Mon, 23 Jun 2008 14:11:12 -0700
6708714: Optimize long LShift on 32-bits x86
Summary: For small (1-3 bits) left long shifts in 32-bits VM use sets of add+addc instructions instead of shld+shl on new AMD cpus.
Reviewed-by: never
Contributed-by: shrinivas.joshi@amd.com
duke@435 | 1 | /* |
duke@435 | 2 | * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
duke@435 | 19 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
duke@435 | 20 | * CA 95054 USA or visit www.sun.com if you need additional information or |
duke@435 | 21 | * have any questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
duke@435 | 25 | // The following classes are used for operations |
duke@435 | 26 | // initiated by a Java thread but that must |
duke@435 | 27 | // take place in the VMThread. |
duke@435 | 28 | |
duke@435 | 29 | #define VM_OP_ENUM(type) VMOp_##type, |
duke@435 | 30 | |
duke@435 | 31 | // Note: When new VM_XXX comes up, add 'XXX' to the template table. |
duke@435 | 32 | #define VM_OPS_DO(template) \ |
duke@435 | 33 | template(Dummy) \ |
duke@435 | 34 | template(ThreadStop) \ |
duke@435 | 35 | template(ThreadDump) \ |
duke@435 | 36 | template(PrintThreads) \ |
duke@435 | 37 | template(FindDeadlocks) \ |
duke@435 | 38 | template(ForceSafepoint) \ |
duke@435 | 39 | template(ForceAsyncSafepoint) \ |
duke@435 | 40 | template(Deoptimize) \ |
duke@435 | 41 | template(DeoptimizeFrame) \ |
duke@435 | 42 | template(DeoptimizeAll) \ |
duke@435 | 43 | template(ZombieAll) \ |
duke@435 | 44 | template(Verify) \ |
duke@435 | 45 | template(PrintJNI) \ |
duke@435 | 46 | template(HeapDumper) \ |
duke@435 | 47 | template(DeoptimizeTheWorld) \ |
duke@435 | 48 | template(GC_HeapInspection) \ |
duke@435 | 49 | template(GenCollectFull) \ |
duke@435 | 50 | template(GenCollectFullConcurrent) \ |
duke@435 | 51 | template(GenCollectForAllocation) \ |
apetrusenko@574 | 52 | template(GenCollectForPermanentAllocation) \ |
duke@435 | 53 | template(ParallelGCFailedAllocation) \ |
duke@435 | 54 | template(ParallelGCFailedPermanentAllocation) \ |
duke@435 | 55 | template(ParallelGCSystemGC) \ |
duke@435 | 56 | template(CMS_Initial_Mark) \ |
duke@435 | 57 | template(CMS_Final_Remark) \ |
duke@435 | 58 | template(EnableBiasedLocking) \ |
duke@435 | 59 | template(RevokeBias) \ |
duke@435 | 60 | template(BulkRevokeBias) \ |
duke@435 | 61 | template(PopulateDumpSharedSpace) \ |
duke@435 | 62 | template(JNIFunctionTableCopier) \ |
duke@435 | 63 | template(RedefineClasses) \ |
duke@435 | 64 | template(GetOwnedMonitorInfo) \ |
duke@435 | 65 | template(GetObjectMonitorUsage) \ |
duke@435 | 66 | template(GetCurrentContendedMonitor) \ |
duke@435 | 67 | template(GetStackTrace) \ |
duke@435 | 68 | template(GetMultipleStackTraces) \ |
duke@435 | 69 | template(GetAllStackTraces) \ |
duke@435 | 70 | template(GetThreadListStackTraces) \ |
duke@435 | 71 | template(GetFrameCount) \ |
duke@435 | 72 | template(GetFrameLocation) \ |
duke@435 | 73 | template(ChangeBreakpoints) \ |
duke@435 | 74 | template(GetOrSetLocal) \ |
duke@435 | 75 | template(GetCurrentLocation) \ |
duke@435 | 76 | template(EnterInterpOnlyMode) \ |
duke@435 | 77 | template(ChangeSingleStep) \ |
duke@435 | 78 | template(HeapWalkOperation) \ |
duke@435 | 79 | template(HeapIterateOperation) \ |
duke@435 | 80 | template(ReportJavaOutOfMemory) \ |
duke@435 | 81 | template(Exit) \ |
duke@435 | 82 | |
duke@435 | 83 | class VM_Operation: public CHeapObj { |
duke@435 | 84 | public: |
duke@435 | 85 | enum Mode { |
duke@435 | 86 | _safepoint, // blocking, safepoint, vm_op C-heap allocated |
duke@435 | 87 | _no_safepoint, // blocking, no safepoint, vm_op C-Heap allocated |
duke@435 | 88 | _concurrent, // non-blocking, no safepoint, vm_op C-Heap allocated |
duke@435 | 89 | _async_safepoint // non-blocking, safepoint, vm_op C-Heap allocated |
duke@435 | 90 | }; |
duke@435 | 91 | |
duke@435 | 92 | enum VMOp_Type { |
duke@435 | 93 | VM_OPS_DO(VM_OP_ENUM) |
duke@435 | 94 | VMOp_Terminating |
duke@435 | 95 | }; |
duke@435 | 96 | |
duke@435 | 97 | private: |
duke@435 | 98 | Thread* _calling_thread; |
duke@435 | 99 | ThreadPriority _priority; |
duke@435 | 100 | long _timestamp; |
duke@435 | 101 | VM_Operation* _next; |
duke@435 | 102 | VM_Operation* _prev; |
duke@435 | 103 | |
duke@435 | 104 | // The VM operation name array |
duke@435 | 105 | static const char* _names[]; |
duke@435 | 106 | |
duke@435 | 107 | public: |
duke@435 | 108 | VM_Operation() { _calling_thread = NULL; _next = NULL; _prev = NULL; } |
duke@435 | 109 | virtual ~VM_Operation() {} |
duke@435 | 110 | |
duke@435 | 111 | // VM operation support (used by VM thread) |
duke@435 | 112 | Thread* calling_thread() const { return _calling_thread; } |
duke@435 | 113 | ThreadPriority priority() { return _priority; } |
duke@435 | 114 | void set_calling_thread(Thread* thread, ThreadPriority priority); |
duke@435 | 115 | |
duke@435 | 116 | long timestamp() const { return _timestamp; } |
duke@435 | 117 | void set_timestamp(long timestamp) { _timestamp = timestamp; } |
duke@435 | 118 | |
duke@435 | 119 | // Called by VM thread - does in turn invoke doit(). Do not override this |
duke@435 | 120 | void evaluate(); |
duke@435 | 121 | |
duke@435 | 122 | // evaluate() is called by the VMThread and in turn calls doit(). |
duke@435 | 123 | // If the thread invoking VMThread::execute((VM_Operation*) is a JavaThread, |
duke@435 | 124 | // doit_prologue() is called in that thread before transferring control to |
duke@435 | 125 | // the VMThread. |
duke@435 | 126 | // If doit_prologue() returns true the VM operation will proceed, and |
duke@435 | 127 | // doit_epilogue() will be called by the JavaThread once the VM operation |
duke@435 | 128 | // completes. If doit_prologue() returns false the VM operation is cancelled. |
duke@435 | 129 | virtual void doit() = 0; |
duke@435 | 130 | virtual bool doit_prologue() { return true; }; |
duke@435 | 131 | virtual void doit_epilogue() {}; // Note: Not called if mode is: _concurrent |
duke@435 | 132 | |
duke@435 | 133 | // Type test |
duke@435 | 134 | virtual bool is_methodCompiler() const { return false; } |
duke@435 | 135 | |
duke@435 | 136 | // Linking |
duke@435 | 137 | VM_Operation *next() const { return _next; } |
duke@435 | 138 | VM_Operation *prev() const { return _prev; } |
duke@435 | 139 | void set_next(VM_Operation *next) { _next = next; } |
duke@435 | 140 | void set_prev(VM_Operation *prev) { _prev = prev; } |
duke@435 | 141 | |
duke@435 | 142 | // Configuration. Override these appropriatly in subclasses. |
duke@435 | 143 | virtual VMOp_Type type() const = 0; |
duke@435 | 144 | virtual Mode evaluation_mode() const { return _safepoint; } |
duke@435 | 145 | virtual bool allow_nested_vm_operations() const { return false; } |
duke@435 | 146 | virtual bool is_cheap_allocated() const { return false; } |
duke@435 | 147 | virtual void oops_do(OopClosure* f) { /* do nothing */ }; |
duke@435 | 148 | |
duke@435 | 149 | // CAUTION: <don't hang yourself with following rope> |
duke@435 | 150 | // If you override these methods, make sure that the evaluation |
duke@435 | 151 | // of these methods is race-free and non-blocking, since these |
duke@435 | 152 | // methods may be evaluated either by the mutators or by the |
duke@435 | 153 | // vm thread, either concurrently with mutators or with the mutators |
duke@435 | 154 | // stopped. In other words, taking locks is verboten, and if there |
duke@435 | 155 | // are any races in evaluating the conditions, they'd better be benign. |
duke@435 | 156 | virtual bool evaluate_at_safepoint() const { |
duke@435 | 157 | return evaluation_mode() == _safepoint || |
duke@435 | 158 | evaluation_mode() == _async_safepoint; |
duke@435 | 159 | } |
duke@435 | 160 | virtual bool evaluate_concurrently() const { |
duke@435 | 161 | return evaluation_mode() == _concurrent || |
duke@435 | 162 | evaluation_mode() == _async_safepoint; |
duke@435 | 163 | } |
duke@435 | 164 | |
duke@435 | 165 | // Debugging |
duke@435 | 166 | void print_on_error(outputStream* st) const; |
duke@435 | 167 | const char* name() const { return _names[type()]; } |
duke@435 | 168 | static const char* name(int type) { |
duke@435 | 169 | assert(type >= 0 && type < VMOp_Terminating, "invalid VM operation type"); |
duke@435 | 170 | return _names[type]; |
duke@435 | 171 | } |
duke@435 | 172 | #ifndef PRODUCT |
duke@435 | 173 | void print_on(outputStream* st) const { print_on_error(st); } |
duke@435 | 174 | #endif |
duke@435 | 175 | }; |
duke@435 | 176 | |
duke@435 | 177 | class VM_ThreadStop: public VM_Operation { |
duke@435 | 178 | private: |
duke@435 | 179 | oop _thread; // The Thread that the Throwable is thrown against |
duke@435 | 180 | oop _throwable; // The Throwable thrown at the target Thread |
duke@435 | 181 | public: |
duke@435 | 182 | // All oops are passed as JNI handles, since there is no guarantee that a GC might happen before the |
duke@435 | 183 | // VM operation is executed. |
duke@435 | 184 | VM_ThreadStop(oop thread, oop throwable) { |
duke@435 | 185 | _thread = thread; |
duke@435 | 186 | _throwable = throwable; |
duke@435 | 187 | } |
duke@435 | 188 | VMOp_Type type() const { return VMOp_ThreadStop; } |
duke@435 | 189 | oop target_thread() const { return _thread; } |
duke@435 | 190 | oop throwable() const { return _throwable;} |
duke@435 | 191 | void doit(); |
duke@435 | 192 | // We deoptimize if top-most frame is compiled - this might require a C2I adapter to be generated |
duke@435 | 193 | bool allow_nested_vm_operations() const { return true; } |
duke@435 | 194 | Mode evaluation_mode() const { return _async_safepoint; } |
duke@435 | 195 | bool is_cheap_allocated() const { return true; } |
duke@435 | 196 | |
duke@435 | 197 | // GC support |
duke@435 | 198 | void oops_do(OopClosure* f) { |
duke@435 | 199 | f->do_oop(&_thread); f->do_oop(&_throwable); |
duke@435 | 200 | } |
duke@435 | 201 | }; |
duke@435 | 202 | |
duke@435 | 203 | // dummy vm op, evaluated just to force a safepoint |
duke@435 | 204 | class VM_ForceSafepoint: public VM_Operation { |
duke@435 | 205 | public: |
duke@435 | 206 | VM_ForceSafepoint() {} |
duke@435 | 207 | void doit() {} |
duke@435 | 208 | VMOp_Type type() const { return VMOp_ForceSafepoint; } |
duke@435 | 209 | }; |
duke@435 | 210 | |
duke@435 | 211 | // dummy vm op, evaluated just to force a safepoint |
duke@435 | 212 | class VM_ForceAsyncSafepoint: public VM_Operation { |
duke@435 | 213 | public: |
duke@435 | 214 | VM_ForceAsyncSafepoint() {} |
duke@435 | 215 | void doit() {} |
duke@435 | 216 | VMOp_Type type() const { return VMOp_ForceAsyncSafepoint; } |
duke@435 | 217 | Mode evaluation_mode() const { return _async_safepoint; } |
duke@435 | 218 | bool is_cheap_allocated() const { return true; } |
duke@435 | 219 | }; |
duke@435 | 220 | |
duke@435 | 221 | class VM_Deoptimize: public VM_Operation { |
duke@435 | 222 | public: |
duke@435 | 223 | VM_Deoptimize() {} |
duke@435 | 224 | VMOp_Type type() const { return VMOp_Deoptimize; } |
duke@435 | 225 | void doit(); |
duke@435 | 226 | bool allow_nested_vm_operations() const { return true; } |
duke@435 | 227 | }; |
duke@435 | 228 | |
duke@435 | 229 | class VM_DeoptimizeFrame: public VM_Operation { |
duke@435 | 230 | private: |
duke@435 | 231 | JavaThread* _thread; |
duke@435 | 232 | intptr_t* _id; |
duke@435 | 233 | public: |
duke@435 | 234 | VM_DeoptimizeFrame(JavaThread* thread, intptr_t* id); |
duke@435 | 235 | VMOp_Type type() const { return VMOp_DeoptimizeFrame; } |
duke@435 | 236 | void doit(); |
duke@435 | 237 | bool allow_nested_vm_operations() const { return true; } |
duke@435 | 238 | }; |
duke@435 | 239 | |
duke@435 | 240 | #ifndef PRODUCT |
duke@435 | 241 | class VM_DeoptimizeAll: public VM_Operation { |
duke@435 | 242 | private: |
duke@435 | 243 | KlassHandle _dependee; |
duke@435 | 244 | public: |
duke@435 | 245 | VM_DeoptimizeAll() {} |
duke@435 | 246 | VMOp_Type type() const { return VMOp_DeoptimizeAll; } |
duke@435 | 247 | void doit(); |
duke@435 | 248 | bool allow_nested_vm_operations() const { return true; } |
duke@435 | 249 | }; |
duke@435 | 250 | |
duke@435 | 251 | |
duke@435 | 252 | class VM_ZombieAll: public VM_Operation { |
duke@435 | 253 | public: |
duke@435 | 254 | VM_ZombieAll() {} |
duke@435 | 255 | VMOp_Type type() const { return VMOp_ZombieAll; } |
duke@435 | 256 | void doit(); |
duke@435 | 257 | bool allow_nested_vm_operations() const { return true; } |
duke@435 | 258 | }; |
duke@435 | 259 | #endif // PRODUCT |
duke@435 | 260 | |
duke@435 | 261 | class VM_Verify: public VM_Operation { |
duke@435 | 262 | private: |
duke@435 | 263 | KlassHandle _dependee; |
duke@435 | 264 | public: |
duke@435 | 265 | VM_Verify() {} |
duke@435 | 266 | VMOp_Type type() const { return VMOp_Verify; } |
duke@435 | 267 | void doit(); |
duke@435 | 268 | }; |
duke@435 | 269 | |
duke@435 | 270 | |
duke@435 | 271 | class VM_PrintThreads: public VM_Operation { |
duke@435 | 272 | private: |
duke@435 | 273 | outputStream* _out; |
duke@435 | 274 | bool _print_concurrent_locks; |
duke@435 | 275 | public: |
duke@435 | 276 | VM_PrintThreads() { _out = tty; _print_concurrent_locks = PrintConcurrentLocks; } |
duke@435 | 277 | VM_PrintThreads(outputStream* out, bool print_concurrent_locks) { _out = out; _print_concurrent_locks = print_concurrent_locks; } |
duke@435 | 278 | VMOp_Type type() const { return VMOp_PrintThreads; } |
duke@435 | 279 | void doit(); |
duke@435 | 280 | bool doit_prologue(); |
duke@435 | 281 | void doit_epilogue(); |
duke@435 | 282 | }; |
duke@435 | 283 | |
duke@435 | 284 | class VM_PrintJNI: public VM_Operation { |
duke@435 | 285 | private: |
duke@435 | 286 | outputStream* _out; |
duke@435 | 287 | public: |
duke@435 | 288 | VM_PrintJNI() { _out = tty; } |
duke@435 | 289 | VM_PrintJNI(outputStream* out) { _out = out; } |
duke@435 | 290 | VMOp_Type type() const { return VMOp_PrintJNI; } |
duke@435 | 291 | void doit(); |
duke@435 | 292 | }; |
duke@435 | 293 | |
duke@435 | 294 | class DeadlockCycle; |
duke@435 | 295 | class VM_FindDeadlocks: public VM_Operation { |
duke@435 | 296 | private: |
duke@435 | 297 | bool _concurrent_locks; |
duke@435 | 298 | DeadlockCycle* _deadlocks; |
duke@435 | 299 | outputStream* _out; |
duke@435 | 300 | |
duke@435 | 301 | public: |
duke@435 | 302 | VM_FindDeadlocks(bool concurrent_locks) : _concurrent_locks(concurrent_locks), _out(NULL), _deadlocks(NULL) {}; |
duke@435 | 303 | VM_FindDeadlocks(outputStream* st) : _concurrent_locks(true), _out(st), _deadlocks(NULL) {}; |
duke@435 | 304 | ~VM_FindDeadlocks(); |
duke@435 | 305 | |
duke@435 | 306 | DeadlockCycle* result() { return _deadlocks; }; |
duke@435 | 307 | VMOp_Type type() const { return VMOp_FindDeadlocks; } |
duke@435 | 308 | void doit(); |
duke@435 | 309 | bool doit_prologue(); |
duke@435 | 310 | }; |
duke@435 | 311 | |
duke@435 | 312 | class ThreadDumpResult; |
duke@435 | 313 | class ThreadSnapshot; |
duke@435 | 314 | class ThreadConcurrentLocks; |
duke@435 | 315 | |
duke@435 | 316 | class VM_ThreadDump : public VM_Operation { |
duke@435 | 317 | private: |
duke@435 | 318 | ThreadDumpResult* _result; |
duke@435 | 319 | int _num_threads; |
duke@435 | 320 | GrowableArray<instanceHandle>* _threads; |
duke@435 | 321 | int _max_depth; |
duke@435 | 322 | bool _with_locked_monitors; |
duke@435 | 323 | bool _with_locked_synchronizers; |
duke@435 | 324 | |
duke@435 | 325 | ThreadSnapshot* snapshot_thread(JavaThread* java_thread, ThreadConcurrentLocks* tcl); |
duke@435 | 326 | |
duke@435 | 327 | public: |
duke@435 | 328 | VM_ThreadDump(ThreadDumpResult* result, |
duke@435 | 329 | int max_depth, // -1 indicates entire stack |
duke@435 | 330 | bool with_locked_monitors, |
duke@435 | 331 | bool with_locked_synchronizers); |
duke@435 | 332 | |
duke@435 | 333 | VM_ThreadDump(ThreadDumpResult* result, |
duke@435 | 334 | GrowableArray<instanceHandle>* threads, |
duke@435 | 335 | int num_threads, // -1 indicates entire stack |
duke@435 | 336 | int max_depth, |
duke@435 | 337 | bool with_locked_monitors, |
duke@435 | 338 | bool with_locked_synchronizers); |
duke@435 | 339 | |
duke@435 | 340 | VMOp_Type type() const { return VMOp_ThreadDump; } |
duke@435 | 341 | void doit(); |
duke@435 | 342 | bool doit_prologue(); |
duke@435 | 343 | void doit_epilogue(); |
duke@435 | 344 | }; |
duke@435 | 345 | |
duke@435 | 346 | |
duke@435 | 347 | class VM_Exit: public VM_Operation { |
duke@435 | 348 | private: |
duke@435 | 349 | int _exit_code; |
duke@435 | 350 | static volatile bool _vm_exited; |
duke@435 | 351 | static Thread * _shutdown_thread; |
duke@435 | 352 | static void wait_if_vm_exited(); |
duke@435 | 353 | public: |
duke@435 | 354 | VM_Exit(int exit_code) { |
duke@435 | 355 | _exit_code = exit_code; |
duke@435 | 356 | } |
duke@435 | 357 | static int wait_for_threads_in_native_to_block(); |
duke@435 | 358 | static int set_vm_exited(); |
duke@435 | 359 | static bool vm_exited() { return _vm_exited; } |
duke@435 | 360 | static void block_if_vm_exited() { |
duke@435 | 361 | if (_vm_exited) { |
duke@435 | 362 | wait_if_vm_exited(); |
duke@435 | 363 | } |
duke@435 | 364 | } |
duke@435 | 365 | VMOp_Type type() const { return VMOp_Exit; } |
duke@435 | 366 | void doit(); |
duke@435 | 367 | }; |