is null"); 8056 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP, 8057 OptoRuntime::dilithiumNttMult_Type(), 8058 stubAddr, stubName, TypePtr::BOTTOM, 8059 result_start, ntta_start, nttb_start); 8060 8061 // return an int 8062 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms)); 8063 set_result(retvalue); 8064 8065 return true; 8066 } 8067 8068 //------------------------------inline_dilithiumMontMulByConstant 8069 bool LibraryCallKit::inline_dilithiumMontMulByConstant() { 8070 address stubAddr; 8071 const char *stubName; 8072 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 8073 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters"); 8074 8075 stubAddr = StubRoutines::dilithiumMontMulByConstant(); 8076 stubName = "dilithiumMontMulByConstant"; 8077 if (!stubAddr) return false; 8078 8079 Node* coeffs = argument(0); 8080 Node* constant = argument(1); 8081 8082 coeffs = must_be_not_null(coeffs, true); 8083 8084 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT); 8085 assert(coeffs_start, "coeffs is null"); 8086 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP, 8087 OptoRuntime::dilithiumMontMulByConstant_Type(), 8088 stubAddr, stubName, TypePtr::BOTTOM, 8089 coeffs_start, constant); 8090 8091 // return an int 8092 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms)); 8093 set_result(retvalue); 8094 return true; 8095 } 8096 8097 8098 //------------------------------inline_dilithiumDecomposePoly 8099 bool LibraryCallKit::inline_dilithiumDecomposePoly() { 8100 address stubAddr; 8101 const char *stubName; 8102 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support"); 8103 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters"); 8104 8105 stubAddr = StubRoutines::dilithiumDecomposePoly(); 8106 stubName = "dilithiumDecomposePoly"; 8107 if (!stubAddr) return false; 8108 8109 Node* input = argument(0); 8110 Node* lowPart = argument(1); 8111 Node* highPart = argument(2); 8112 Node* twoGamma2 = argument(3); 8113 Node* multiplier = argument(4); 8114 8115 input = must_be_not_null(input, true); 8116 lowPart = must_be_not_null(lowPart, true); 8117 highPart = must_be_not_null(highPart, true); 8118 8119 Node* input_start = array_element_address(input, intcon(0), T_INT); 8120 assert(input_start, "input is null"); 8121 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT); 8122 assert(lowPart_start, "lowPart is null"); 8123 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT); 8124 assert(highPart_start, "highPart is null"); 8125 8126 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP, 8127 OptoRuntime::dilithiumDecomposePoly_Type(), 8128 stubAddr, stubName, TypePtr::BOTTOM, 8129 input_start, lowPart_start, highPart_start, 8130 twoGamma2, multiplier); 8131 8132 // return an int 8133 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms)); 8134 set_result(retvalue); 8135 return true; 8136 } 8137 8138 bool LibraryCallKit::inline_base64_encodeBlock() { 8139 address stubAddr; 8140 const char *stubName; 8141 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 8142 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters"); 8143 stubAddr = StubRoutines::base64_encodeBlock(); 8144 stubName = "encodeBlock"; 8145 8146 if (!stubAddr) return false; 8147 Node* base64obj = argument(0); 8148 Node* src = argument(1); 8149 Node* offset = argument(2); 8150 Node* len = argument(3); 8151 Node* dest = argument(4); 8152 Node* dp = argument(5); 8153 Node* isURL = argument(6); 8154 8155 src = must_be_not_null(src, true); 8156 dest = must_be_not_null(dest, true); 8157 8158 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 8159 assert(src_start, "source array is null"); 8160 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 8161 assert(dest_start, "destination array is null"); 8162 8163 Node* base64 = make_runtime_call(RC_LEAF, 8164 OptoRuntime::base64_encodeBlock_Type(), 8165 stubAddr, stubName, TypePtr::BOTTOM, 8166 src_start, offset, len, dest_start, dp, isURL); 8167 return true; 8168 } 8169 8170 bool LibraryCallKit::inline_base64_decodeBlock() { 8171 address stubAddr; 8172 const char *stubName; 8173 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 8174 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters"); 8175 stubAddr = StubRoutines::base64_decodeBlock(); 8176 stubName = "decodeBlock"; 8177 8178 if (!stubAddr) return false; 8179 Node* base64obj = argument(0); 8180 Node* src = argument(1); 8181 Node* src_offset = argument(2); 8182 Node* len = argument(3); 8183 Node* dest = argument(4); 8184 Node* dest_offset = argument(5); 8185 Node* isURL = argument(6); 8186 Node* isMIME = argument(7); 8187 8188 src = must_be_not_null(src, true); 8189 dest = must_be_not_null(dest, true); 8190 8191 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 8192 assert(src_start, "source array is null"); 8193 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 8194 assert(dest_start, "destination array is null"); 8195 8196 Node* call = make_runtime_call(RC_LEAF, 8197 OptoRuntime::base64_decodeBlock_Type(), 8198 stubAddr, stubName, TypePtr::BOTTOM, 8199 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME); 8200 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 8201 set_result(result); 8202 return true; 8203 } 8204 8205 bool LibraryCallKit::inline_poly1305_processBlocks() { 8206 address stubAddr; 8207 const char *stubName; 8208 assert(UsePoly1305Intrinsics, "need Poly intrinsics support"); 8209 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size()); 8210 stubAddr = StubRoutines::poly1305_processBlocks(); 8211 stubName = "poly1305_processBlocks"; 8212 8213 if (!stubAddr) return false; 8214 null_check_receiver(); // null-check receiver 8215 if (stopped()) return true; 8216 8217 Node* input = argument(1); 8218 Node* input_offset = argument(2); 8219 Node* len = argument(3); 8220 Node* alimbs = argument(4); 8221 Node* rlimbs = argument(5); 8222 8223 input = must_be_not_null(input, true); 8224 alimbs = must_be_not_null(alimbs, true); 8225 rlimbs = must_be_not_null(rlimbs, true); 8226 8227 Node* input_start = array_element_address(input, input_offset, T_BYTE); 8228 assert(input_start, "input array is null"); 8229 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG); 8230 assert(acc_start, "acc array is null"); 8231 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG); 8232 assert(r_start, "r array is null"); 8233 8234 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP, 8235 OptoRuntime::poly1305_processBlocks_Type(), 8236 stubAddr, stubName, TypePtr::BOTTOM, 8237 input_start, len, acc_start, r_start); 8238 return true; 8239 } 8240 8241 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() { 8242 address stubAddr; 8243 const char *stubName; 8244 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support"); 8245 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size()); 8246 stubAddr = StubRoutines::intpoly_montgomeryMult_P256(); 8247 stubName = "intpoly_montgomeryMult_P256"; 8248 8249 if (!stubAddr) return false; 8250 null_check_receiver(); // null-check receiver 8251 if (stopped()) return true; 8252 8253 Node* a = argument(1); 8254 Node* b = argument(2); 8255 Node* r = argument(3); 8256 8257 a = must_be_not_null(a, true); 8258 b = must_be_not_null(b, true); 8259 r = must_be_not_null(r, true); 8260 8261 Node* a_start = array_element_address(a, intcon(0), T_LONG); 8262 assert(a_start, "a array is null"); 8263 Node* b_start = array_element_address(b, intcon(0), T_LONG); 8264 assert(b_start, "b array is null"); 8265 Node* r_start = array_element_address(r, intcon(0), T_LONG); 8266 assert(r_start, "r array is null"); 8267 8268 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP, 8269 OptoRuntime::intpoly_montgomeryMult_P256_Type(), 8270 stubAddr, stubName, TypePtr::BOTTOM, 8271 a_start, b_start, r_start); 8272 return true; 8273 } 8274 8275 bool LibraryCallKit::inline_intpoly_assign() { 8276 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support"); 8277 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size()); 8278 const char *stubName = "intpoly_assign"; 8279 address stubAddr = StubRoutines::intpoly_assign(); 8280 if (!stubAddr) return false; 8281 8282 Node* set = argument(0); 8283 Node* a = argument(1); 8284 Node* b = argument(2); 8285 Node* arr_length = load_array_length(a); 8286 8287 a = must_be_not_null(a, true); 8288 b = must_be_not_null(b, true); 8289 8290 Node* a_start = array_element_address(a, intcon(0), T_LONG); 8291 assert(a_start, "a array is null"); 8292 Node* b_start = array_element_address(b, intcon(0), T_LONG); 8293 assert(b_start, "b array is null"); 8294 8295 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP, 8296 OptoRuntime::intpoly_assign_Type(), 8297 stubAddr, stubName, TypePtr::BOTTOM, 8298 set, a_start, b_start, arr_length); 8299 return true; 8300 } 8301 8302 //------------------------------inline_digestBase_implCompress----------------------- 8303 // 8304 // Calculate MD5 for single-block byte[] array. 8305 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs) 8306 // 8307 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 8308 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 8309 // 8310 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 8311 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 8312 // 8313 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 8314 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 8315 // 8316 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array. 8317 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs) 8318 // 8319 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) { 8320 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 8321 8322 Node* digestBase_obj = argument(0); 8323 Node* src = argument(1); // type oop 8324 Node* ofs = argument(2); // type int 8325 8326 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 8327 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 8328 // failed array check 8329 return false; 8330 } 8331 // Figure out the size and type of the elements we will be copying. 8332 BasicType src_elem = src_type->elem()->array_element_basic_type(); 8333 if (src_elem != T_BYTE) { 8334 return false; 8335 } 8336 // 'src_start' points to src array + offset 8337 src = must_be_not_null(src, true); 8338 Node* src_start = array_element_address(src, ofs, src_elem); 8339 Node* state = nullptr; 8340 Node* block_size = nullptr; 8341 address stubAddr; 8342 const char *stubName; 8343 8344 switch(id) { 8345 case vmIntrinsics::_md5_implCompress: 8346 assert(UseMD5Intrinsics, "need MD5 instruction support"); 8347 state = get_state_from_digest_object(digestBase_obj, T_INT); 8348 stubAddr = StubRoutines::md5_implCompress(); 8349 stubName = "md5_implCompress"; 8350 break; 8351 case vmIntrinsics::_sha_implCompress: 8352 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 8353 state = get_state_from_digest_object(digestBase_obj, T_INT); 8354 stubAddr = StubRoutines::sha1_implCompress(); 8355 stubName = "sha1_implCompress"; 8356 break; 8357 case vmIntrinsics::_sha2_implCompress: 8358 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 8359 state = get_state_from_digest_object(digestBase_obj, T_INT); 8360 stubAddr = StubRoutines::sha256_implCompress(); 8361 stubName = "sha256_implCompress"; 8362 break; 8363 case vmIntrinsics::_sha5_implCompress: 8364 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 8365 state = get_state_from_digest_object(digestBase_obj, T_LONG); 8366 stubAddr = StubRoutines::sha512_implCompress(); 8367 stubName = "sha512_implCompress"; 8368 break; 8369 case vmIntrinsics::_sha3_implCompress: 8370 assert(UseSHA3Intrinsics, "need SHA3 instruction support"); 8371 state = get_state_from_digest_object(digestBase_obj, T_LONG); 8372 stubAddr = StubRoutines::sha3_implCompress(); 8373 stubName = "sha3_implCompress"; 8374 block_size = get_block_size_from_digest_object(digestBase_obj); 8375 if (block_size == nullptr) return false; 8376 break; 8377 default: 8378 fatal_unexpected_iid(id); 8379 return false; 8380 } 8381 if (state == nullptr) return false; 8382 8383 assert(stubAddr != nullptr, "Stub %s is not generated", stubName); 8384 if (stubAddr == nullptr) return false; 8385 8386 // Call the stub. 8387 Node* call; 8388 if (block_size == nullptr) { 8389 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false), 8390 stubAddr, stubName, TypePtr::BOTTOM, 8391 src_start, state); 8392 } else { 8393 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true), 8394 stubAddr, stubName, TypePtr::BOTTOM, 8395 src_start, state, block_size); 8396 } 8397 8398 return true; 8399 } 8400 8401 //------------------------------inline_double_keccak 8402 bool LibraryCallKit::inline_double_keccak() { 8403 address stubAddr; 8404 const char *stubName; 8405 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support"); 8406 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters"); 8407 8408 stubAddr = StubRoutines::double_keccak(); 8409 stubName = "double_keccak"; 8410 if (!stubAddr) return false; 8411 8412 Node* status0 = argument(0); 8413 Node* status1 = argument(1); 8414 8415 status0 = must_be_not_null(status0, true); 8416 status1 = must_be_not_null(status1, true); 8417 8418 Node* status0_start = array_element_address(status0, intcon(0), T_LONG); 8419 assert(status0_start, "status0 is null"); 8420 Node* status1_start = array_element_address(status1, intcon(0), T_LONG); 8421 assert(status1_start, "status1 is null"); 8422 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP, 8423 OptoRuntime::double_keccak_Type(), 8424 stubAddr, stubName, TypePtr::BOTTOM, 8425 status0_start, status1_start); 8426 // return an int 8427 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms)); 8428 set_result(retvalue); 8429 return true; 8430 } 8431 8432 8433 //------------------------------inline_digestBase_implCompressMB----------------------- 8434 // 8435 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array. 8436 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 8437 // 8438 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 8439 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics, 8440 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support"); 8441 assert((uint)predicate < 5, "sanity"); 8442 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 8443 8444 Node* digestBase_obj = argument(0); // The receiver was checked for null already. 8445 Node* src = argument(1); // byte[] array 8446 Node* ofs = argument(2); // type int 8447 Node* limit = argument(3); // type int 8448 8449 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr(); 8450 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) { 8451 // failed array check 8452 return false; 8453 } 8454 // Figure out the size and type of the elements we will be copying. 8455 BasicType src_elem = src_type->elem()->array_element_basic_type(); 8456 if (src_elem != T_BYTE) { 8457 return false; 8458 } 8459 // 'src_start' points to src array + offset 8460 src = must_be_not_null(src, false); 8461 Node* src_start = array_element_address(src, ofs, src_elem); 8462 8463 const char* klass_digestBase_name = nullptr; 8464 const char* stub_name = nullptr; 8465 address stub_addr = nullptr; 8466 BasicType elem_type = T_INT; 8467 8468 switch (predicate) { 8469 case 0: 8470 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) { 8471 klass_digestBase_name = "sun/security/provider/MD5"; 8472 stub_name = "md5_implCompressMB"; 8473 stub_addr = StubRoutines::md5_implCompressMB(); 8474 } 8475 break; 8476 case 1: 8477 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) { 8478 klass_digestBase_name = "sun/security/provider/SHA"; 8479 stub_name = "sha1_implCompressMB"; 8480 stub_addr = StubRoutines::sha1_implCompressMB(); 8481 } 8482 break; 8483 case 2: 8484 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) { 8485 klass_digestBase_name = "sun/security/provider/SHA2"; 8486 stub_name = "sha256_implCompressMB"; 8487 stub_addr = StubRoutines::sha256_implCompressMB(); 8488 } 8489 break; 8490 case 3: 8491 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) { 8492 klass_digestBase_name = "sun/security/provider/SHA5"; 8493 stub_name = "sha512_implCompressMB"; 8494 stub_addr = StubRoutines::sha512_implCompressMB(); 8495 elem_type = T_LONG; 8496 } 8497 break; 8498 case 4: 8499 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) { 8500 klass_digestBase_name = "sun/security/provider/SHA3"; 8501 stub_name = "sha3_implCompressMB"; 8502 stub_addr = StubRoutines::sha3_implCompressMB(); 8503 elem_type = T_LONG; 8504 } 8505 break; 8506 default: 8507 fatal("unknown DigestBase intrinsic predicate: %d", predicate); 8508 } 8509 if (klass_digestBase_name != nullptr) { 8510 assert(stub_addr != nullptr, "Stub is generated"); 8511 if (stub_addr == nullptr) return false; 8512 8513 // get DigestBase klass to lookup for SHA klass 8514 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 8515 assert(tinst != nullptr, "digestBase_obj is not instance???"); 8516 assert(tinst->is_loaded(), "DigestBase is not loaded"); 8517 8518 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name)); 8519 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded"); 8520 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass(); 8521 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit); 8522 } 8523 return false; 8524 } 8525 8526 //------------------------------inline_digestBase_implCompressMB----------------------- 8527 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase, 8528 BasicType elem_type, address stubAddr, const char *stubName, 8529 Node* src_start, Node* ofs, Node* limit) { 8530 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase); 8531 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull); 8532 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 8533 digest_obj = _gvn.transform(digest_obj); 8534 8535 Node* state = get_state_from_digest_object(digest_obj, elem_type); 8536 if (state == nullptr) return false; 8537 8538 Node* block_size = nullptr; 8539 if (strcmp("sha3_implCompressMB", stubName) == 0) { 8540 block_size = get_block_size_from_digest_object(digest_obj); 8541 if (block_size == nullptr) return false; 8542 } 8543 8544 // Call the stub. 8545 Node* call; 8546 if (block_size == nullptr) { 8547 call = make_runtime_call(RC_LEAF|RC_NO_FP, 8548 OptoRuntime::digestBase_implCompressMB_Type(false), 8549 stubAddr, stubName, TypePtr::BOTTOM, 8550 src_start, state, ofs, limit); 8551 } else { 8552 call = make_runtime_call(RC_LEAF|RC_NO_FP, 8553 OptoRuntime::digestBase_implCompressMB_Type(true), 8554 stubAddr, stubName, TypePtr::BOTTOM, 8555 src_start, state, block_size, ofs, limit); 8556 } 8557 8558 // return ofs (int) 8559 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 8560 set_result(result); 8561 8562 return true; 8563 } 8564 8565 //------------------------------inline_galoisCounterMode_AESCrypt----------------------- 8566 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() { 8567 assert(UseAES, "need AES instruction support"); 8568 address stubAddr = nullptr; 8569 const char *stubName = nullptr; 8570 stubAddr = StubRoutines::galoisCounterMode_AESCrypt(); 8571 stubName = "galoisCounterMode_AESCrypt"; 8572 8573 if (stubAddr == nullptr) return false; 8574 8575 Node* in = argument(0); 8576 Node* inOfs = argument(1); 8577 Node* len = argument(2); 8578 Node* ct = argument(3); 8579 Node* ctOfs = argument(4); 8580 Node* out = argument(5); 8581 Node* outOfs = argument(6); 8582 Node* gctr_object = argument(7); 8583 Node* ghash_object = argument(8); 8584 8585 // (1) in, ct and out are arrays. 8586 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr(); 8587 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr(); 8588 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr(); 8589 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM && 8590 ct_type != nullptr && ct_type->elem() != Type::BOTTOM && 8591 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange"); 8592 8593 // checks are the responsibility of the caller 8594 Node* in_start = in; 8595 Node* ct_start = ct; 8596 Node* out_start = out; 8597 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) { 8598 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, ""); 8599 in_start = array_element_address(in, inOfs, T_BYTE); 8600 ct_start = array_element_address(ct, ctOfs, T_BYTE); 8601 out_start = array_element_address(out, outOfs, T_BYTE); 8602 } 8603 8604 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 8605 // (because of the predicated logic executed earlier). 8606 // so we cast it here safely. 8607 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 8608 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 8609 Node* counter = load_field_from_object(gctr_object, "counter", "[B"); 8610 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J"); 8611 Node* state = load_field_from_object(ghash_object, "state", "[J"); 8612 8613 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) { 8614 return false; 8615 } 8616 // cast it to what we know it will be at runtime 8617 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr(); 8618 assert(tinst != nullptr, "GCTR obj is null"); 8619 assert(tinst->is_loaded(), "GCTR obj is not loaded"); 8620 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 8621 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 8622 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 8623 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 8624 const TypeOopPtr* xtype = aklass->as_instance_type(); 8625 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 8626 aescrypt_object = _gvn.transform(aescrypt_object); 8627 // we need to get the start of the aescrypt_object's expanded key array 8628 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 8629 if (k_start == nullptr) return false; 8630 // similarly, get the start address of the r vector 8631 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE); 8632 Node* state_start = array_element_address(state, intcon(0), T_LONG); 8633 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG); 8634 8635 8636 // Call the stub, passing params 8637 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 8638 OptoRuntime::galoisCounterMode_aescrypt_Type(), 8639 stubAddr, stubName, TypePtr::BOTTOM, 8640 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start); 8641 8642 // return cipher length (int) 8643 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms)); 8644 set_result(retvalue); 8645 8646 return true; 8647 } 8648 8649 //----------------------------inline_galoisCounterMode_AESCrypt_predicate---------------------------- 8650 // Return node representing slow path of predicate check. 8651 // the pseudo code we want to emulate with this predicate is: 8652 // for encryption: 8653 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 8654 // for decryption: 8655 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 8656 // note cipher==plain is more conservative than the original java code but that's OK 8657 // 8658 8659 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() { 8660 // The receiver was checked for null already. 8661 Node* objGCTR = argument(7); 8662 // Load embeddedCipher field of GCTR object. 8663 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;"); 8664 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null"); 8665 8666 // get AESCrypt klass for instanceOf check 8667 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 8668 // will have same classloader as CipherBlockChaining object 8669 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr(); 8670 assert(tinst != nullptr, "GCTR obj is null"); 8671 assert(tinst->is_loaded(), "GCTR obj is not loaded"); 8672 8673 // we want to do an instanceof comparison against the AESCrypt class 8674 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 8675 if (!klass_AESCrypt->is_loaded()) { 8676 // if AESCrypt is not even loaded, we never take the intrinsic fast path 8677 Node* ctrl = control(); 8678 set_control(top()); // no regular fast path 8679 return ctrl; 8680 } 8681 8682 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 8683 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 8684 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 8685 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 8686 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 8687 8688 return instof_false; // even if it is null 8689 } 8690 8691 //------------------------------get_state_from_digest_object----------------------- 8692 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) { 8693 const char* state_type; 8694 switch (elem_type) { 8695 case T_BYTE: state_type = "[B"; break; 8696 case T_INT: state_type = "[I"; break; 8697 case T_LONG: state_type = "[J"; break; 8698 default: ShouldNotReachHere(); 8699 } 8700 Node* digest_state = load_field_from_object(digest_object, "state", state_type); 8701 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3"); 8702 if (digest_state == nullptr) return (Node *) nullptr; 8703 8704 // now have the array, need to get the start address of the state array 8705 Node* state = array_element_address(digest_state, intcon(0), elem_type); 8706 return state; 8707 } 8708 8709 //------------------------------get_block_size_from_sha3_object---------------------------------- 8710 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) { 8711 Node* block_size = load_field_from_object(digest_object, "blockSize", "I"); 8712 assert (block_size != nullptr, "sanity"); 8713 return block_size; 8714 } 8715 8716 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 8717 // Return node representing slow path of predicate check. 8718 // the pseudo code we want to emulate with this predicate is: 8719 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath 8720 // 8721 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 8722 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics, 8723 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support"); 8724 assert((uint)predicate < 5, "sanity"); 8725 8726 // The receiver was checked for null already. 8727 Node* digestBaseObj = argument(0); 8728 8729 // get DigestBase klass for instanceOf check 8730 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 8731 assert(tinst != nullptr, "digestBaseObj is null"); 8732 assert(tinst->is_loaded(), "DigestBase is not loaded"); 8733 8734 const char* klass_name = nullptr; 8735 switch (predicate) { 8736 case 0: 8737 if (UseMD5Intrinsics) { 8738 // we want to do an instanceof comparison against the MD5 class 8739 klass_name = "sun/security/provider/MD5"; 8740 } 8741 break; 8742 case 1: 8743 if (UseSHA1Intrinsics) { 8744 // we want to do an instanceof comparison against the SHA class 8745 klass_name = "sun/security/provider/SHA"; 8746 } 8747 break; 8748 case 2: 8749 if (UseSHA256Intrinsics) { 8750 // we want to do an instanceof comparison against the SHA2 class 8751 klass_name = "sun/security/provider/SHA2"; 8752 } 8753 break; 8754 case 3: 8755 if (UseSHA512Intrinsics) { 8756 // we want to do an instanceof comparison against the SHA5 class 8757 klass_name = "sun/security/provider/SHA5"; 8758 } 8759 break; 8760 case 4: 8761 if (UseSHA3Intrinsics) { 8762 // we want to do an instanceof comparison against the SHA3 class 8763 klass_name = "sun/security/provider/SHA3"; 8764 } 8765 break; 8766 default: 8767 fatal("unknown SHA intrinsic predicate: %d", predicate); 8768 } 8769 8770 ciKlass* klass = nullptr; 8771 if (klass_name != nullptr) { 8772 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name)); 8773 } 8774 if ((klass == nullptr) || !klass->is_loaded()) { 8775 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 8776 Node* ctrl = control(); 8777 set_control(top()); // no intrinsic path 8778 return ctrl; 8779 } 8780 ciInstanceKlass* instklass = klass->as_instance_klass(); 8781 8782 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass))); 8783 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 8784 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 8785 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN); 8786 8787 return instof_false; // even if it is null 8788 } 8789 8790 //-------------inline_fma----------------------------------- 8791 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) { 8792 Node *a = nullptr; 8793 Node *b = nullptr; 8794 Node *c = nullptr; 8795 Node* result = nullptr; 8796 switch (id) { 8797 case vmIntrinsics::_fmaD: 8798 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each."); 8799 // no receiver since it is static method 8800 a = argument(0); 8801 b = argument(2); 8802 c = argument(4); 8803 result = _gvn.transform(new FmaDNode(a, b, c)); 8804 break; 8805 case vmIntrinsics::_fmaF: 8806 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each."); 8807 a = argument(0); 8808 b = argument(1); 8809 c = argument(2); 8810 result = _gvn.transform(new FmaFNode(a, b, c)); 8811 break; 8812 default: 8813 fatal_unexpected_iid(id); break; 8814 } 8815 set_result(result); 8816 return true; 8817 } 8818 8819 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) { 8820 // argument(0) is receiver 8821 Node* codePoint = argument(1); 8822 Node* n = nullptr; 8823 8824 switch (id) { 8825 case vmIntrinsics::_isDigit : 8826 n = new DigitNode(control(), codePoint); 8827 break; 8828 case vmIntrinsics::_isLowerCase : 8829 n = new LowerCaseNode(control(), codePoint); 8830 break; 8831 case vmIntrinsics::_isUpperCase : 8832 n = new UpperCaseNode(control(), codePoint); 8833 break; 8834 case vmIntrinsics::_isWhitespace : 8835 n = new WhitespaceNode(control(), codePoint); 8836 break; 8837 default: 8838 fatal_unexpected_iid(id); 8839 } 8840 8841 set_result(_gvn.transform(n)); 8842 return true; 8843 } 8844 8845 bool LibraryCallKit::inline_profileBoolean() { 8846 Node* counts = argument(1); 8847 const TypeAryPtr* ary = nullptr; 8848 ciArray* aobj = nullptr; 8849 if (counts->is_Con() 8850 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr 8851 && (aobj = ary->const_oop()->as_array()) != nullptr 8852 && (aobj->length() == 2)) { 8853 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 8854 jint false_cnt = aobj->element_value(0).as_int(); 8855 jint true_cnt = aobj->element_value(1).as_int(); 8856 8857 if (C->log() != nullptr) { 8858 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 8859 false_cnt, true_cnt); 8860 } 8861 8862 if (false_cnt + true_cnt == 0) { 8863 // According to profile, never executed. 8864 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 8865 Deoptimization::Action_reinterpret); 8866 return true; 8867 } 8868 8869 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 8870 // is a number of each value occurrences. 8871 Node* result = argument(0); 8872 if (false_cnt == 0 || true_cnt == 0) { 8873 // According to profile, one value has been never seen. 8874 int expected_val = (false_cnt == 0) ? 1 : 0; 8875 8876 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 8877 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 8878 8879 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 8880 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 8881 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 8882 8883 { // Slow path: uncommon trap for never seen value and then reexecute 8884 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 8885 // the value has been seen at least once. 8886 PreserveJVMState pjvms(this); 8887 PreserveReexecuteState preexecs(this); 8888 jvms()->set_should_reexecute(true); 8889 8890 set_control(slow_path); 8891 set_i_o(i_o()); 8892 8893 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 8894 Deoptimization::Action_reinterpret); 8895 } 8896 // The guard for never seen value enables sharpening of the result and 8897 // returning a constant. It allows to eliminate branches on the same value 8898 // later on. 8899 set_control(fast_path); 8900 result = intcon(expected_val); 8901 } 8902 // Stop profiling. 8903 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 8904 // By replacing method body with profile data (represented as ProfileBooleanNode 8905 // on IR level) we effectively disable profiling. 8906 // It enables full speed execution once optimized code is generated. 8907 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 8908 C->record_for_igvn(profile); 8909 set_result(profile); 8910 return true; 8911 } else { 8912 // Continue profiling. 8913 // Profile data isn't available at the moment. So, execute method's bytecode version. 8914 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 8915 // is compiled and counters aren't available since corresponding MethodHandle 8916 // isn't a compile-time constant. 8917 return false; 8918 } 8919 } 8920 8921 bool LibraryCallKit::inline_isCompileConstant() { 8922 Node* n = argument(0); 8923 set_result(n->is_Con() ? intcon(1) : intcon(0)); 8924 return true; 8925 } 8926 8927 //------------------------------- inline_getObjectSize/sizeOf -------------------------------------- 8928 // 8929 // Calculate the runtime size of the object/array. 8930 // 8931 // long java.lang.Runtime.sizeOf(Object obj) 8932 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize); 8933 // 8934 bool LibraryCallKit::inline_sizeOf() { 8935 Node* obj = argument(0); 8936 return inline_sizeOf_impl(obj); 8937 } 8938 8939 bool LibraryCallKit::inline_getObjectSize() { 8940 Node* obj = argument(3); 8941 return inline_sizeOf_impl(obj); 8942 } 8943 8944 bool LibraryCallKit::inline_sizeOf_impl(Node* obj) { 8945 Node* klass_node = load_object_klass(obj); 8946 8947 jint layout_con = Klass::_lh_neutral_value; 8948 Node* layout_val = get_layout_helper(klass_node, layout_con); 8949 int layout_is_con = (layout_val == nullptr); 8950 8951 if (layout_is_con) { 8952 // Layout helper is constant, can figure out things at compile time. 8953 8954 if (Klass::layout_helper_is_instance(layout_con)) { 8955 // Instance case: layout_con contains the size itself. 8956 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con)); 8957 set_result(size); 8958 } else { 8959 // Array case: size is round(header + element_size*arraylength). 8960 // Since arraylength is different for every array instance, we have to 8961 // compute the whole thing at runtime. 8962 8963 Node* arr_length = load_array_length(obj); 8964 8965 int round_mask = MinObjAlignmentInBytes - 1; 8966 int hsize = Klass::layout_helper_header_size(layout_con); 8967 int eshift = Klass::layout_helper_log2_element_size(layout_con); 8968 8969 if ((round_mask & ~right_n_bits(eshift)) == 0) { 8970 round_mask = 0; // strength-reduce it if it goes away completely 8971 } 8972 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded"); 8973 Node* header_size = intcon(hsize + round_mask); 8974 8975 Node* lengthx = ConvI2X(arr_length); 8976 Node* headerx = ConvI2X(header_size); 8977 8978 Node* abody = lengthx; 8979 if (eshift != 0) { 8980 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift))); 8981 } 8982 Node* size = _gvn.transform( new AddXNode(headerx, abody) ); 8983 if (round_mask != 0) { 8984 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) ); 8985 } 8986 size = ConvX2L(size); 8987 set_result(size); 8988 } 8989 } else { 8990 // Layout helper is not constant, need to test for array-ness at runtime. 8991 8992 enum { _instance_path = 1, _array_path, PATH_LIMIT }; 8993 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 8994 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG); 8995 record_for_igvn(result_reg); 8996 8997 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj); 8998 if (array_ctl != nullptr) { 8999 // Array case: size is round(header + element_size*arraylength). 9000 // Since arraylength is different for every array instance, we have to 9001 // compute the whole thing at runtime. 9002 9003 PreserveJVMState pjvms(this); 9004 set_control(array_ctl); 9005 Node* arr_length = load_array_length(obj); 9006 9007 int round_mask = MinObjAlignmentInBytes - 1; 9008 Node* mask = intcon(round_mask); 9009 9010 Node* hss = intcon(Klass::_lh_header_size_shift); 9011 Node* hsm = intcon(Klass::_lh_header_size_mask); 9012 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss)); 9013 header_size = _gvn.transform(new AndINode(header_size, hsm)); 9014 header_size = _gvn.transform(new AddINode(header_size, mask)); 9015 9016 // There is no need to mask or shift this value. 9017 // The semantics of LShiftINode include an implicit mask to 0x1F. 9018 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place"); 9019 Node* elem_shift = layout_val; 9020 9021 Node* lengthx = ConvI2X(arr_length); 9022 Node* headerx = ConvI2X(header_size); 9023 9024 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift)); 9025 Node* size = _gvn.transform(new AddXNode(headerx, abody)); 9026 if (round_mask != 0) { 9027 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask))); 9028 } 9029 size = ConvX2L(size); 9030 9031 result_reg->init_req(_array_path, control()); 9032 result_val->init_req(_array_path, size); 9033 } 9034 9035 if (!stopped()) { 9036 // Instance case: the layout helper gives us instance size almost directly, 9037 // but we need to mask out the _lh_instance_slow_path_bit. 9038 Node* size = ConvI2X(layout_val); 9039 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit"); 9040 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong)); 9041 size = _gvn.transform(new AndXNode(size, mask)); 9042 size = ConvX2L(size); 9043 9044 result_reg->init_req(_instance_path, control()); 9045 result_val->init_req(_instance_path, size); 9046 } 9047 9048 set_result(result_reg, result_val); 9049 } 9050 9051 return true; 9052 } 9053 9054 //------------------------------- inline_blackhole -------------------------------------- 9055 // 9056 // Make sure all arguments to this node are alive. 9057 // This matches methods that were requested to be blackholed through compile commands. 9058 // 9059 bool LibraryCallKit::inline_blackhole() { 9060 assert(callee()->is_static(), "Should have been checked before: only static methods here"); 9061 assert(callee()->is_empty(), "Should have been checked before: only empty methods here"); 9062 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here"); 9063 9064 // Blackhole node pinches only the control, not memory. This allows 9065 // the blackhole to be pinned in the loop that computes blackholed 9066 // values, but have no other side effects, like breaking the optimizations 9067 // across the blackhole. 9068 9069 Node* bh = _gvn.transform(new BlackholeNode(control())); 9070 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control))); 9071 9072 // Bind call arguments as blackhole arguments to keep them alive 9073 uint nargs = callee()->arg_size(); 9074 for (uint i = 0; i < nargs; i++) { 9075 bh->add_req(argument(i)); 9076 } 9077 9078 return true; 9079 } 9080 9081 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) { 9082 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr(); 9083 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) { 9084 return nullptr; // box klass is not Float16 9085 } 9086 9087 // Null check; get notnull casted pointer 9088 Node* null_ctl = top(); 9089 Node* not_null_box = null_check_oop(box, &null_ctl, true); 9090 // If not_null_box is dead, only null-path is taken 9091 if (stopped()) { 9092 set_control(null_ctl); 9093 return nullptr; 9094 } 9095 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, ""); 9096 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 9097 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes()); 9098 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP); 9099 } 9100 9101 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) { 9102 PreserveReexecuteState preexecs(this); 9103 jvms()->set_should_reexecute(true); 9104 9105 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type(); 9106 Node* klass_node = makecon(klass_type); 9107 Node* box = new_instance(klass_node); 9108 9109 Node* value_field = basic_plus_adr(box, field->offset_in_bytes()); 9110 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr(); 9111 9112 Node* field_store = _gvn.transform(access_store_at(box, 9113 value_field, 9114 value_adr_type, 9115 value, 9116 TypeInt::SHORT, 9117 T_SHORT, 9118 IN_HEAP)); 9119 set_memory(field_store, value_adr_type); 9120 return box; 9121 } 9122 9123 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) { 9124 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) || 9125 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) { 9126 return false; 9127 } 9128 9129 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr(); 9130 if (box_type == nullptr || box_type->const_oop() == nullptr) { 9131 return false; 9132 } 9133 9134 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass(); 9135 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass); 9136 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(), 9137 ciSymbols::short_signature(), 9138 false); 9139 assert(field != nullptr, ""); 9140 9141 // Transformed nodes 9142 Node* fld1 = nullptr; 9143 Node* fld2 = nullptr; 9144 Node* fld3 = nullptr; 9145 switch(num_args) { 9146 case 3: 9147 fld3 = unbox_fp16_value(float16_box_type, field, argument(3)); 9148 if (fld3 == nullptr) { 9149 return false; 9150 } 9151 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3)); 9152 // fall-through 9153 case 2: 9154 fld2 = unbox_fp16_value(float16_box_type, field, argument(2)); 9155 if (fld2 == nullptr) { 9156 return false; 9157 } 9158 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2)); 9159 // fall-through 9160 case 1: 9161 fld1 = unbox_fp16_value(float16_box_type, field, argument(1)); 9162 if (fld1 == nullptr) { 9163 return false; 9164 } 9165 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1)); 9166 break; 9167 default: fatal("Unsupported number of arguments %d", num_args); 9168 } 9169 9170 Node* result = nullptr; 9171 switch (id) { 9172 // Unary operations 9173 case vmIntrinsics::_sqrt_float16: 9174 result = _gvn.transform(new SqrtHFNode(C, control(), fld1)); 9175 break; 9176 // Ternary operations 9177 case vmIntrinsics::_fma_float16: 9178 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3)); 9179 break; 9180 default: 9181 fatal_unexpected_iid(id); 9182 break; 9183 } 9184 result = _gvn.transform(new ReinterpretHF2SNode(result)); 9185 set_result(box_fp16_value(float16_box_type, field, result)); 9186 return true; 9187 } 9188 9189 bool LibraryCallKit::inline_timestamp(bool serial) { 9190 insert_mem_bar(Op_MemBarCPUOrder); 9191 Node* node = serial ? 9192 _gvn.transform(new TimestampSerialNode(control())) : 9193 _gvn.transform(new TimestampNode(control())); 9194 set_result(node); 9195 insert_mem_bar(Op_MemBarCPUOrder); 9196 return true; 9197 } 9198 9199 bool LibraryCallKit::inline_addressOf() { 9200 Node* obj = argument(0); 9201 Node* raw_val = _gvn.transform(new CastP2XNode(nullptr, obj)); 9202 Node* long_val = ConvX2L(raw_val); 9203 set_result(long_val); 9204 return true; 9205 }