1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/share/vm/c1/c1_IR.cpp Sat Dec 01 00:00:00 2007 +0000
1.3 @@ -0,0 +1,1323 @@
1.4 +/*
1.5 + * Copyright 1999-2006 Sun Microsystems, Inc. All Rights Reserved.
1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
1.7 + *
1.8 + * This code is free software; you can redistribute it and/or modify it
1.9 + * under the terms of the GNU General Public License version 2 only, as
1.10 + * published by the Free Software Foundation.
1.11 + *
1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT
1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
1.15 + * version 2 for more details (a copy is included in the LICENSE file that
1.16 + * accompanied this code).
1.17 + *
1.18 + * You should have received a copy of the GNU General Public License version
1.19 + * 2 along with this work; if not, write to the Free Software Foundation,
1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
1.21 + *
1.22 + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
1.23 + * CA 95054 USA or visit www.sun.com if you need additional information or
1.24 + * have any questions.
1.25 + *
1.26 + */
1.27 +
1.28 +# include "incls/_precompiled.incl"
1.29 +# include "incls/_c1_IR.cpp.incl"
1.30 +
1.31 +
1.32 +// Implementation of XHandlers
1.33 +//
1.34 +// Note: This code could eventually go away if we are
1.35 +// just using the ciExceptionHandlerStream.
1.36 +
1.37 +XHandlers::XHandlers(ciMethod* method) : _list(method->exception_table_length()) {
1.38 + ciExceptionHandlerStream s(method);
1.39 + while (!s.is_done()) {
1.40 + _list.append(new XHandler(s.handler()));
1.41 + s.next();
1.42 + }
1.43 + assert(s.count() == method->exception_table_length(), "exception table lengths inconsistent");
1.44 +}
1.45 +
1.46 +// deep copy of all XHandler contained in list
1.47 +XHandlers::XHandlers(XHandlers* other) :
1.48 + _list(other->length())
1.49 +{
1.50 + for (int i = 0; i < other->length(); i++) {
1.51 + _list.append(new XHandler(other->handler_at(i)));
1.52 + }
1.53 +}
1.54 +
1.55 +// Returns whether a particular exception type can be caught. Also
1.56 +// returns true if klass is unloaded or any exception handler
1.57 +// classes are unloaded. type_is_exact indicates whether the throw
1.58 +// is known to be exactly that class or it might throw a subtype.
1.59 +bool XHandlers::could_catch(ciInstanceKlass* klass, bool type_is_exact) const {
1.60 + // the type is unknown so be conservative
1.61 + if (!klass->is_loaded()) {
1.62 + return true;
1.63 + }
1.64 +
1.65 + for (int i = 0; i < length(); i++) {
1.66 + XHandler* handler = handler_at(i);
1.67 + if (handler->is_catch_all()) {
1.68 + // catch of ANY
1.69 + return true;
1.70 + }
1.71 + ciInstanceKlass* handler_klass = handler->catch_klass();
1.72 + // if it's unknown it might be catchable
1.73 + if (!handler_klass->is_loaded()) {
1.74 + return true;
1.75 + }
1.76 + // if the throw type is definitely a subtype of the catch type
1.77 + // then it can be caught.
1.78 + if (klass->is_subtype_of(handler_klass)) {
1.79 + return true;
1.80 + }
1.81 + if (!type_is_exact) {
1.82 + // If the type isn't exactly known then it can also be caught by
1.83 + // catch statements where the inexact type is a subtype of the
1.84 + // catch type.
1.85 + // given: foo extends bar extends Exception
1.86 + // throw bar can be caught by catch foo, catch bar, and catch
1.87 + // Exception, however it can't be caught by any handlers without
1.88 + // bar in its type hierarchy.
1.89 + if (handler_klass->is_subtype_of(klass)) {
1.90 + return true;
1.91 + }
1.92 + }
1.93 + }
1.94 +
1.95 + return false;
1.96 +}
1.97 +
1.98 +
1.99 +bool XHandlers::equals(XHandlers* others) const {
1.100 + if (others == NULL) return false;
1.101 + if (length() != others->length()) return false;
1.102 +
1.103 + for (int i = 0; i < length(); i++) {
1.104 + if (!handler_at(i)->equals(others->handler_at(i))) return false;
1.105 + }
1.106 + return true;
1.107 +}
1.108 +
1.109 +bool XHandler::equals(XHandler* other) const {
1.110 + assert(entry_pco() != -1 && other->entry_pco() != -1, "must have entry_pco");
1.111 +
1.112 + if (entry_pco() != other->entry_pco()) return false;
1.113 + if (scope_count() != other->scope_count()) return false;
1.114 + if (_desc != other->_desc) return false;
1.115 +
1.116 + assert(entry_block() == other->entry_block(), "entry_block must be equal when entry_pco is equal");
1.117 + return true;
1.118 +}
1.119 +
1.120 +
1.121 +// Implementation of IRScope
1.122 +
1.123 +BlockBegin* IRScope::header_block(BlockBegin* entry, BlockBegin::Flag f, ValueStack* state) {
1.124 + if (entry == NULL) return NULL;
1.125 + assert(entry->is_set(f), "entry/flag mismatch");
1.126 + // create header block
1.127 + BlockBegin* h = new BlockBegin(entry->bci());
1.128 + BlockEnd* g = new Goto(entry, false);
1.129 + h->set_next(g, entry->bci());
1.130 + h->set_end(g);
1.131 + h->set(f);
1.132 + // setup header block end state
1.133 + ValueStack* s = state->copy(); // can use copy since stack is empty (=> no phis)
1.134 + assert(s->stack_is_empty(), "must have empty stack at entry point");
1.135 + g->set_state(s);
1.136 + return h;
1.137 +}
1.138 +
1.139 +
1.140 +BlockBegin* IRScope::build_graph(Compilation* compilation, int osr_bci) {
1.141 + GraphBuilder gm(compilation, this);
1.142 + NOT_PRODUCT(if (PrintValueNumbering && Verbose) gm.print_stats());
1.143 + if (compilation->bailed_out()) return NULL;
1.144 + return gm.start();
1.145 +}
1.146 +
1.147 +
1.148 +IRScope::IRScope(Compilation* compilation, IRScope* caller, int caller_bci, ciMethod* method, int osr_bci, bool create_graph)
1.149 +: _callees(2)
1.150 +, _compilation(compilation)
1.151 +, _lock_stack_size(-1)
1.152 +, _requires_phi_function(method->max_locals())
1.153 +{
1.154 + _caller = caller;
1.155 + _caller_bci = caller == NULL ? -1 : caller_bci;
1.156 + _caller_state = NULL; // Must be set later if needed
1.157 + _level = caller == NULL ? 0 : caller->level() + 1;
1.158 + _method = method;
1.159 + _xhandlers = new XHandlers(method);
1.160 + _number_of_locks = 0;
1.161 + _monitor_pairing_ok = method->has_balanced_monitors();
1.162 + _start = NULL;
1.163 +
1.164 + if (osr_bci == -1) {
1.165 + _requires_phi_function.clear();
1.166 + } else {
1.167 + // selective creation of phi functions is not possibel in osr-methods
1.168 + _requires_phi_function.set_range(0, method->max_locals());
1.169 + }
1.170 +
1.171 + assert(method->holder()->is_loaded() , "method holder must be loaded");
1.172 +
1.173 + // build graph if monitor pairing is ok
1.174 + if (create_graph && monitor_pairing_ok()) _start = build_graph(compilation, osr_bci);
1.175 +}
1.176 +
1.177 +
1.178 +int IRScope::max_stack() const {
1.179 + int my_max = method()->max_stack();
1.180 + int callee_max = 0;
1.181 + for (int i = 0; i < number_of_callees(); i++) {
1.182 + callee_max = MAX2(callee_max, callee_no(i)->max_stack());
1.183 + }
1.184 + return my_max + callee_max;
1.185 +}
1.186 +
1.187 +
1.188 +void IRScope::compute_lock_stack_size() {
1.189 + if (!InlineMethodsWithExceptionHandlers) {
1.190 + _lock_stack_size = 0;
1.191 + return;
1.192 + }
1.193 +
1.194 + // Figure out whether we have to preserve expression stack elements
1.195 + // for parent scopes, and if so, how many
1.196 + IRScope* cur_scope = this;
1.197 + while (cur_scope != NULL && !cur_scope->xhandlers()->has_handlers()) {
1.198 + cur_scope = cur_scope->caller();
1.199 + }
1.200 + _lock_stack_size = (cur_scope == NULL ? 0 :
1.201 + (cur_scope->caller_state() == NULL ? 0 :
1.202 + cur_scope->caller_state()->stack_size()));
1.203 +}
1.204 +
1.205 +int IRScope::top_scope_bci() const {
1.206 + assert(!is_top_scope(), "no correct answer for top scope possible");
1.207 + const IRScope* scope = this;
1.208 + while (!scope->caller()->is_top_scope()) {
1.209 + scope = scope->caller();
1.210 + }
1.211 + return scope->caller_bci();
1.212 +}
1.213 +
1.214 +
1.215 +
1.216 +// Implementation of CodeEmitInfo
1.217 +
1.218 +// Stack must be NON-null
1.219 +CodeEmitInfo::CodeEmitInfo(int bci, ValueStack* stack, XHandlers* exception_handlers)
1.220 + : _scope(stack->scope())
1.221 + , _bci(bci)
1.222 + , _scope_debug_info(NULL)
1.223 + , _oop_map(NULL)
1.224 + , _stack(stack)
1.225 + , _exception_handlers(exception_handlers)
1.226 + , _next(NULL)
1.227 + , _id(-1) {
1.228 + assert(_stack != NULL, "must be non null");
1.229 + assert(_bci == SynchronizationEntryBCI || Bytecodes::is_defined(scope()->method()->java_code_at_bci(_bci)), "make sure bci points at a real bytecode");
1.230 +}
1.231 +
1.232 +
1.233 +CodeEmitInfo::CodeEmitInfo(CodeEmitInfo* info, bool lock_stack_only)
1.234 + : _scope(info->_scope)
1.235 + , _exception_handlers(NULL)
1.236 + , _bci(info->_bci)
1.237 + , _scope_debug_info(NULL)
1.238 + , _oop_map(NULL) {
1.239 + if (lock_stack_only) {
1.240 + if (info->_stack != NULL) {
1.241 + _stack = info->_stack->copy_locks();
1.242 + } else {
1.243 + _stack = NULL;
1.244 + }
1.245 + } else {
1.246 + _stack = info->_stack;
1.247 + }
1.248 +
1.249 + // deep copy of exception handlers
1.250 + if (info->_exception_handlers != NULL) {
1.251 + _exception_handlers = new XHandlers(info->_exception_handlers);
1.252 + }
1.253 +}
1.254 +
1.255 +
1.256 +void CodeEmitInfo::record_debug_info(DebugInformationRecorder* recorder, int pc_offset) {
1.257 + // record the safepoint before recording the debug info for enclosing scopes
1.258 + recorder->add_safepoint(pc_offset, _oop_map->deep_copy());
1.259 + _scope_debug_info->record_debug_info(recorder, pc_offset);
1.260 + recorder->end_safepoint(pc_offset);
1.261 +}
1.262 +
1.263 +
1.264 +void CodeEmitInfo::add_register_oop(LIR_Opr opr) {
1.265 + assert(_oop_map != NULL, "oop map must already exist");
1.266 + assert(opr->is_single_cpu(), "should not call otherwise");
1.267 +
1.268 + int frame_size = frame_map()->framesize();
1.269 + int arg_count = frame_map()->oop_map_arg_count();
1.270 + VMReg name = frame_map()->regname(opr);
1.271 + _oop_map->set_oop(name);
1.272 +}
1.273 +
1.274 +
1.275 +
1.276 +
1.277 +// Implementation of IR
1.278 +
1.279 +IR::IR(Compilation* compilation, ciMethod* method, int osr_bci) :
1.280 + _locals_size(in_WordSize(-1))
1.281 + , _num_loops(0) {
1.282 + // initialize data structures
1.283 + ValueType::initialize();
1.284 + Instruction::initialize();
1.285 + BlockBegin::initialize();
1.286 + GraphBuilder::initialize();
1.287 + // setup IR fields
1.288 + _compilation = compilation;
1.289 + _top_scope = new IRScope(compilation, NULL, -1, method, osr_bci, true);
1.290 + _code = NULL;
1.291 +}
1.292 +
1.293 +
1.294 +void IR::optimize() {
1.295 + Optimizer opt(this);
1.296 + if (DoCEE) {
1.297 + opt.eliminate_conditional_expressions();
1.298 +#ifndef PRODUCT
1.299 + if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after CEE"); print(true); }
1.300 + if (PrintIR || PrintIR1 ) { tty->print_cr("IR after CEE"); print(false); }
1.301 +#endif
1.302 + }
1.303 + if (EliminateBlocks) {
1.304 + opt.eliminate_blocks();
1.305 +#ifndef PRODUCT
1.306 + if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after block elimination"); print(true); }
1.307 + if (PrintIR || PrintIR1 ) { tty->print_cr("IR after block elimination"); print(false); }
1.308 +#endif
1.309 + }
1.310 + if (EliminateNullChecks) {
1.311 + opt.eliminate_null_checks();
1.312 +#ifndef PRODUCT
1.313 + if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after null check elimination"); print(true); }
1.314 + if (PrintIR || PrintIR1 ) { tty->print_cr("IR after null check elimination"); print(false); }
1.315 +#endif
1.316 + }
1.317 +}
1.318 +
1.319 +
1.320 +static int sort_pairs(BlockPair** a, BlockPair** b) {
1.321 + if ((*a)->from() == (*b)->from()) {
1.322 + return (*a)->to()->block_id() - (*b)->to()->block_id();
1.323 + } else {
1.324 + return (*a)->from()->block_id() - (*b)->from()->block_id();
1.325 + }
1.326 +}
1.327 +
1.328 +
1.329 +class CriticalEdgeFinder: public BlockClosure {
1.330 + BlockPairList blocks;
1.331 + IR* _ir;
1.332 +
1.333 + public:
1.334 + CriticalEdgeFinder(IR* ir): _ir(ir) {}
1.335 + void block_do(BlockBegin* bb) {
1.336 + BlockEnd* be = bb->end();
1.337 + int nos = be->number_of_sux();
1.338 + if (nos >= 2) {
1.339 + for (int i = 0; i < nos; i++) {
1.340 + BlockBegin* sux = be->sux_at(i);
1.341 + if (sux->number_of_preds() >= 2) {
1.342 + blocks.append(new BlockPair(bb, sux));
1.343 + }
1.344 + }
1.345 + }
1.346 + }
1.347 +
1.348 + void split_edges() {
1.349 + BlockPair* last_pair = NULL;
1.350 + blocks.sort(sort_pairs);
1.351 + for (int i = 0; i < blocks.length(); i++) {
1.352 + BlockPair* pair = blocks.at(i);
1.353 + if (last_pair != NULL && pair->is_same(last_pair)) continue;
1.354 + BlockBegin* from = pair->from();
1.355 + BlockBegin* to = pair->to();
1.356 + BlockBegin* split = from->insert_block_between(to);
1.357 +#ifndef PRODUCT
1.358 + if ((PrintIR || PrintIR1) && Verbose) {
1.359 + tty->print_cr("Split critical edge B%d -> B%d (new block B%d)",
1.360 + from->block_id(), to->block_id(), split->block_id());
1.361 + }
1.362 +#endif
1.363 + last_pair = pair;
1.364 + }
1.365 + }
1.366 +};
1.367 +
1.368 +void IR::split_critical_edges() {
1.369 + CriticalEdgeFinder cef(this);
1.370 +
1.371 + iterate_preorder(&cef);
1.372 + cef.split_edges();
1.373 +}
1.374 +
1.375 +
1.376 +class UseCountComputer: public AllStatic {
1.377 + private:
1.378 + static void update_use_count(Value* n) {
1.379 + // Local instructions and Phis for expression stack values at the
1.380 + // start of basic blocks are not added to the instruction list
1.381 + if ((*n)->bci() == -99 && (*n)->as_Local() == NULL &&
1.382 + (*n)->as_Phi() == NULL) {
1.383 + assert(false, "a node was not appended to the graph");
1.384 + Compilation::current_compilation()->bailout("a node was not appended to the graph");
1.385 + }
1.386 + // use n's input if not visited before
1.387 + if (!(*n)->is_pinned() && !(*n)->has_uses()) {
1.388 + // note: a) if the instruction is pinned, it will be handled by compute_use_count
1.389 + // b) if the instruction has uses, it was touched before
1.390 + // => in both cases we don't need to update n's values
1.391 + uses_do(n);
1.392 + }
1.393 + // use n
1.394 + (*n)->_use_count++;
1.395 + }
1.396 +
1.397 + static Values* worklist;
1.398 + static int depth;
1.399 + enum {
1.400 + max_recurse_depth = 20
1.401 + };
1.402 +
1.403 + static void uses_do(Value* n) {
1.404 + depth++;
1.405 + if (depth > max_recurse_depth) {
1.406 + // don't allow the traversal to recurse too deeply
1.407 + worklist->push(*n);
1.408 + } else {
1.409 + (*n)->input_values_do(update_use_count);
1.410 + // special handling for some instructions
1.411 + if ((*n)->as_BlockEnd() != NULL) {
1.412 + // note on BlockEnd:
1.413 + // must 'use' the stack only if the method doesn't
1.414 + // terminate, however, in those cases stack is empty
1.415 + (*n)->state_values_do(update_use_count);
1.416 + }
1.417 + }
1.418 + depth--;
1.419 + }
1.420 +
1.421 + static void basic_compute_use_count(BlockBegin* b) {
1.422 + depth = 0;
1.423 + // process all pinned nodes as the roots of expression trees
1.424 + for (Instruction* n = b; n != NULL; n = n->next()) {
1.425 + if (n->is_pinned()) uses_do(&n);
1.426 + }
1.427 + assert(depth == 0, "should have counted back down");
1.428 +
1.429 + // now process any unpinned nodes which recursed too deeply
1.430 + while (worklist->length() > 0) {
1.431 + Value t = worklist->pop();
1.432 + if (!t->is_pinned()) {
1.433 + // compute the use count
1.434 + uses_do(&t);
1.435 +
1.436 + // pin the instruction so that LIRGenerator doesn't recurse
1.437 + // too deeply during it's evaluation.
1.438 + t->pin();
1.439 + }
1.440 + }
1.441 + assert(depth == 0, "should have counted back down");
1.442 + }
1.443 +
1.444 + public:
1.445 + static void compute(BlockList* blocks) {
1.446 + worklist = new Values();
1.447 + blocks->blocks_do(basic_compute_use_count);
1.448 + worklist = NULL;
1.449 + }
1.450 +};
1.451 +
1.452 +
1.453 +Values* UseCountComputer::worklist = NULL;
1.454 +int UseCountComputer::depth = 0;
1.455 +
1.456 +// helper macro for short definition of trace-output inside code
1.457 +#ifndef PRODUCT
1.458 + #define TRACE_LINEAR_SCAN(level, code) \
1.459 + if (TraceLinearScanLevel >= level) { \
1.460 + code; \
1.461 + }
1.462 +#else
1.463 + #define TRACE_LINEAR_SCAN(level, code)
1.464 +#endif
1.465 +
1.466 +class ComputeLinearScanOrder : public StackObj {
1.467 + private:
1.468 + int _max_block_id; // the highest block_id of a block
1.469 + int _num_blocks; // total number of blocks (smaller than _max_block_id)
1.470 + int _num_loops; // total number of loops
1.471 + bool _iterative_dominators;// method requires iterative computation of dominatiors
1.472 +
1.473 + BlockList* _linear_scan_order; // the resulting list of blocks in correct order
1.474 +
1.475 + BitMap _visited_blocks; // used for recursive processing of blocks
1.476 + BitMap _active_blocks; // used for recursive processing of blocks
1.477 + BitMap _dominator_blocks; // temproary BitMap used for computation of dominator
1.478 + intArray _forward_branches; // number of incoming forward branches for each block
1.479 + BlockList _loop_end_blocks; // list of all loop end blocks collected during count_edges
1.480 + BitMap2D _loop_map; // two-dimensional bit set: a bit is set if a block is contained in a loop
1.481 + BlockList _work_list; // temporary list (used in mark_loops and compute_order)
1.482 +
1.483 + // accessors for _visited_blocks and _active_blocks
1.484 + void init_visited() { _active_blocks.clear(); _visited_blocks.clear(); }
1.485 + bool is_visited(BlockBegin* b) const { return _visited_blocks.at(b->block_id()); }
1.486 + bool is_active(BlockBegin* b) const { return _active_blocks.at(b->block_id()); }
1.487 + void set_visited(BlockBegin* b) { assert(!is_visited(b), "already set"); _visited_blocks.set_bit(b->block_id()); }
1.488 + void set_active(BlockBegin* b) { assert(!is_active(b), "already set"); _active_blocks.set_bit(b->block_id()); }
1.489 + void clear_active(BlockBegin* b) { assert(is_active(b), "not already"); _active_blocks.clear_bit(b->block_id()); }
1.490 +
1.491 + // accessors for _forward_branches
1.492 + void inc_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) + 1); }
1.493 + int dec_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) - 1); return _forward_branches.at(b->block_id()); }
1.494 +
1.495 + // accessors for _loop_map
1.496 + bool is_block_in_loop (int loop_idx, BlockBegin* b) const { return _loop_map.at(loop_idx, b->block_id()); }
1.497 + void set_block_in_loop (int loop_idx, BlockBegin* b) { _loop_map.set_bit(loop_idx, b->block_id()); }
1.498 + void clear_block_in_loop(int loop_idx, int block_id) { _loop_map.clear_bit(loop_idx, block_id); }
1.499 +
1.500 + // count edges between blocks
1.501 + void count_edges(BlockBegin* cur, BlockBegin* parent);
1.502 +
1.503 + // loop detection
1.504 + void mark_loops();
1.505 + void clear_non_natural_loops(BlockBegin* start_block);
1.506 + void assign_loop_depth(BlockBegin* start_block);
1.507 +
1.508 + // computation of final block order
1.509 + BlockBegin* common_dominator(BlockBegin* a, BlockBegin* b);
1.510 + void compute_dominator(BlockBegin* cur, BlockBegin* parent);
1.511 + int compute_weight(BlockBegin* cur);
1.512 + bool ready_for_processing(BlockBegin* cur);
1.513 + void sort_into_work_list(BlockBegin* b);
1.514 + void append_block(BlockBegin* cur);
1.515 + void compute_order(BlockBegin* start_block);
1.516 +
1.517 + // fixup of dominators for non-natural loops
1.518 + bool compute_dominators_iter();
1.519 + void compute_dominators();
1.520 +
1.521 + // debug functions
1.522 + NOT_PRODUCT(void print_blocks();)
1.523 + DEBUG_ONLY(void verify();)
1.524 +
1.525 + public:
1.526 + ComputeLinearScanOrder(BlockBegin* start_block);
1.527 +
1.528 + // accessors for final result
1.529 + BlockList* linear_scan_order() const { return _linear_scan_order; }
1.530 + int num_loops() const { return _num_loops; }
1.531 +};
1.532 +
1.533 +
1.534 +ComputeLinearScanOrder::ComputeLinearScanOrder(BlockBegin* start_block) :
1.535 + _max_block_id(BlockBegin::number_of_blocks()),
1.536 + _num_blocks(0),
1.537 + _num_loops(0),
1.538 + _iterative_dominators(false),
1.539 + _visited_blocks(_max_block_id),
1.540 + _active_blocks(_max_block_id),
1.541 + _dominator_blocks(_max_block_id),
1.542 + _forward_branches(_max_block_id, 0),
1.543 + _loop_end_blocks(8),
1.544 + _work_list(8),
1.545 + _linear_scan_order(NULL), // initialized later with correct size
1.546 + _loop_map(0, 0) // initialized later with correct size
1.547 +{
1.548 + TRACE_LINEAR_SCAN(2, "***** computing linear-scan block order");
1.549 +
1.550 + init_visited();
1.551 + count_edges(start_block, NULL);
1.552 +
1.553 + if (_num_loops > 0) {
1.554 + mark_loops();
1.555 + clear_non_natural_loops(start_block);
1.556 + assign_loop_depth(start_block);
1.557 + }
1.558 +
1.559 + compute_order(start_block);
1.560 + compute_dominators();
1.561 +
1.562 + NOT_PRODUCT(print_blocks());
1.563 + DEBUG_ONLY(verify());
1.564 +}
1.565 +
1.566 +
1.567 +// Traverse the CFG:
1.568 +// * count total number of blocks
1.569 +// * count all incoming edges and backward incoming edges
1.570 +// * number loop header blocks
1.571 +// * create a list with all loop end blocks
1.572 +void ComputeLinearScanOrder::count_edges(BlockBegin* cur, BlockBegin* parent) {
1.573 + TRACE_LINEAR_SCAN(3, tty->print_cr("Enter count_edges for block B%d coming from B%d", cur->block_id(), parent != NULL ? parent->block_id() : -1));
1.574 + assert(cur->dominator() == NULL, "dominator already initialized");
1.575 +
1.576 + if (is_active(cur)) {
1.577 + TRACE_LINEAR_SCAN(3, tty->print_cr("backward branch"));
1.578 + assert(is_visited(cur), "block must be visisted when block is active");
1.579 + assert(parent != NULL, "must have parent");
1.580 + assert(parent->number_of_sux() == 1, "loop end blocks must have one successor (critical edges are split)");
1.581 +
1.582 + cur->set(BlockBegin::linear_scan_loop_header_flag);
1.583 + cur->set(BlockBegin::backward_branch_target_flag);
1.584 +
1.585 + parent->set(BlockBegin::linear_scan_loop_end_flag);
1.586 + _loop_end_blocks.append(parent);
1.587 + return;
1.588 + }
1.589 +
1.590 + // increment number of incoming forward branches
1.591 + inc_forward_branches(cur);
1.592 +
1.593 + if (is_visited(cur)) {
1.594 + TRACE_LINEAR_SCAN(3, tty->print_cr("block already visited"));
1.595 + return;
1.596 + }
1.597 +
1.598 + _num_blocks++;
1.599 + set_visited(cur);
1.600 + set_active(cur);
1.601 +
1.602 + // recursive call for all successors
1.603 + int i;
1.604 + for (i = cur->number_of_sux() - 1; i >= 0; i--) {
1.605 + count_edges(cur->sux_at(i), cur);
1.606 + }
1.607 + for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
1.608 + count_edges(cur->exception_handler_at(i), cur);
1.609 + }
1.610 +
1.611 + clear_active(cur);
1.612 +
1.613 + // Each loop has a unique number.
1.614 + // When multiple loops are nested, assign_loop_depth assumes that the
1.615 + // innermost loop has the lowest number. This is guaranteed by setting
1.616 + // the loop number after the recursive calls for the successors above
1.617 + // have returned.
1.618 + if (cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
1.619 + assert(cur->loop_index() == -1, "cannot set loop-index twice");
1.620 + TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops));
1.621 +
1.622 + cur->set_loop_index(_num_loops);
1.623 + _num_loops++;
1.624 + }
1.625 +
1.626 + TRACE_LINEAR_SCAN(3, tty->print_cr("Finished count_edges for block B%d", cur->block_id()));
1.627 +}
1.628 +
1.629 +
1.630 +void ComputeLinearScanOrder::mark_loops() {
1.631 + TRACE_LINEAR_SCAN(3, tty->print_cr("----- marking loops"));
1.632 +
1.633 + _loop_map = BitMap2D(_num_loops, _max_block_id);
1.634 + _loop_map.clear();
1.635 +
1.636 + for (int i = _loop_end_blocks.length() - 1; i >= 0; i--) {
1.637 + BlockBegin* loop_end = _loop_end_blocks.at(i);
1.638 + BlockBegin* loop_start = loop_end->sux_at(0);
1.639 + int loop_idx = loop_start->loop_index();
1.640 +
1.641 + TRACE_LINEAR_SCAN(3, tty->print_cr("Processing loop from B%d to B%d (loop %d):", loop_start->block_id(), loop_end->block_id(), loop_idx));
1.642 + assert(loop_end->is_set(BlockBegin::linear_scan_loop_end_flag), "loop end flag must be set");
1.643 + assert(loop_end->number_of_sux() == 1, "incorrect number of successors");
1.644 + assert(loop_start->is_set(BlockBegin::linear_scan_loop_header_flag), "loop header flag must be set");
1.645 + assert(loop_idx >= 0 && loop_idx < _num_loops, "loop index not set");
1.646 + assert(_work_list.is_empty(), "work list must be empty before processing");
1.647 +
1.648 + // add the end-block of the loop to the working list
1.649 + _work_list.push(loop_end);
1.650 + set_block_in_loop(loop_idx, loop_end);
1.651 + do {
1.652 + BlockBegin* cur = _work_list.pop();
1.653 +
1.654 + TRACE_LINEAR_SCAN(3, tty->print_cr(" processing B%d", cur->block_id()));
1.655 + assert(is_block_in_loop(loop_idx, cur), "bit in loop map must be set when block is in work list");
1.656 +
1.657 + // recursive processing of all predecessors ends when start block of loop is reached
1.658 + if (cur != loop_start && !cur->is_set(BlockBegin::osr_entry_flag)) {
1.659 + for (int j = cur->number_of_preds() - 1; j >= 0; j--) {
1.660 + BlockBegin* pred = cur->pred_at(j);
1.661 +
1.662 + if (!is_block_in_loop(loop_idx, pred) /*&& !pred->is_set(BlockBeginosr_entry_flag)*/) {
1.663 + // this predecessor has not been processed yet, so add it to work list
1.664 + TRACE_LINEAR_SCAN(3, tty->print_cr(" pushing B%d", pred->block_id()));
1.665 + _work_list.push(pred);
1.666 + set_block_in_loop(loop_idx, pred);
1.667 + }
1.668 + }
1.669 + }
1.670 + } while (!_work_list.is_empty());
1.671 + }
1.672 +}
1.673 +
1.674 +
1.675 +// check for non-natural loops (loops where the loop header does not dominate
1.676 +// all other loop blocks = loops with mulitple entries).
1.677 +// such loops are ignored
1.678 +void ComputeLinearScanOrder::clear_non_natural_loops(BlockBegin* start_block) {
1.679 + for (int i = _num_loops - 1; i >= 0; i--) {
1.680 + if (is_block_in_loop(i, start_block)) {
1.681 + // loop i contains the entry block of the method
1.682 + // -> this is not a natural loop, so ignore it
1.683 + TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i));
1.684 +
1.685 + for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) {
1.686 + clear_block_in_loop(i, block_id);
1.687 + }
1.688 + _iterative_dominators = true;
1.689 + }
1.690 + }
1.691 +}
1.692 +
1.693 +void ComputeLinearScanOrder::assign_loop_depth(BlockBegin* start_block) {
1.694 + TRACE_LINEAR_SCAN(3, "----- computing loop-depth and weight");
1.695 + init_visited();
1.696 +
1.697 + assert(_work_list.is_empty(), "work list must be empty before processing");
1.698 + _work_list.append(start_block);
1.699 +
1.700 + do {
1.701 + BlockBegin* cur = _work_list.pop();
1.702 +
1.703 + if (!is_visited(cur)) {
1.704 + set_visited(cur);
1.705 + TRACE_LINEAR_SCAN(4, tty->print_cr("Computing loop depth for block B%d", cur->block_id()));
1.706 +
1.707 + // compute loop-depth and loop-index for the block
1.708 + assert(cur->loop_depth() == 0, "cannot set loop-depth twice");
1.709 + int i;
1.710 + int loop_depth = 0;
1.711 + int min_loop_idx = -1;
1.712 + for (i = _num_loops - 1; i >= 0; i--) {
1.713 + if (is_block_in_loop(i, cur)) {
1.714 + loop_depth++;
1.715 + min_loop_idx = i;
1.716 + }
1.717 + }
1.718 + cur->set_loop_depth(loop_depth);
1.719 + cur->set_loop_index(min_loop_idx);
1.720 +
1.721 + // append all unvisited successors to work list
1.722 + for (i = cur->number_of_sux() - 1; i >= 0; i--) {
1.723 + _work_list.append(cur->sux_at(i));
1.724 + }
1.725 + for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
1.726 + _work_list.append(cur->exception_handler_at(i));
1.727 + }
1.728 + }
1.729 + } while (!_work_list.is_empty());
1.730 +}
1.731 +
1.732 +
1.733 +BlockBegin* ComputeLinearScanOrder::common_dominator(BlockBegin* a, BlockBegin* b) {
1.734 + assert(a != NULL && b != NULL, "must have input blocks");
1.735 +
1.736 + _dominator_blocks.clear();
1.737 + while (a != NULL) {
1.738 + _dominator_blocks.set_bit(a->block_id());
1.739 + assert(a->dominator() != NULL || a == _linear_scan_order->at(0), "dominator must be initialized");
1.740 + a = a->dominator();
1.741 + }
1.742 + while (b != NULL && !_dominator_blocks.at(b->block_id())) {
1.743 + assert(b->dominator() != NULL || b == _linear_scan_order->at(0), "dominator must be initialized");
1.744 + b = b->dominator();
1.745 + }
1.746 +
1.747 + assert(b != NULL, "could not find dominator");
1.748 + return b;
1.749 +}
1.750 +
1.751 +void ComputeLinearScanOrder::compute_dominator(BlockBegin* cur, BlockBegin* parent) {
1.752 + if (cur->dominator() == NULL) {
1.753 + TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: initializing dominator of B%d to B%d", cur->block_id(), parent->block_id()));
1.754 + cur->set_dominator(parent);
1.755 +
1.756 + } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) {
1.757 + TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: computing dominator of B%d: common dominator of B%d and B%d is B%d", cur->block_id(), parent->block_id(), cur->dominator()->block_id(), common_dominator(cur->dominator(), parent)->block_id()));
1.758 + assert(cur->number_of_preds() > 1, "");
1.759 + cur->set_dominator(common_dominator(cur->dominator(), parent));
1.760 + }
1.761 +}
1.762 +
1.763 +
1.764 +int ComputeLinearScanOrder::compute_weight(BlockBegin* cur) {
1.765 + BlockBegin* single_sux = NULL;
1.766 + if (cur->number_of_sux() == 1) {
1.767 + single_sux = cur->sux_at(0);
1.768 + }
1.769 +
1.770 + // limit loop-depth to 15 bit (only for security reason, it will never be so big)
1.771 + int weight = (cur->loop_depth() & 0x7FFF) << 16;
1.772 +
1.773 + // general macro for short definition of weight flags
1.774 + // the first instance of INC_WEIGHT_IF has the highest priority
1.775 + int cur_bit = 15;
1.776 + #define INC_WEIGHT_IF(condition) if ((condition)) { weight |= (1 << cur_bit); } cur_bit--;
1.777 +
1.778 + // this is necessery for the (very rare) case that two successing blocks have
1.779 + // the same loop depth, but a different loop index (can happen for endless loops
1.780 + // with exception handlers)
1.781 + INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_header_flag));
1.782 +
1.783 + // loop end blocks (blocks that end with a backward branch) are added
1.784 + // after all other blocks of the loop.
1.785 + INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_end_flag));
1.786 +
1.787 + // critical edge split blocks are prefered because than they have a bigger
1.788 + // proability to be completely empty
1.789 + INC_WEIGHT_IF(cur->is_set(BlockBegin::critical_edge_split_flag));
1.790 +
1.791 + // exceptions should not be thrown in normal control flow, so these blocks
1.792 + // are added as late as possible
1.793 + INC_WEIGHT_IF(cur->end()->as_Throw() == NULL && (single_sux == NULL || single_sux->end()->as_Throw() == NULL));
1.794 + INC_WEIGHT_IF(cur->end()->as_Return() == NULL && (single_sux == NULL || single_sux->end()->as_Return() == NULL));
1.795 +
1.796 + // exceptions handlers are added as late as possible
1.797 + INC_WEIGHT_IF(!cur->is_set(BlockBegin::exception_entry_flag));
1.798 +
1.799 + // guarantee that weight is > 0
1.800 + weight |= 1;
1.801 +
1.802 + #undef INC_WEIGHT_IF
1.803 + assert(cur_bit >= 0, "too many flags");
1.804 + assert(weight > 0, "weight cannot become negative");
1.805 +
1.806 + return weight;
1.807 +}
1.808 +
1.809 +bool ComputeLinearScanOrder::ready_for_processing(BlockBegin* cur) {
1.810 + // Discount the edge just traveled.
1.811 + // When the number drops to zero, all forward branches were processed
1.812 + if (dec_forward_branches(cur) != 0) {
1.813 + return false;
1.814 + }
1.815 +
1.816 + assert(_linear_scan_order->index_of(cur) == -1, "block already processed (block can be ready only once)");
1.817 + assert(_work_list.index_of(cur) == -1, "block already in work-list (block can be ready only once)");
1.818 + return true;
1.819 +}
1.820 +
1.821 +void ComputeLinearScanOrder::sort_into_work_list(BlockBegin* cur) {
1.822 + assert(_work_list.index_of(cur) == -1, "block already in work list");
1.823 +
1.824 + int cur_weight = compute_weight(cur);
1.825 +
1.826 + // the linear_scan_number is used to cache the weight of a block
1.827 + cur->set_linear_scan_number(cur_weight);
1.828 +
1.829 +#ifndef PRODUCT
1.830 + if (StressLinearScan) {
1.831 + _work_list.insert_before(0, cur);
1.832 + return;
1.833 + }
1.834 +#endif
1.835 +
1.836 + _work_list.append(NULL); // provide space for new element
1.837 +
1.838 + int insert_idx = _work_list.length() - 1;
1.839 + while (insert_idx > 0 && _work_list.at(insert_idx - 1)->linear_scan_number() > cur_weight) {
1.840 + _work_list.at_put(insert_idx, _work_list.at(insert_idx - 1));
1.841 + insert_idx--;
1.842 + }
1.843 + _work_list.at_put(insert_idx, cur);
1.844 +
1.845 + TRACE_LINEAR_SCAN(3, tty->print_cr("Sorted B%d into worklist. new worklist:", cur->block_id()));
1.846 + TRACE_LINEAR_SCAN(3, for (int i = 0; i < _work_list.length(); i++) tty->print_cr("%8d B%2d weight:%6x", i, _work_list.at(i)->block_id(), _work_list.at(i)->linear_scan_number()));
1.847 +
1.848 +#ifdef ASSERT
1.849 + for (int i = 0; i < _work_list.length(); i++) {
1.850 + assert(_work_list.at(i)->linear_scan_number() > 0, "weight not set");
1.851 + assert(i == 0 || _work_list.at(i - 1)->linear_scan_number() <= _work_list.at(i)->linear_scan_number(), "incorrect order in worklist");
1.852 + }
1.853 +#endif
1.854 +}
1.855 +
1.856 +void ComputeLinearScanOrder::append_block(BlockBegin* cur) {
1.857 + TRACE_LINEAR_SCAN(3, tty->print_cr("appending block B%d (weight 0x%6x) to linear-scan order", cur->block_id(), cur->linear_scan_number()));
1.858 + assert(_linear_scan_order->index_of(cur) == -1, "cannot add the same block twice");
1.859 +
1.860 + // currently, the linear scan order and code emit order are equal.
1.861 + // therefore the linear_scan_number and the weight of a block must also
1.862 + // be equal.
1.863 + cur->set_linear_scan_number(_linear_scan_order->length());
1.864 + _linear_scan_order->append(cur);
1.865 +}
1.866 +
1.867 +void ComputeLinearScanOrder::compute_order(BlockBegin* start_block) {
1.868 + TRACE_LINEAR_SCAN(3, "----- computing final block order");
1.869 +
1.870 + // the start block is always the first block in the linear scan order
1.871 + _linear_scan_order = new BlockList(_num_blocks);
1.872 + append_block(start_block);
1.873 +
1.874 + assert(start_block->end()->as_Base() != NULL, "start block must end with Base-instruction");
1.875 + BlockBegin* std_entry = ((Base*)start_block->end())->std_entry();
1.876 + BlockBegin* osr_entry = ((Base*)start_block->end())->osr_entry();
1.877 +
1.878 + BlockBegin* sux_of_osr_entry = NULL;
1.879 + if (osr_entry != NULL) {
1.880 + // special handling for osr entry:
1.881 + // ignore the edge between the osr entry and its successor for processing
1.882 + // the osr entry block is added manually below
1.883 + assert(osr_entry->number_of_sux() == 1, "osr entry must have exactly one successor");
1.884 + assert(osr_entry->sux_at(0)->number_of_preds() >= 2, "sucessor of osr entry must have two predecessors (otherwise it is not present in normal control flow");
1.885 +
1.886 + sux_of_osr_entry = osr_entry->sux_at(0);
1.887 + dec_forward_branches(sux_of_osr_entry);
1.888 +
1.889 + compute_dominator(osr_entry, start_block);
1.890 + _iterative_dominators = true;
1.891 + }
1.892 + compute_dominator(std_entry, start_block);
1.893 +
1.894 + // start processing with standard entry block
1.895 + assert(_work_list.is_empty(), "list must be empty before processing");
1.896 +
1.897 + if (ready_for_processing(std_entry)) {
1.898 + sort_into_work_list(std_entry);
1.899 + } else {
1.900 + assert(false, "the std_entry must be ready for processing (otherwise, the method has no start block)");
1.901 + }
1.902 +
1.903 + do {
1.904 + BlockBegin* cur = _work_list.pop();
1.905 +
1.906 + if (cur == sux_of_osr_entry) {
1.907 + // the osr entry block is ignored in normal processing, it is never added to the
1.908 + // work list. Instead, it is added as late as possible manually here.
1.909 + append_block(osr_entry);
1.910 + compute_dominator(cur, osr_entry);
1.911 + }
1.912 + append_block(cur);
1.913 +
1.914 + int i;
1.915 + int num_sux = cur->number_of_sux();
1.916 + // changed loop order to get "intuitive" order of if- and else-blocks
1.917 + for (i = 0; i < num_sux; i++) {
1.918 + BlockBegin* sux = cur->sux_at(i);
1.919 + compute_dominator(sux, cur);
1.920 + if (ready_for_processing(sux)) {
1.921 + sort_into_work_list(sux);
1.922 + }
1.923 + }
1.924 + num_sux = cur->number_of_exception_handlers();
1.925 + for (i = 0; i < num_sux; i++) {
1.926 + BlockBegin* sux = cur->exception_handler_at(i);
1.927 + compute_dominator(sux, cur);
1.928 + if (ready_for_processing(sux)) {
1.929 + sort_into_work_list(sux);
1.930 + }
1.931 + }
1.932 + } while (_work_list.length() > 0);
1.933 +}
1.934 +
1.935 +
1.936 +bool ComputeLinearScanOrder::compute_dominators_iter() {
1.937 + bool changed = false;
1.938 + int num_blocks = _linear_scan_order->length();
1.939 +
1.940 + assert(_linear_scan_order->at(0)->dominator() == NULL, "must not have dominator");
1.941 + assert(_linear_scan_order->at(0)->number_of_preds() == 0, "must not have predecessors");
1.942 + for (int i = 1; i < num_blocks; i++) {
1.943 + BlockBegin* block = _linear_scan_order->at(i);
1.944 +
1.945 + BlockBegin* dominator = block->pred_at(0);
1.946 + int num_preds = block->number_of_preds();
1.947 + for (int i = 1; i < num_preds; i++) {
1.948 + dominator = common_dominator(dominator, block->pred_at(i));
1.949 + }
1.950 +
1.951 + if (dominator != block->dominator()) {
1.952 + TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: updating dominator of B%d from B%d to B%d", block->block_id(), block->dominator()->block_id(), dominator->block_id()));
1.953 +
1.954 + block->set_dominator(dominator);
1.955 + changed = true;
1.956 + }
1.957 + }
1.958 + return changed;
1.959 +}
1.960 +
1.961 +void ComputeLinearScanOrder::compute_dominators() {
1.962 + TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing dominators (iterative computation reqired: %d)", _iterative_dominators));
1.963 +
1.964 + // iterative computation of dominators is only required for methods with non-natural loops
1.965 + // and OSR-methods. For all other methods, the dominators computed when generating the
1.966 + // linear scan block order are correct.
1.967 + if (_iterative_dominators) {
1.968 + do {
1.969 + TRACE_LINEAR_SCAN(1, tty->print_cr("DOM: next iteration of fix-point calculation"));
1.970 + } while (compute_dominators_iter());
1.971 + }
1.972 +
1.973 + // check that dominators are correct
1.974 + assert(!compute_dominators_iter(), "fix point not reached");
1.975 +}
1.976 +
1.977 +
1.978 +#ifndef PRODUCT
1.979 +void ComputeLinearScanOrder::print_blocks() {
1.980 + if (TraceLinearScanLevel >= 2) {
1.981 + tty->print_cr("----- loop information:");
1.982 + for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1.983 + BlockBegin* cur = _linear_scan_order->at(block_idx);
1.984 +
1.985 + tty->print("%4d: B%2d: ", cur->linear_scan_number(), cur->block_id());
1.986 + for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1.987 + tty->print ("%d ", is_block_in_loop(loop_idx, cur));
1.988 + }
1.989 + tty->print_cr(" -> loop_index: %2d, loop_depth: %2d", cur->loop_index(), cur->loop_depth());
1.990 + }
1.991 + }
1.992 +
1.993 + if (TraceLinearScanLevel >= 1) {
1.994 + tty->print_cr("----- linear-scan block order:");
1.995 + for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1.996 + BlockBegin* cur = _linear_scan_order->at(block_idx);
1.997 + tty->print("%4d: B%2d loop: %2d depth: %2d", cur->linear_scan_number(), cur->block_id(), cur->loop_index(), cur->loop_depth());
1.998 +
1.999 + tty->print(cur->is_set(BlockBegin::exception_entry_flag) ? " ex" : " ");
1.1000 + tty->print(cur->is_set(BlockBegin::critical_edge_split_flag) ? " ce" : " ");
1.1001 + tty->print(cur->is_set(BlockBegin::linear_scan_loop_header_flag) ? " lh" : " ");
1.1002 + tty->print(cur->is_set(BlockBegin::linear_scan_loop_end_flag) ? " le" : " ");
1.1003 +
1.1004 + if (cur->dominator() != NULL) {
1.1005 + tty->print(" dom: B%d ", cur->dominator()->block_id());
1.1006 + } else {
1.1007 + tty->print(" dom: NULL ");
1.1008 + }
1.1009 +
1.1010 + if (cur->number_of_preds() > 0) {
1.1011 + tty->print(" preds: ");
1.1012 + for (int j = 0; j < cur->number_of_preds(); j++) {
1.1013 + BlockBegin* pred = cur->pred_at(j);
1.1014 + tty->print("B%d ", pred->block_id());
1.1015 + }
1.1016 + }
1.1017 + if (cur->number_of_sux() > 0) {
1.1018 + tty->print(" sux: ");
1.1019 + for (int j = 0; j < cur->number_of_sux(); j++) {
1.1020 + BlockBegin* sux = cur->sux_at(j);
1.1021 + tty->print("B%d ", sux->block_id());
1.1022 + }
1.1023 + }
1.1024 + if (cur->number_of_exception_handlers() > 0) {
1.1025 + tty->print(" ex: ");
1.1026 + for (int j = 0; j < cur->number_of_exception_handlers(); j++) {
1.1027 + BlockBegin* ex = cur->exception_handler_at(j);
1.1028 + tty->print("B%d ", ex->block_id());
1.1029 + }
1.1030 + }
1.1031 + tty->cr();
1.1032 + }
1.1033 + }
1.1034 +}
1.1035 +#endif
1.1036 +
1.1037 +#ifdef ASSERT
1.1038 +void ComputeLinearScanOrder::verify() {
1.1039 + assert(_linear_scan_order->length() == _num_blocks, "wrong number of blocks in list");
1.1040 +
1.1041 + if (StressLinearScan) {
1.1042 + // blocks are scrambled when StressLinearScan is used
1.1043 + return;
1.1044 + }
1.1045 +
1.1046 + // check that all successors of a block have a higher linear-scan-number
1.1047 + // and that all predecessors of a block have a lower linear-scan-number
1.1048 + // (only backward branches of loops are ignored)
1.1049 + int i;
1.1050 + for (i = 0; i < _linear_scan_order->length(); i++) {
1.1051 + BlockBegin* cur = _linear_scan_order->at(i);
1.1052 +
1.1053 + assert(cur->linear_scan_number() == i, "incorrect linear_scan_number");
1.1054 + assert(cur->linear_scan_number() >= 0 && cur->linear_scan_number() == _linear_scan_order->index_of(cur), "incorrect linear_scan_number");
1.1055 +
1.1056 + int j;
1.1057 + for (j = cur->number_of_sux() - 1; j >= 0; j--) {
1.1058 + BlockBegin* sux = cur->sux_at(j);
1.1059 +
1.1060 + assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->index_of(sux), "incorrect linear_scan_number");
1.1061 + if (!cur->is_set(BlockBegin::linear_scan_loop_end_flag)) {
1.1062 + assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order");
1.1063 + }
1.1064 + if (cur->loop_depth() == sux->loop_depth()) {
1.1065 + assert(cur->loop_index() == sux->loop_index() || sux->is_set(BlockBegin::linear_scan_loop_header_flag), "successing blocks with same loop depth must have same loop index");
1.1066 + }
1.1067 + }
1.1068 +
1.1069 + for (j = cur->number_of_preds() - 1; j >= 0; j--) {
1.1070 + BlockBegin* pred = cur->pred_at(j);
1.1071 +
1.1072 + assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->index_of(pred), "incorrect linear_scan_number");
1.1073 + if (!cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
1.1074 + assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order");
1.1075 + }
1.1076 + if (cur->loop_depth() == pred->loop_depth()) {
1.1077 + assert(cur->loop_index() == pred->loop_index() || cur->is_set(BlockBegin::linear_scan_loop_header_flag), "successing blocks with same loop depth must have same loop index");
1.1078 + }
1.1079 +
1.1080 + assert(cur->dominator()->linear_scan_number() <= cur->pred_at(j)->linear_scan_number(), "dominator must be before predecessors");
1.1081 + }
1.1082 +
1.1083 + // check dominator
1.1084 + if (i == 0) {
1.1085 + assert(cur->dominator() == NULL, "first block has no dominator");
1.1086 + } else {
1.1087 + assert(cur->dominator() != NULL, "all but first block must have dominator");
1.1088 + }
1.1089 + assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0), "Single predecessor must also be dominator");
1.1090 + }
1.1091 +
1.1092 + // check that all loops are continuous
1.1093 + for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1.1094 + int block_idx = 0;
1.1095 + assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "the first block must not be present in any loop");
1.1096 +
1.1097 + // skip blocks before the loop
1.1098 + while (block_idx < _num_blocks && !is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1.1099 + block_idx++;
1.1100 + }
1.1101 + // skip blocks of loop
1.1102 + while (block_idx < _num_blocks && is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1.1103 + block_idx++;
1.1104 + }
1.1105 + // after the first non-loop block, there must not be another loop-block
1.1106 + while (block_idx < _num_blocks) {
1.1107 + assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "loop not continuous in linear-scan order");
1.1108 + block_idx++;
1.1109 + }
1.1110 + }
1.1111 +}
1.1112 +#endif
1.1113 +
1.1114 +
1.1115 +void IR::compute_code() {
1.1116 + assert(is_valid(), "IR must be valid");
1.1117 +
1.1118 + ComputeLinearScanOrder compute_order(start());
1.1119 + _num_loops = compute_order.num_loops();
1.1120 + _code = compute_order.linear_scan_order();
1.1121 +}
1.1122 +
1.1123 +
1.1124 +void IR::compute_use_counts() {
1.1125 + // make sure all values coming out of this block get evaluated.
1.1126 + int num_blocks = _code->length();
1.1127 + for (int i = 0; i < num_blocks; i++) {
1.1128 + _code->at(i)->end()->state()->pin_stack_for_linear_scan();
1.1129 + }
1.1130 +
1.1131 + // compute use counts
1.1132 + UseCountComputer::compute(_code);
1.1133 +}
1.1134 +
1.1135 +
1.1136 +void IR::iterate_preorder(BlockClosure* closure) {
1.1137 + assert(is_valid(), "IR must be valid");
1.1138 + start()->iterate_preorder(closure);
1.1139 +}
1.1140 +
1.1141 +
1.1142 +void IR::iterate_postorder(BlockClosure* closure) {
1.1143 + assert(is_valid(), "IR must be valid");
1.1144 + start()->iterate_postorder(closure);
1.1145 +}
1.1146 +
1.1147 +void IR::iterate_linear_scan_order(BlockClosure* closure) {
1.1148 + linear_scan_order()->iterate_forward(closure);
1.1149 +}
1.1150 +
1.1151 +
1.1152 +#ifndef PRODUCT
1.1153 +class BlockPrinter: public BlockClosure {
1.1154 + private:
1.1155 + InstructionPrinter* _ip;
1.1156 + bool _cfg_only;
1.1157 + bool _live_only;
1.1158 +
1.1159 + public:
1.1160 + BlockPrinter(InstructionPrinter* ip, bool cfg_only, bool live_only = false) {
1.1161 + _ip = ip;
1.1162 + _cfg_only = cfg_only;
1.1163 + _live_only = live_only;
1.1164 + }
1.1165 +
1.1166 + virtual void block_do(BlockBegin* block) {
1.1167 + if (_cfg_only) {
1.1168 + _ip->print_instr(block); tty->cr();
1.1169 + } else {
1.1170 + block->print_block(*_ip, _live_only);
1.1171 + }
1.1172 + }
1.1173 +};
1.1174 +
1.1175 +
1.1176 +void IR::print(BlockBegin* start, bool cfg_only, bool live_only) {
1.1177 + ttyLocker ttyl;
1.1178 + InstructionPrinter ip(!cfg_only);
1.1179 + BlockPrinter bp(&ip, cfg_only, live_only);
1.1180 + start->iterate_preorder(&bp);
1.1181 + tty->cr();
1.1182 +}
1.1183 +
1.1184 +void IR::print(bool cfg_only, bool live_only) {
1.1185 + if (is_valid()) {
1.1186 + print(start(), cfg_only, live_only);
1.1187 + } else {
1.1188 + tty->print_cr("invalid IR");
1.1189 + }
1.1190 +}
1.1191 +
1.1192 +
1.1193 +define_array(BlockListArray, BlockList*)
1.1194 +define_stack(BlockListList, BlockListArray)
1.1195 +
1.1196 +class PredecessorValidator : public BlockClosure {
1.1197 + private:
1.1198 + BlockListList* _predecessors;
1.1199 + BlockList* _blocks;
1.1200 +
1.1201 + static int cmp(BlockBegin** a, BlockBegin** b) {
1.1202 + return (*a)->block_id() - (*b)->block_id();
1.1203 + }
1.1204 +
1.1205 + public:
1.1206 + PredecessorValidator(IR* hir) {
1.1207 + ResourceMark rm;
1.1208 + _predecessors = new BlockListList(BlockBegin::number_of_blocks(), NULL);
1.1209 + _blocks = new BlockList();
1.1210 +
1.1211 + int i;
1.1212 + hir->start()->iterate_preorder(this);
1.1213 + if (hir->code() != NULL) {
1.1214 + assert(hir->code()->length() == _blocks->length(), "must match");
1.1215 + for (i = 0; i < _blocks->length(); i++) {
1.1216 + assert(hir->code()->contains(_blocks->at(i)), "should be in both lists");
1.1217 + }
1.1218 + }
1.1219 +
1.1220 + for (i = 0; i < _blocks->length(); i++) {
1.1221 + BlockBegin* block = _blocks->at(i);
1.1222 + BlockList* preds = _predecessors->at(block->block_id());
1.1223 + if (preds == NULL) {
1.1224 + assert(block->number_of_preds() == 0, "should be the same");
1.1225 + continue;
1.1226 + }
1.1227 +
1.1228 + // clone the pred list so we can mutate it
1.1229 + BlockList* pred_copy = new BlockList();
1.1230 + int j;
1.1231 + for (j = 0; j < block->number_of_preds(); j++) {
1.1232 + pred_copy->append(block->pred_at(j));
1.1233 + }
1.1234 + // sort them in the same order
1.1235 + preds->sort(cmp);
1.1236 + pred_copy->sort(cmp);
1.1237 + int length = MIN2(preds->length(), block->number_of_preds());
1.1238 + for (j = 0; j < block->number_of_preds(); j++) {
1.1239 + assert(preds->at(j) == pred_copy->at(j), "must match");
1.1240 + }
1.1241 +
1.1242 + assert(preds->length() == block->number_of_preds(), "should be the same");
1.1243 + }
1.1244 + }
1.1245 +
1.1246 + virtual void block_do(BlockBegin* block) {
1.1247 + _blocks->append(block);
1.1248 + BlockEnd* be = block->end();
1.1249 + int n = be->number_of_sux();
1.1250 + int i;
1.1251 + for (i = 0; i < n; i++) {
1.1252 + BlockBegin* sux = be->sux_at(i);
1.1253 + assert(!sux->is_set(BlockBegin::exception_entry_flag), "must not be xhandler");
1.1254 +
1.1255 + BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1.1256 + if (preds == NULL) {
1.1257 + preds = new BlockList();
1.1258 + _predecessors->at_put(sux->block_id(), preds);
1.1259 + }
1.1260 + preds->append(block);
1.1261 + }
1.1262 +
1.1263 + n = block->number_of_exception_handlers();
1.1264 + for (i = 0; i < n; i++) {
1.1265 + BlockBegin* sux = block->exception_handler_at(i);
1.1266 + assert(sux->is_set(BlockBegin::exception_entry_flag), "must be xhandler");
1.1267 +
1.1268 + BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1.1269 + if (preds == NULL) {
1.1270 + preds = new BlockList();
1.1271 + _predecessors->at_put(sux->block_id(), preds);
1.1272 + }
1.1273 + preds->append(block);
1.1274 + }
1.1275 + }
1.1276 +};
1.1277 +
1.1278 +void IR::verify() {
1.1279 +#ifdef ASSERT
1.1280 + PredecessorValidator pv(this);
1.1281 +#endif
1.1282 +}
1.1283 +
1.1284 +#endif // PRODUCT
1.1285 +
1.1286 +void SubstitutionResolver::substitute(Value* v) {
1.1287 + Value v0 = *v;
1.1288 + if (v0) {
1.1289 + Value vs = v0->subst();
1.1290 + if (vs != v0) {
1.1291 + *v = v0->subst();
1.1292 + }
1.1293 + }
1.1294 +}
1.1295 +
1.1296 +#ifdef ASSERT
1.1297 +void check_substitute(Value* v) {
1.1298 + Value v0 = *v;
1.1299 + if (v0) {
1.1300 + Value vs = v0->subst();
1.1301 + assert(vs == v0, "missed substitution");
1.1302 + }
1.1303 +}
1.1304 +#endif
1.1305 +
1.1306 +
1.1307 +void SubstitutionResolver::block_do(BlockBegin* block) {
1.1308 + Instruction* last = NULL;
1.1309 + for (Instruction* n = block; n != NULL;) {
1.1310 + n->values_do(substitute);
1.1311 + // need to remove this instruction from the instruction stream
1.1312 + if (n->subst() != n) {
1.1313 + assert(last != NULL, "must have last");
1.1314 + last->set_next(n->next(), n->next()->bci());
1.1315 + } else {
1.1316 + last = n;
1.1317 + }
1.1318 + n = last->next();
1.1319 + }
1.1320 +
1.1321 +#ifdef ASSERT
1.1322 + if (block->state()) block->state()->values_do(check_substitute);
1.1323 + block->block_values_do(check_substitute);
1.1324 + if (block->end() && block->end()->state()) block->end()->state()->values_do(check_substitute);
1.1325 +#endif
1.1326 +}
