1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/share/vm/c1/c1_IR.cpp Wed Apr 27 01:25:04 2016 +0800
1.3 @@ -0,0 +1,1405 @@
1.4 +/*
1.5 + * Copyright (c) 1999, 2013, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
1.23 + * or visit www.oracle.com if you need additional information or have any
1.24 + * questions.
1.25 + *
1.26 + */
1.27 +
1.28 +#include "precompiled.hpp"
1.29 +#include "c1/c1_Compilation.hpp"
1.30 +#include "c1/c1_FrameMap.hpp"
1.31 +#include "c1/c1_GraphBuilder.hpp"
1.32 +#include "c1/c1_IR.hpp"
1.33 +#include "c1/c1_InstructionPrinter.hpp"
1.34 +#include "c1/c1_Optimizer.hpp"
1.35 +#include "utilities/bitMap.inline.hpp"
1.36 +
1.37 +
1.38 +// Implementation of XHandlers
1.39 +//
1.40 +// Note: This code could eventually go away if we are
1.41 +// just using the ciExceptionHandlerStream.
1.42 +
1.43 +XHandlers::XHandlers(ciMethod* method) : _list(method->exception_table_length()) {
1.44 + ciExceptionHandlerStream s(method);
1.45 + while (!s.is_done()) {
1.46 + _list.append(new XHandler(s.handler()));
1.47 + s.next();
1.48 + }
1.49 + assert(s.count() == method->exception_table_length(), "exception table lengths inconsistent");
1.50 +}
1.51 +
1.52 +// deep copy of all XHandler contained in list
1.53 +XHandlers::XHandlers(XHandlers* other) :
1.54 + _list(other->length())
1.55 +{
1.56 + for (int i = 0; i < other->length(); i++) {
1.57 + _list.append(new XHandler(other->handler_at(i)));
1.58 + }
1.59 +}
1.60 +
1.61 +// Returns whether a particular exception type can be caught. Also
1.62 +// returns true if klass is unloaded or any exception handler
1.63 +// classes are unloaded. type_is_exact indicates whether the throw
1.64 +// is known to be exactly that class or it might throw a subtype.
1.65 +bool XHandlers::could_catch(ciInstanceKlass* klass, bool type_is_exact) const {
1.66 + // the type is unknown so be conservative
1.67 + if (!klass->is_loaded()) {
1.68 + return true;
1.69 + }
1.70 +
1.71 + for (int i = 0; i < length(); i++) {
1.72 + XHandler* handler = handler_at(i);
1.73 + if (handler->is_catch_all()) {
1.74 + // catch of ANY
1.75 + return true;
1.76 + }
1.77 + ciInstanceKlass* handler_klass = handler->catch_klass();
1.78 + // if it's unknown it might be catchable
1.79 + if (!handler_klass->is_loaded()) {
1.80 + return true;
1.81 + }
1.82 + // if the throw type is definitely a subtype of the catch type
1.83 + // then it can be caught.
1.84 + if (klass->is_subtype_of(handler_klass)) {
1.85 + return true;
1.86 + }
1.87 + if (!type_is_exact) {
1.88 + // If the type isn't exactly known then it can also be caught by
1.89 + // catch statements where the inexact type is a subtype of the
1.90 + // catch type.
1.91 + // given: foo extends bar extends Exception
1.92 + // throw bar can be caught by catch foo, catch bar, and catch
1.93 + // Exception, however it can't be caught by any handlers without
1.94 + // bar in its type hierarchy.
1.95 + if (handler_klass->is_subtype_of(klass)) {
1.96 + return true;
1.97 + }
1.98 + }
1.99 + }
1.100 +
1.101 + return false;
1.102 +}
1.103 +
1.104 +
1.105 +bool XHandlers::equals(XHandlers* others) const {
1.106 + if (others == NULL) return false;
1.107 + if (length() != others->length()) return false;
1.108 +
1.109 + for (int i = 0; i < length(); i++) {
1.110 + if (!handler_at(i)->equals(others->handler_at(i))) return false;
1.111 + }
1.112 + return true;
1.113 +}
1.114 +
1.115 +bool XHandler::equals(XHandler* other) const {
1.116 + assert(entry_pco() != -1 && other->entry_pco() != -1, "must have entry_pco");
1.117 +
1.118 + if (entry_pco() != other->entry_pco()) return false;
1.119 + if (scope_count() != other->scope_count()) return false;
1.120 + if (_desc != other->_desc) return false;
1.121 +
1.122 + assert(entry_block() == other->entry_block(), "entry_block must be equal when entry_pco is equal");
1.123 + return true;
1.124 +}
1.125 +
1.126 +
1.127 +// Implementation of IRScope
1.128 +BlockBegin* IRScope::build_graph(Compilation* compilation, int osr_bci) {
1.129 + GraphBuilder gm(compilation, this);
1.130 + NOT_PRODUCT(if (PrintValueNumbering && Verbose) gm.print_stats());
1.131 + if (compilation->bailed_out()) return NULL;
1.132 + return gm.start();
1.133 +}
1.134 +
1.135 +
1.136 +IRScope::IRScope(Compilation* compilation, IRScope* caller, int caller_bci, ciMethod* method, int osr_bci, bool create_graph)
1.137 +: _callees(2)
1.138 +, _compilation(compilation)
1.139 +, _requires_phi_function(method->max_locals())
1.140 +{
1.141 + _caller = caller;
1.142 + _level = caller == NULL ? 0 : caller->level() + 1;
1.143 + _method = method;
1.144 + _xhandlers = new XHandlers(method);
1.145 + _number_of_locks = 0;
1.146 + _monitor_pairing_ok = method->has_balanced_monitors();
1.147 + _wrote_final = false;
1.148 + _start = NULL;
1.149 +
1.150 + if (osr_bci == -1) {
1.151 + _requires_phi_function.clear();
1.152 + } else {
1.153 + // selective creation of phi functions is not possibel in osr-methods
1.154 + _requires_phi_function.set_range(0, method->max_locals());
1.155 + }
1.156 +
1.157 + assert(method->holder()->is_loaded() , "method holder must be loaded");
1.158 +
1.159 + // build graph if monitor pairing is ok
1.160 + if (create_graph && monitor_pairing_ok()) _start = build_graph(compilation, osr_bci);
1.161 +}
1.162 +
1.163 +
1.164 +int IRScope::max_stack() const {
1.165 + int my_max = method()->max_stack();
1.166 + int callee_max = 0;
1.167 + for (int i = 0; i < number_of_callees(); i++) {
1.168 + callee_max = MAX2(callee_max, callee_no(i)->max_stack());
1.169 + }
1.170 + return my_max + callee_max;
1.171 +}
1.172 +
1.173 +
1.174 +bool IRScopeDebugInfo::should_reexecute() {
1.175 + ciMethod* cur_method = scope()->method();
1.176 + int cur_bci = bci();
1.177 + if (cur_method != NULL && cur_bci != SynchronizationEntryBCI) {
1.178 + Bytecodes::Code code = cur_method->java_code_at_bci(cur_bci);
1.179 + return Interpreter::bytecode_should_reexecute(code);
1.180 + } else
1.181 + return false;
1.182 +}
1.183 +
1.184 +
1.185 +// Implementation of CodeEmitInfo
1.186 +
1.187 +// Stack must be NON-null
1.188 +CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception)
1.189 + : _scope(stack->scope())
1.190 + , _scope_debug_info(NULL)
1.191 + , _oop_map(NULL)
1.192 + , _stack(stack)
1.193 + , _exception_handlers(exception_handlers)
1.194 + , _is_method_handle_invoke(false)
1.195 + , _deoptimize_on_exception(deoptimize_on_exception) {
1.196 + assert(_stack != NULL, "must be non null");
1.197 +}
1.198 +
1.199 +
1.200 +CodeEmitInfo::CodeEmitInfo(CodeEmitInfo* info, ValueStack* stack)
1.201 + : _scope(info->_scope)
1.202 + , _exception_handlers(NULL)
1.203 + , _scope_debug_info(NULL)
1.204 + , _oop_map(NULL)
1.205 + , _stack(stack == NULL ? info->_stack : stack)
1.206 + , _is_method_handle_invoke(info->_is_method_handle_invoke)
1.207 + , _deoptimize_on_exception(info->_deoptimize_on_exception) {
1.208 +
1.209 + // deep copy of exception handlers
1.210 + if (info->_exception_handlers != NULL) {
1.211 + _exception_handlers = new XHandlers(info->_exception_handlers);
1.212 + }
1.213 +}
1.214 +
1.215 +
1.216 +void CodeEmitInfo::record_debug_info(DebugInformationRecorder* recorder, int pc_offset) {
1.217 + // record the safepoint before recording the debug info for enclosing scopes
1.218 + recorder->add_safepoint(pc_offset, _oop_map->deep_copy());
1.219 + _scope_debug_info->record_debug_info(recorder, pc_offset, true/*topmost*/, _is_method_handle_invoke);
1.220 + recorder->end_safepoint(pc_offset);
1.221 +}
1.222 +
1.223 +
1.224 +void CodeEmitInfo::add_register_oop(LIR_Opr opr) {
1.225 + assert(_oop_map != NULL, "oop map must already exist");
1.226 + assert(opr->is_single_cpu(), "should not call otherwise");
1.227 +
1.228 + VMReg name = frame_map()->regname(opr);
1.229 + _oop_map->set_oop(name);
1.230 +}
1.231 +
1.232 +// Mirror the stack size calculation in the deopt code
1.233 +// How much stack space would we need at this point in the program in
1.234 +// case of deoptimization?
1.235 +int CodeEmitInfo::interpreter_frame_size() const {
1.236 + ValueStack* state = _stack;
1.237 + int size = 0;
1.238 + int callee_parameters = 0;
1.239 + int callee_locals = 0;
1.240 + int extra_args = state->scope()->method()->max_stack() - state->stack_size();
1.241 +
1.242 + while (state != NULL) {
1.243 + int locks = state->locks_size();
1.244 + int temps = state->stack_size();
1.245 + bool is_top_frame = (state == _stack);
1.246 + ciMethod* method = state->scope()->method();
1.247 +
1.248 + int frame_size = BytesPerWord * Interpreter::size_activation(method->max_stack(),
1.249 + temps + callee_parameters,
1.250 + extra_args,
1.251 + locks,
1.252 + callee_parameters,
1.253 + callee_locals,
1.254 + is_top_frame);
1.255 + size += frame_size;
1.256 +
1.257 + callee_parameters = method->size_of_parameters();
1.258 + callee_locals = method->max_locals();
1.259 + extra_args = 0;
1.260 + state = state->caller_state();
1.261 + }
1.262 + return size + Deoptimization::last_frame_adjust(0, callee_locals) * BytesPerWord;
1.263 +}
1.264 +
1.265 +// Implementation of IR
1.266 +
1.267 +IR::IR(Compilation* compilation, ciMethod* method, int osr_bci) :
1.268 + _locals_size(in_WordSize(-1))
1.269 + , _num_loops(0) {
1.270 + // setup IR fields
1.271 + _compilation = compilation;
1.272 + _top_scope = new IRScope(compilation, NULL, -1, method, osr_bci, true);
1.273 + _code = NULL;
1.274 +}
1.275 +
1.276 +
1.277 +void IR::optimize_blocks() {
1.278 + Optimizer opt(this);
1.279 + if (!compilation()->profile_branches()) {
1.280 + if (DoCEE) {
1.281 + opt.eliminate_conditional_expressions();
1.282 +#ifndef PRODUCT
1.283 + if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after CEE"); print(true); }
1.284 + if (PrintIR || PrintIR1 ) { tty->print_cr("IR after CEE"); print(false); }
1.285 +#endif
1.286 + }
1.287 + if (EliminateBlocks) {
1.288 + opt.eliminate_blocks();
1.289 +#ifndef PRODUCT
1.290 + if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after block elimination"); print(true); }
1.291 + if (PrintIR || PrintIR1 ) { tty->print_cr("IR after block elimination"); print(false); }
1.292 +#endif
1.293 + }
1.294 + }
1.295 +}
1.296 +
1.297 +void IR::eliminate_null_checks() {
1.298 + Optimizer opt(this);
1.299 + if (EliminateNullChecks) {
1.300 + opt.eliminate_null_checks();
1.301 +#ifndef PRODUCT
1.302 + if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after null check elimination"); print(true); }
1.303 + if (PrintIR || PrintIR1 ) { tty->print_cr("IR after null check elimination"); print(false); }
1.304 +#endif
1.305 + }
1.306 +}
1.307 +
1.308 +
1.309 +static int sort_pairs(BlockPair** a, BlockPair** b) {
1.310 + if ((*a)->from() == (*b)->from()) {
1.311 + return (*a)->to()->block_id() - (*b)->to()->block_id();
1.312 + } else {
1.313 + return (*a)->from()->block_id() - (*b)->from()->block_id();
1.314 + }
1.315 +}
1.316 +
1.317 +
1.318 +class CriticalEdgeFinder: public BlockClosure {
1.319 + BlockPairList blocks;
1.320 + IR* _ir;
1.321 +
1.322 + public:
1.323 + CriticalEdgeFinder(IR* ir): _ir(ir) {}
1.324 + void block_do(BlockBegin* bb) {
1.325 + BlockEnd* be = bb->end();
1.326 + int nos = be->number_of_sux();
1.327 + if (nos >= 2) {
1.328 + for (int i = 0; i < nos; i++) {
1.329 + BlockBegin* sux = be->sux_at(i);
1.330 + if (sux->number_of_preds() >= 2) {
1.331 + blocks.append(new BlockPair(bb, sux));
1.332 + }
1.333 + }
1.334 + }
1.335 + }
1.336 +
1.337 + void split_edges() {
1.338 + BlockPair* last_pair = NULL;
1.339 + blocks.sort(sort_pairs);
1.340 + for (int i = 0; i < blocks.length(); i++) {
1.341 + BlockPair* pair = blocks.at(i);
1.342 + if (last_pair != NULL && pair->is_same(last_pair)) continue;
1.343 + BlockBegin* from = pair->from();
1.344 + BlockBegin* to = pair->to();
1.345 + BlockBegin* split = from->insert_block_between(to);
1.346 +#ifndef PRODUCT
1.347 + if ((PrintIR || PrintIR1) && Verbose) {
1.348 + tty->print_cr("Split critical edge B%d -> B%d (new block B%d)",
1.349 + from->block_id(), to->block_id(), split->block_id());
1.350 + }
1.351 +#endif
1.352 + last_pair = pair;
1.353 + }
1.354 + }
1.355 +};
1.356 +
1.357 +void IR::split_critical_edges() {
1.358 + CriticalEdgeFinder cef(this);
1.359 +
1.360 + iterate_preorder(&cef);
1.361 + cef.split_edges();
1.362 +}
1.363 +
1.364 +
1.365 +class UseCountComputer: public ValueVisitor, BlockClosure {
1.366 + private:
1.367 + void visit(Value* n) {
1.368 + // Local instructions and Phis for expression stack values at the
1.369 + // start of basic blocks are not added to the instruction list
1.370 + if (!(*n)->is_linked() && (*n)->can_be_linked()) {
1.371 + assert(false, "a node was not appended to the graph");
1.372 + Compilation::current()->bailout("a node was not appended to the graph");
1.373 + }
1.374 + // use n's input if not visited before
1.375 + if (!(*n)->is_pinned() && !(*n)->has_uses()) {
1.376 + // note: a) if the instruction is pinned, it will be handled by compute_use_count
1.377 + // b) if the instruction has uses, it was touched before
1.378 + // => in both cases we don't need to update n's values
1.379 + uses_do(n);
1.380 + }
1.381 + // use n
1.382 + (*n)->_use_count++;
1.383 + }
1.384 +
1.385 + Values* worklist;
1.386 + int depth;
1.387 + enum {
1.388 + max_recurse_depth = 20
1.389 + };
1.390 +
1.391 + void uses_do(Value* n) {
1.392 + depth++;
1.393 + if (depth > max_recurse_depth) {
1.394 + // don't allow the traversal to recurse too deeply
1.395 + worklist->push(*n);
1.396 + } else {
1.397 + (*n)->input_values_do(this);
1.398 + // special handling for some instructions
1.399 + if ((*n)->as_BlockEnd() != NULL) {
1.400 + // note on BlockEnd:
1.401 + // must 'use' the stack only if the method doesn't
1.402 + // terminate, however, in those cases stack is empty
1.403 + (*n)->state_values_do(this);
1.404 + }
1.405 + }
1.406 + depth--;
1.407 + }
1.408 +
1.409 + void block_do(BlockBegin* b) {
1.410 + depth = 0;
1.411 + // process all pinned nodes as the roots of expression trees
1.412 + for (Instruction* n = b; n != NULL; n = n->next()) {
1.413 + if (n->is_pinned()) uses_do(&n);
1.414 + }
1.415 + assert(depth == 0, "should have counted back down");
1.416 +
1.417 + // now process any unpinned nodes which recursed too deeply
1.418 + while (worklist->length() > 0) {
1.419 + Value t = worklist->pop();
1.420 + if (!t->is_pinned()) {
1.421 + // compute the use count
1.422 + uses_do(&t);
1.423 +
1.424 + // pin the instruction so that LIRGenerator doesn't recurse
1.425 + // too deeply during it's evaluation.
1.426 + t->pin();
1.427 + }
1.428 + }
1.429 + assert(depth == 0, "should have counted back down");
1.430 + }
1.431 +
1.432 + UseCountComputer() {
1.433 + worklist = new Values();
1.434 + depth = 0;
1.435 + }
1.436 +
1.437 + public:
1.438 + static void compute(BlockList* blocks) {
1.439 + UseCountComputer ucc;
1.440 + blocks->iterate_backward(&ucc);
1.441 + }
1.442 +};
1.443 +
1.444 +
1.445 +// helper macro for short definition of trace-output inside code
1.446 +#ifndef PRODUCT
1.447 + #define TRACE_LINEAR_SCAN(level, code) \
1.448 + if (TraceLinearScanLevel >= level) { \
1.449 + code; \
1.450 + }
1.451 +#else
1.452 + #define TRACE_LINEAR_SCAN(level, code)
1.453 +#endif
1.454 +
1.455 +class ComputeLinearScanOrder : public StackObj {
1.456 + private:
1.457 + int _max_block_id; // the highest block_id of a block
1.458 + int _num_blocks; // total number of blocks (smaller than _max_block_id)
1.459 + int _num_loops; // total number of loops
1.460 + bool _iterative_dominators;// method requires iterative computation of dominatiors
1.461 +
1.462 + BlockList* _linear_scan_order; // the resulting list of blocks in correct order
1.463 +
1.464 + BitMap _visited_blocks; // used for recursive processing of blocks
1.465 + BitMap _active_blocks; // used for recursive processing of blocks
1.466 + BitMap _dominator_blocks; // temproary BitMap used for computation of dominator
1.467 + intArray _forward_branches; // number of incoming forward branches for each block
1.468 + BlockList _loop_end_blocks; // list of all loop end blocks collected during count_edges
1.469 + BitMap2D _loop_map; // two-dimensional bit set: a bit is set if a block is contained in a loop
1.470 + BlockList _work_list; // temporary list (used in mark_loops and compute_order)
1.471 + BlockList _loop_headers;
1.472 +
1.473 + Compilation* _compilation;
1.474 +
1.475 + // accessors for _visited_blocks and _active_blocks
1.476 + void init_visited() { _active_blocks.clear(); _visited_blocks.clear(); }
1.477 + bool is_visited(BlockBegin* b) const { return _visited_blocks.at(b->block_id()); }
1.478 + bool is_active(BlockBegin* b) const { return _active_blocks.at(b->block_id()); }
1.479 + void set_visited(BlockBegin* b) { assert(!is_visited(b), "already set"); _visited_blocks.set_bit(b->block_id()); }
1.480 + void set_active(BlockBegin* b) { assert(!is_active(b), "already set"); _active_blocks.set_bit(b->block_id()); }
1.481 + void clear_active(BlockBegin* b) { assert(is_active(b), "not already"); _active_blocks.clear_bit(b->block_id()); }
1.482 +
1.483 + // accessors for _forward_branches
1.484 + void inc_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) + 1); }
1.485 + 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()); }
1.486 +
1.487 + // accessors for _loop_map
1.488 + bool is_block_in_loop (int loop_idx, BlockBegin* b) const { return _loop_map.at(loop_idx, b->block_id()); }
1.489 + void set_block_in_loop (int loop_idx, BlockBegin* b) { _loop_map.set_bit(loop_idx, b->block_id()); }
1.490 + void clear_block_in_loop(int loop_idx, int block_id) { _loop_map.clear_bit(loop_idx, block_id); }
1.491 +
1.492 + // count edges between blocks
1.493 + void count_edges(BlockBegin* cur, BlockBegin* parent);
1.494 +
1.495 + // loop detection
1.496 + void mark_loops();
1.497 + void clear_non_natural_loops(BlockBegin* start_block);
1.498 + void assign_loop_depth(BlockBegin* start_block);
1.499 +
1.500 + // computation of final block order
1.501 + BlockBegin* common_dominator(BlockBegin* a, BlockBegin* b);
1.502 + void compute_dominator(BlockBegin* cur, BlockBegin* parent);
1.503 + int compute_weight(BlockBegin* cur);
1.504 + bool ready_for_processing(BlockBegin* cur);
1.505 + void sort_into_work_list(BlockBegin* b);
1.506 + void append_block(BlockBegin* cur);
1.507 + void compute_order(BlockBegin* start_block);
1.508 +
1.509 + // fixup of dominators for non-natural loops
1.510 + bool compute_dominators_iter();
1.511 + void compute_dominators();
1.512 +
1.513 + // debug functions
1.514 + NOT_PRODUCT(void print_blocks();)
1.515 + DEBUG_ONLY(void verify();)
1.516 +
1.517 + Compilation* compilation() const { return _compilation; }
1.518 + public:
1.519 + ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block);
1.520 +
1.521 + // accessors for final result
1.522 + BlockList* linear_scan_order() const { return _linear_scan_order; }
1.523 + int num_loops() const { return _num_loops; }
1.524 +};
1.525 +
1.526 +
1.527 +ComputeLinearScanOrder::ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block) :
1.528 + _max_block_id(BlockBegin::number_of_blocks()),
1.529 + _num_blocks(0),
1.530 + _num_loops(0),
1.531 + _iterative_dominators(false),
1.532 + _visited_blocks(_max_block_id),
1.533 + _active_blocks(_max_block_id),
1.534 + _dominator_blocks(_max_block_id),
1.535 + _forward_branches(_max_block_id, 0),
1.536 + _loop_end_blocks(8),
1.537 + _work_list(8),
1.538 + _linear_scan_order(NULL), // initialized later with correct size
1.539 + _loop_map(0, 0), // initialized later with correct size
1.540 + _compilation(c)
1.541 +{
1.542 + TRACE_LINEAR_SCAN(2, tty->print_cr("***** computing linear-scan block order"));
1.543 +
1.544 + init_visited();
1.545 + count_edges(start_block, NULL);
1.546 +
1.547 + if (compilation()->is_profiling()) {
1.548 + ciMethod *method = compilation()->method();
1.549 + if (!method->is_accessor()) {
1.550 + ciMethodData* md = method->method_data_or_null();
1.551 + assert(md != NULL, "Sanity");
1.552 + md->set_compilation_stats(_num_loops, _num_blocks);
1.553 + }
1.554 + }
1.555 +
1.556 + if (_num_loops > 0) {
1.557 + mark_loops();
1.558 + clear_non_natural_loops(start_block);
1.559 + assign_loop_depth(start_block);
1.560 + }
1.561 +
1.562 + compute_order(start_block);
1.563 + compute_dominators();
1.564 +
1.565 + NOT_PRODUCT(print_blocks());
1.566 + DEBUG_ONLY(verify());
1.567 +}
1.568 +
1.569 +
1.570 +// Traverse the CFG:
1.571 +// * count total number of blocks
1.572 +// * count all incoming edges and backward incoming edges
1.573 +// * number loop header blocks
1.574 +// * create a list with all loop end blocks
1.575 +void ComputeLinearScanOrder::count_edges(BlockBegin* cur, BlockBegin* parent) {
1.576 + 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));
1.577 + assert(cur->dominator() == NULL, "dominator already initialized");
1.578 +
1.579 + if (is_active(cur)) {
1.580 + TRACE_LINEAR_SCAN(3, tty->print_cr("backward branch"));
1.581 + assert(is_visited(cur), "block must be visisted when block is active");
1.582 + assert(parent != NULL, "must have parent");
1.583 +
1.584 + cur->set(BlockBegin::linear_scan_loop_header_flag);
1.585 + cur->set(BlockBegin::backward_branch_target_flag);
1.586 +
1.587 + parent->set(BlockBegin::linear_scan_loop_end_flag);
1.588 +
1.589 + // When a loop header is also the start of an exception handler, then the backward branch is
1.590 + // an exception edge. Because such edges are usually critical edges which cannot be split, the
1.591 + // loop must be excluded here from processing.
1.592 + if (cur->is_set(BlockBegin::exception_entry_flag)) {
1.593 + // Make sure that dominators are correct in this weird situation
1.594 + _iterative_dominators = true;
1.595 + return;
1.596 + }
1.597 + assert(parent->number_of_sux() == 1 && parent->sux_at(0) == cur,
1.598 + "loop end blocks must have one successor (critical edges are split)");
1.599 +
1.600 + _loop_end_blocks.append(parent);
1.601 + return;
1.602 + }
1.603 +
1.604 + // increment number of incoming forward branches
1.605 + inc_forward_branches(cur);
1.606 +
1.607 + if (is_visited(cur)) {
1.608 + TRACE_LINEAR_SCAN(3, tty->print_cr("block already visited"));
1.609 + return;
1.610 + }
1.611 +
1.612 + _num_blocks++;
1.613 + set_visited(cur);
1.614 + set_active(cur);
1.615 +
1.616 + // recursive call for all successors
1.617 + int i;
1.618 + for (i = cur->number_of_sux() - 1; i >= 0; i--) {
1.619 + count_edges(cur->sux_at(i), cur);
1.620 + }
1.621 + for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
1.622 + count_edges(cur->exception_handler_at(i), cur);
1.623 + }
1.624 +
1.625 + clear_active(cur);
1.626 +
1.627 + // Each loop has a unique number.
1.628 + // When multiple loops are nested, assign_loop_depth assumes that the
1.629 + // innermost loop has the lowest number. This is guaranteed by setting
1.630 + // the loop number after the recursive calls for the successors above
1.631 + // have returned.
1.632 + if (cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
1.633 + assert(cur->loop_index() == -1, "cannot set loop-index twice");
1.634 + TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops));
1.635 +
1.636 + cur->set_loop_index(_num_loops);
1.637 + _loop_headers.append(cur);
1.638 + _num_loops++;
1.639 + }
1.640 +
1.641 + TRACE_LINEAR_SCAN(3, tty->print_cr("Finished count_edges for block B%d", cur->block_id()));
1.642 +}
1.643 +
1.644 +
1.645 +void ComputeLinearScanOrder::mark_loops() {
1.646 + TRACE_LINEAR_SCAN(3, tty->print_cr("----- marking loops"));
1.647 +
1.648 + _loop_map = BitMap2D(_num_loops, _max_block_id);
1.649 + _loop_map.clear();
1.650 +
1.651 + for (int i = _loop_end_blocks.length() - 1; i >= 0; i--) {
1.652 + BlockBegin* loop_end = _loop_end_blocks.at(i);
1.653 + BlockBegin* loop_start = loop_end->sux_at(0);
1.654 + int loop_idx = loop_start->loop_index();
1.655 +
1.656 + 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));
1.657 + assert(loop_end->is_set(BlockBegin::linear_scan_loop_end_flag), "loop end flag must be set");
1.658 + assert(loop_end->number_of_sux() == 1, "incorrect number of successors");
1.659 + assert(loop_start->is_set(BlockBegin::linear_scan_loop_header_flag), "loop header flag must be set");
1.660 + assert(loop_idx >= 0 && loop_idx < _num_loops, "loop index not set");
1.661 + assert(_work_list.is_empty(), "work list must be empty before processing");
1.662 +
1.663 + // add the end-block of the loop to the working list
1.664 + _work_list.push(loop_end);
1.665 + set_block_in_loop(loop_idx, loop_end);
1.666 + do {
1.667 + BlockBegin* cur = _work_list.pop();
1.668 +
1.669 + TRACE_LINEAR_SCAN(3, tty->print_cr(" processing B%d", cur->block_id()));
1.670 + assert(is_block_in_loop(loop_idx, cur), "bit in loop map must be set when block is in work list");
1.671 +
1.672 + // recursive processing of all predecessors ends when start block of loop is reached
1.673 + if (cur != loop_start && !cur->is_set(BlockBegin::osr_entry_flag)) {
1.674 + for (int j = cur->number_of_preds() - 1; j >= 0; j--) {
1.675 + BlockBegin* pred = cur->pred_at(j);
1.676 +
1.677 + if (!is_block_in_loop(loop_idx, pred) /*&& !pred->is_set(BlockBeginosr_entry_flag)*/) {
1.678 + // this predecessor has not been processed yet, so add it to work list
1.679 + TRACE_LINEAR_SCAN(3, tty->print_cr(" pushing B%d", pred->block_id()));
1.680 + _work_list.push(pred);
1.681 + set_block_in_loop(loop_idx, pred);
1.682 + }
1.683 + }
1.684 + }
1.685 + } while (!_work_list.is_empty());
1.686 + }
1.687 +}
1.688 +
1.689 +
1.690 +// check for non-natural loops (loops where the loop header does not dominate
1.691 +// all other loop blocks = loops with mulitple entries).
1.692 +// such loops are ignored
1.693 +void ComputeLinearScanOrder::clear_non_natural_loops(BlockBegin* start_block) {
1.694 + for (int i = _num_loops - 1; i >= 0; i--) {
1.695 + if (is_block_in_loop(i, start_block)) {
1.696 + // loop i contains the entry block of the method
1.697 + // -> this is not a natural loop, so ignore it
1.698 + TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i));
1.699 +
1.700 + BlockBegin *loop_header = _loop_headers.at(i);
1.701 + assert(loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Must be loop header");
1.702 +
1.703 + for (int j = 0; j < loop_header->number_of_preds(); j++) {
1.704 + BlockBegin *pred = loop_header->pred_at(j);
1.705 + pred->clear(BlockBegin::linear_scan_loop_end_flag);
1.706 + }
1.707 +
1.708 + loop_header->clear(BlockBegin::linear_scan_loop_header_flag);
1.709 +
1.710 + for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) {
1.711 + clear_block_in_loop(i, block_id);
1.712 + }
1.713 + _iterative_dominators = true;
1.714 + }
1.715 + }
1.716 +}
1.717 +
1.718 +void ComputeLinearScanOrder::assign_loop_depth(BlockBegin* start_block) {
1.719 + TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing loop-depth and weight"));
1.720 + init_visited();
1.721 +
1.722 + assert(_work_list.is_empty(), "work list must be empty before processing");
1.723 + _work_list.append(start_block);
1.724 +
1.725 + do {
1.726 + BlockBegin* cur = _work_list.pop();
1.727 +
1.728 + if (!is_visited(cur)) {
1.729 + set_visited(cur);
1.730 + TRACE_LINEAR_SCAN(4, tty->print_cr("Computing loop depth for block B%d", cur->block_id()));
1.731 +
1.732 + // compute loop-depth and loop-index for the block
1.733 + assert(cur->loop_depth() == 0, "cannot set loop-depth twice");
1.734 + int i;
1.735 + int loop_depth = 0;
1.736 + int min_loop_idx = -1;
1.737 + for (i = _num_loops - 1; i >= 0; i--) {
1.738 + if (is_block_in_loop(i, cur)) {
1.739 + loop_depth++;
1.740 + min_loop_idx = i;
1.741 + }
1.742 + }
1.743 + cur->set_loop_depth(loop_depth);
1.744 + cur->set_loop_index(min_loop_idx);
1.745 +
1.746 + // append all unvisited successors to work list
1.747 + for (i = cur->number_of_sux() - 1; i >= 0; i--) {
1.748 + _work_list.append(cur->sux_at(i));
1.749 + }
1.750 + for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
1.751 + _work_list.append(cur->exception_handler_at(i));
1.752 + }
1.753 + }
1.754 + } while (!_work_list.is_empty());
1.755 +}
1.756 +
1.757 +
1.758 +BlockBegin* ComputeLinearScanOrder::common_dominator(BlockBegin* a, BlockBegin* b) {
1.759 + assert(a != NULL && b != NULL, "must have input blocks");
1.760 +
1.761 + _dominator_blocks.clear();
1.762 + while (a != NULL) {
1.763 + _dominator_blocks.set_bit(a->block_id());
1.764 + assert(a->dominator() != NULL || a == _linear_scan_order->at(0), "dominator must be initialized");
1.765 + a = a->dominator();
1.766 + }
1.767 + while (b != NULL && !_dominator_blocks.at(b->block_id())) {
1.768 + assert(b->dominator() != NULL || b == _linear_scan_order->at(0), "dominator must be initialized");
1.769 + b = b->dominator();
1.770 + }
1.771 +
1.772 + assert(b != NULL, "could not find dominator");
1.773 + return b;
1.774 +}
1.775 +
1.776 +void ComputeLinearScanOrder::compute_dominator(BlockBegin* cur, BlockBegin* parent) {
1.777 + if (cur->dominator() == NULL) {
1.778 + TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: initializing dominator of B%d to B%d", cur->block_id(), parent->block_id()));
1.779 + cur->set_dominator(parent);
1.780 +
1.781 + } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) {
1.782 + 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()));
1.783 + // Does not hold for exception blocks
1.784 + assert(cur->number_of_preds() > 1 || cur->is_set(BlockBegin::exception_entry_flag), "");
1.785 + cur->set_dominator(common_dominator(cur->dominator(), parent));
1.786 + }
1.787 +
1.788 + // Additional edge to xhandler of all our successors
1.789 + // range check elimination needs that the state at the end of a
1.790 + // block be valid in every block it dominates so cur must dominate
1.791 + // the exception handlers of its successors.
1.792 + int num_cur_xhandler = cur->number_of_exception_handlers();
1.793 + for (int j = 0; j < num_cur_xhandler; j++) {
1.794 + BlockBegin* xhandler = cur->exception_handler_at(j);
1.795 + compute_dominator(xhandler, parent);
1.796 + }
1.797 +}
1.798 +
1.799 +
1.800 +int ComputeLinearScanOrder::compute_weight(BlockBegin* cur) {
1.801 + BlockBegin* single_sux = NULL;
1.802 + if (cur->number_of_sux() == 1) {
1.803 + single_sux = cur->sux_at(0);
1.804 + }
1.805 +
1.806 + // limit loop-depth to 15 bit (only for security reason, it will never be so big)
1.807 + int weight = (cur->loop_depth() & 0x7FFF) << 16;
1.808 +
1.809 + // general macro for short definition of weight flags
1.810 + // the first instance of INC_WEIGHT_IF has the highest priority
1.811 + int cur_bit = 15;
1.812 + #define INC_WEIGHT_IF(condition) if ((condition)) { weight |= (1 << cur_bit); } cur_bit--;
1.813 +
1.814 + // this is necessery for the (very rare) case that two successing blocks have
1.815 + // the same loop depth, but a different loop index (can happen for endless loops
1.816 + // with exception handlers)
1.817 + INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_header_flag));
1.818 +
1.819 + // loop end blocks (blocks that end with a backward branch) are added
1.820 + // after all other blocks of the loop.
1.821 + INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_end_flag));
1.822 +
1.823 + // critical edge split blocks are prefered because than they have a bigger
1.824 + // proability to be completely empty
1.825 + INC_WEIGHT_IF(cur->is_set(BlockBegin::critical_edge_split_flag));
1.826 +
1.827 + // exceptions should not be thrown in normal control flow, so these blocks
1.828 + // are added as late as possible
1.829 + INC_WEIGHT_IF(cur->end()->as_Throw() == NULL && (single_sux == NULL || single_sux->end()->as_Throw() == NULL));
1.830 + INC_WEIGHT_IF(cur->end()->as_Return() == NULL && (single_sux == NULL || single_sux->end()->as_Return() == NULL));
1.831 +
1.832 + // exceptions handlers are added as late as possible
1.833 + INC_WEIGHT_IF(!cur->is_set(BlockBegin::exception_entry_flag));
1.834 +
1.835 + // guarantee that weight is > 0
1.836 + weight |= 1;
1.837 +
1.838 + #undef INC_WEIGHT_IF
1.839 + assert(cur_bit >= 0, "too many flags");
1.840 + assert(weight > 0, "weight cannot become negative");
1.841 +
1.842 + return weight;
1.843 +}
1.844 +
1.845 +bool ComputeLinearScanOrder::ready_for_processing(BlockBegin* cur) {
1.846 + // Discount the edge just traveled.
1.847 + // When the number drops to zero, all forward branches were processed
1.848 + if (dec_forward_branches(cur) != 0) {
1.849 + return false;
1.850 + }
1.851 +
1.852 + assert(_linear_scan_order->index_of(cur) == -1, "block already processed (block can be ready only once)");
1.853 + assert(_work_list.index_of(cur) == -1, "block already in work-list (block can be ready only once)");
1.854 + return true;
1.855 +}
1.856 +
1.857 +void ComputeLinearScanOrder::sort_into_work_list(BlockBegin* cur) {
1.858 + assert(_work_list.index_of(cur) == -1, "block already in work list");
1.859 +
1.860 + int cur_weight = compute_weight(cur);
1.861 +
1.862 + // the linear_scan_number is used to cache the weight of a block
1.863 + cur->set_linear_scan_number(cur_weight);
1.864 +
1.865 +#ifndef PRODUCT
1.866 + if (StressLinearScan) {
1.867 + _work_list.insert_before(0, cur);
1.868 + return;
1.869 + }
1.870 +#endif
1.871 +
1.872 + _work_list.append(NULL); // provide space for new element
1.873 +
1.874 + int insert_idx = _work_list.length() - 1;
1.875 + while (insert_idx > 0 && _work_list.at(insert_idx - 1)->linear_scan_number() > cur_weight) {
1.876 + _work_list.at_put(insert_idx, _work_list.at(insert_idx - 1));
1.877 + insert_idx--;
1.878 + }
1.879 + _work_list.at_put(insert_idx, cur);
1.880 +
1.881 + TRACE_LINEAR_SCAN(3, tty->print_cr("Sorted B%d into worklist. new worklist:", cur->block_id()));
1.882 + 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()));
1.883 +
1.884 +#ifdef ASSERT
1.885 + for (int i = 0; i < _work_list.length(); i++) {
1.886 + assert(_work_list.at(i)->linear_scan_number() > 0, "weight not set");
1.887 + assert(i == 0 || _work_list.at(i - 1)->linear_scan_number() <= _work_list.at(i)->linear_scan_number(), "incorrect order in worklist");
1.888 + }
1.889 +#endif
1.890 +}
1.891 +
1.892 +void ComputeLinearScanOrder::append_block(BlockBegin* cur) {
1.893 + 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()));
1.894 + assert(_linear_scan_order->index_of(cur) == -1, "cannot add the same block twice");
1.895 +
1.896 + // currently, the linear scan order and code emit order are equal.
1.897 + // therefore the linear_scan_number and the weight of a block must also
1.898 + // be equal.
1.899 + cur->set_linear_scan_number(_linear_scan_order->length());
1.900 + _linear_scan_order->append(cur);
1.901 +}
1.902 +
1.903 +void ComputeLinearScanOrder::compute_order(BlockBegin* start_block) {
1.904 + TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing final block order"));
1.905 +
1.906 + // the start block is always the first block in the linear scan order
1.907 + _linear_scan_order = new BlockList(_num_blocks);
1.908 + append_block(start_block);
1.909 +
1.910 + assert(start_block->end()->as_Base() != NULL, "start block must end with Base-instruction");
1.911 + BlockBegin* std_entry = ((Base*)start_block->end())->std_entry();
1.912 + BlockBegin* osr_entry = ((Base*)start_block->end())->osr_entry();
1.913 +
1.914 + BlockBegin* sux_of_osr_entry = NULL;
1.915 + if (osr_entry != NULL) {
1.916 + // special handling for osr entry:
1.917 + // ignore the edge between the osr entry and its successor for processing
1.918 + // the osr entry block is added manually below
1.919 + assert(osr_entry->number_of_sux() == 1, "osr entry must have exactly one successor");
1.920 + 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");
1.921 +
1.922 + sux_of_osr_entry = osr_entry->sux_at(0);
1.923 + dec_forward_branches(sux_of_osr_entry);
1.924 +
1.925 + compute_dominator(osr_entry, start_block);
1.926 + _iterative_dominators = true;
1.927 + }
1.928 + compute_dominator(std_entry, start_block);
1.929 +
1.930 + // start processing with standard entry block
1.931 + assert(_work_list.is_empty(), "list must be empty before processing");
1.932 +
1.933 + if (ready_for_processing(std_entry)) {
1.934 + sort_into_work_list(std_entry);
1.935 + } else {
1.936 + assert(false, "the std_entry must be ready for processing (otherwise, the method has no start block)");
1.937 + }
1.938 +
1.939 + do {
1.940 + BlockBegin* cur = _work_list.pop();
1.941 +
1.942 + if (cur == sux_of_osr_entry) {
1.943 + // the osr entry block is ignored in normal processing, it is never added to the
1.944 + // work list. Instead, it is added as late as possible manually here.
1.945 + append_block(osr_entry);
1.946 + compute_dominator(cur, osr_entry);
1.947 + }
1.948 + append_block(cur);
1.949 +
1.950 + int i;
1.951 + int num_sux = cur->number_of_sux();
1.952 + // changed loop order to get "intuitive" order of if- and else-blocks
1.953 + for (i = 0; i < num_sux; i++) {
1.954 + BlockBegin* sux = cur->sux_at(i);
1.955 + compute_dominator(sux, cur);
1.956 + if (ready_for_processing(sux)) {
1.957 + sort_into_work_list(sux);
1.958 + }
1.959 + }
1.960 + num_sux = cur->number_of_exception_handlers();
1.961 + for (i = 0; i < num_sux; i++) {
1.962 + BlockBegin* sux = cur->exception_handler_at(i);
1.963 + if (ready_for_processing(sux)) {
1.964 + sort_into_work_list(sux);
1.965 + }
1.966 + }
1.967 + } while (_work_list.length() > 0);
1.968 +}
1.969 +
1.970 +
1.971 +bool ComputeLinearScanOrder::compute_dominators_iter() {
1.972 + bool changed = false;
1.973 + int num_blocks = _linear_scan_order->length();
1.974 +
1.975 + assert(_linear_scan_order->at(0)->dominator() == NULL, "must not have dominator");
1.976 + assert(_linear_scan_order->at(0)->number_of_preds() == 0, "must not have predecessors");
1.977 + for (int i = 1; i < num_blocks; i++) {
1.978 + BlockBegin* block = _linear_scan_order->at(i);
1.979 +
1.980 + BlockBegin* dominator = block->pred_at(0);
1.981 + int num_preds = block->number_of_preds();
1.982 +
1.983 + TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: Processing B%d", block->block_id()));
1.984 +
1.985 + for (int j = 0; j < num_preds; j++) {
1.986 +
1.987 + BlockBegin *pred = block->pred_at(j);
1.988 + TRACE_LINEAR_SCAN(4, tty->print_cr(" DOM: Subrocessing B%d", pred->block_id()));
1.989 +
1.990 + if (block->is_set(BlockBegin::exception_entry_flag)) {
1.991 + dominator = common_dominator(dominator, pred);
1.992 + int num_pred_preds = pred->number_of_preds();
1.993 + for (int k = 0; k < num_pred_preds; k++) {
1.994 + dominator = common_dominator(dominator, pred->pred_at(k));
1.995 + }
1.996 + } else {
1.997 + dominator = common_dominator(dominator, pred);
1.998 + }
1.999 + }
1.1000 +
1.1001 + if (dominator != block->dominator()) {
1.1002 + 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()));
1.1003 +
1.1004 + block->set_dominator(dominator);
1.1005 + changed = true;
1.1006 + }
1.1007 + }
1.1008 + return changed;
1.1009 +}
1.1010 +
1.1011 +void ComputeLinearScanOrder::compute_dominators() {
1.1012 + TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing dominators (iterative computation reqired: %d)", _iterative_dominators));
1.1013 +
1.1014 + // iterative computation of dominators is only required for methods with non-natural loops
1.1015 + // and OSR-methods. For all other methods, the dominators computed when generating the
1.1016 + // linear scan block order are correct.
1.1017 + if (_iterative_dominators) {
1.1018 + do {
1.1019 + TRACE_LINEAR_SCAN(1, tty->print_cr("DOM: next iteration of fix-point calculation"));
1.1020 + } while (compute_dominators_iter());
1.1021 + }
1.1022 +
1.1023 + // check that dominators are correct
1.1024 + assert(!compute_dominators_iter(), "fix point not reached");
1.1025 +
1.1026 + // Add Blocks to dominates-Array
1.1027 + int num_blocks = _linear_scan_order->length();
1.1028 + for (int i = 0; i < num_blocks; i++) {
1.1029 + BlockBegin* block = _linear_scan_order->at(i);
1.1030 +
1.1031 + BlockBegin *dom = block->dominator();
1.1032 + if (dom) {
1.1033 + assert(dom->dominator_depth() != -1, "Dominator must have been visited before");
1.1034 + dom->dominates()->append(block);
1.1035 + block->set_dominator_depth(dom->dominator_depth() + 1);
1.1036 + } else {
1.1037 + block->set_dominator_depth(0);
1.1038 + }
1.1039 + }
1.1040 +}
1.1041 +
1.1042 +
1.1043 +#ifndef PRODUCT
1.1044 +void ComputeLinearScanOrder::print_blocks() {
1.1045 + if (TraceLinearScanLevel >= 2) {
1.1046 + tty->print_cr("----- loop information:");
1.1047 + for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1.1048 + BlockBegin* cur = _linear_scan_order->at(block_idx);
1.1049 +
1.1050 + tty->print("%4d: B%2d: ", cur->linear_scan_number(), cur->block_id());
1.1051 + for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1.1052 + tty->print ("%d ", is_block_in_loop(loop_idx, cur));
1.1053 + }
1.1054 + tty->print_cr(" -> loop_index: %2d, loop_depth: %2d", cur->loop_index(), cur->loop_depth());
1.1055 + }
1.1056 + }
1.1057 +
1.1058 + if (TraceLinearScanLevel >= 1) {
1.1059 + tty->print_cr("----- linear-scan block order:");
1.1060 + for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1.1061 + BlockBegin* cur = _linear_scan_order->at(block_idx);
1.1062 + tty->print("%4d: B%2d loop: %2d depth: %2d", cur->linear_scan_number(), cur->block_id(), cur->loop_index(), cur->loop_depth());
1.1063 +
1.1064 + tty->print(cur->is_set(BlockBegin::exception_entry_flag) ? " ex" : " ");
1.1065 + tty->print(cur->is_set(BlockBegin::critical_edge_split_flag) ? " ce" : " ");
1.1066 + tty->print(cur->is_set(BlockBegin::linear_scan_loop_header_flag) ? " lh" : " ");
1.1067 + tty->print(cur->is_set(BlockBegin::linear_scan_loop_end_flag) ? " le" : " ");
1.1068 +
1.1069 + if (cur->dominator() != NULL) {
1.1070 + tty->print(" dom: B%d ", cur->dominator()->block_id());
1.1071 + } else {
1.1072 + tty->print(" dom: NULL ");
1.1073 + }
1.1074 +
1.1075 + if (cur->number_of_preds() > 0) {
1.1076 + tty->print(" preds: ");
1.1077 + for (int j = 0; j < cur->number_of_preds(); j++) {
1.1078 + BlockBegin* pred = cur->pred_at(j);
1.1079 + tty->print("B%d ", pred->block_id());
1.1080 + }
1.1081 + }
1.1082 + if (cur->number_of_sux() > 0) {
1.1083 + tty->print(" sux: ");
1.1084 + for (int j = 0; j < cur->number_of_sux(); j++) {
1.1085 + BlockBegin* sux = cur->sux_at(j);
1.1086 + tty->print("B%d ", sux->block_id());
1.1087 + }
1.1088 + }
1.1089 + if (cur->number_of_exception_handlers() > 0) {
1.1090 + tty->print(" ex: ");
1.1091 + for (int j = 0; j < cur->number_of_exception_handlers(); j++) {
1.1092 + BlockBegin* ex = cur->exception_handler_at(j);
1.1093 + tty->print("B%d ", ex->block_id());
1.1094 + }
1.1095 + }
1.1096 + tty->cr();
1.1097 + }
1.1098 + }
1.1099 +}
1.1100 +#endif
1.1101 +
1.1102 +#ifdef ASSERT
1.1103 +void ComputeLinearScanOrder::verify() {
1.1104 + assert(_linear_scan_order->length() == _num_blocks, "wrong number of blocks in list");
1.1105 +
1.1106 + if (StressLinearScan) {
1.1107 + // blocks are scrambled when StressLinearScan is used
1.1108 + return;
1.1109 + }
1.1110 +
1.1111 + // check that all successors of a block have a higher linear-scan-number
1.1112 + // and that all predecessors of a block have a lower linear-scan-number
1.1113 + // (only backward branches of loops are ignored)
1.1114 + int i;
1.1115 + for (i = 0; i < _linear_scan_order->length(); i++) {
1.1116 + BlockBegin* cur = _linear_scan_order->at(i);
1.1117 +
1.1118 + assert(cur->linear_scan_number() == i, "incorrect linear_scan_number");
1.1119 + assert(cur->linear_scan_number() >= 0 && cur->linear_scan_number() == _linear_scan_order->index_of(cur), "incorrect linear_scan_number");
1.1120 +
1.1121 + int j;
1.1122 + for (j = cur->number_of_sux() - 1; j >= 0; j--) {
1.1123 + BlockBegin* sux = cur->sux_at(j);
1.1124 +
1.1125 + assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->index_of(sux), "incorrect linear_scan_number");
1.1126 + if (!sux->is_set(BlockBegin::backward_branch_target_flag)) {
1.1127 + assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order");
1.1128 + }
1.1129 + if (cur->loop_depth() == sux->loop_depth()) {
1.1130 + 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");
1.1131 + }
1.1132 + }
1.1133 +
1.1134 + for (j = cur->number_of_preds() - 1; j >= 0; j--) {
1.1135 + BlockBegin* pred = cur->pred_at(j);
1.1136 +
1.1137 + assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->index_of(pred), "incorrect linear_scan_number");
1.1138 + if (!cur->is_set(BlockBegin::backward_branch_target_flag)) {
1.1139 + assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order");
1.1140 + }
1.1141 + if (cur->loop_depth() == pred->loop_depth()) {
1.1142 + 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");
1.1143 + }
1.1144 +
1.1145 + assert(cur->dominator()->linear_scan_number() <= cur->pred_at(j)->linear_scan_number(), "dominator must be before predecessors");
1.1146 + }
1.1147 +
1.1148 + // check dominator
1.1149 + if (i == 0) {
1.1150 + assert(cur->dominator() == NULL, "first block has no dominator");
1.1151 + } else {
1.1152 + assert(cur->dominator() != NULL, "all but first block must have dominator");
1.1153 + }
1.1154 + // Assertion does not hold for exception handlers
1.1155 + assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0) || cur->is_set(BlockBegin::exception_entry_flag), "Single predecessor must also be dominator");
1.1156 + }
1.1157 +
1.1158 + // check that all loops are continuous
1.1159 + for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1.1160 + int block_idx = 0;
1.1161 + assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "the first block must not be present in any loop");
1.1162 +
1.1163 + // skip blocks before the loop
1.1164 + while (block_idx < _num_blocks && !is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1.1165 + block_idx++;
1.1166 + }
1.1167 + // skip blocks of loop
1.1168 + while (block_idx < _num_blocks && is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1.1169 + block_idx++;
1.1170 + }
1.1171 + // after the first non-loop block, there must not be another loop-block
1.1172 + while (block_idx < _num_blocks) {
1.1173 + assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "loop not continuous in linear-scan order");
1.1174 + block_idx++;
1.1175 + }
1.1176 + }
1.1177 +}
1.1178 +#endif
1.1179 +
1.1180 +
1.1181 +void IR::compute_code() {
1.1182 + assert(is_valid(), "IR must be valid");
1.1183 +
1.1184 + ComputeLinearScanOrder compute_order(compilation(), start());
1.1185 + _num_loops = compute_order.num_loops();
1.1186 + _code = compute_order.linear_scan_order();
1.1187 +}
1.1188 +
1.1189 +
1.1190 +void IR::compute_use_counts() {
1.1191 + // make sure all values coming out of this block get evaluated.
1.1192 + int num_blocks = _code->length();
1.1193 + for (int i = 0; i < num_blocks; i++) {
1.1194 + _code->at(i)->end()->state()->pin_stack_for_linear_scan();
1.1195 + }
1.1196 +
1.1197 + // compute use counts
1.1198 + UseCountComputer::compute(_code);
1.1199 +}
1.1200 +
1.1201 +
1.1202 +void IR::iterate_preorder(BlockClosure* closure) {
1.1203 + assert(is_valid(), "IR must be valid");
1.1204 + start()->iterate_preorder(closure);
1.1205 +}
1.1206 +
1.1207 +
1.1208 +void IR::iterate_postorder(BlockClosure* closure) {
1.1209 + assert(is_valid(), "IR must be valid");
1.1210 + start()->iterate_postorder(closure);
1.1211 +}
1.1212 +
1.1213 +void IR::iterate_linear_scan_order(BlockClosure* closure) {
1.1214 + linear_scan_order()->iterate_forward(closure);
1.1215 +}
1.1216 +
1.1217 +
1.1218 +#ifndef PRODUCT
1.1219 +class BlockPrinter: public BlockClosure {
1.1220 + private:
1.1221 + InstructionPrinter* _ip;
1.1222 + bool _cfg_only;
1.1223 + bool _live_only;
1.1224 +
1.1225 + public:
1.1226 + BlockPrinter(InstructionPrinter* ip, bool cfg_only, bool live_only = false) {
1.1227 + _ip = ip;
1.1228 + _cfg_only = cfg_only;
1.1229 + _live_only = live_only;
1.1230 + }
1.1231 +
1.1232 + virtual void block_do(BlockBegin* block) {
1.1233 + if (_cfg_only) {
1.1234 + _ip->print_instr(block); tty->cr();
1.1235 + } else {
1.1236 + block->print_block(*_ip, _live_only);
1.1237 + }
1.1238 + }
1.1239 +};
1.1240 +
1.1241 +
1.1242 +void IR::print(BlockBegin* start, bool cfg_only, bool live_only) {
1.1243 + ttyLocker ttyl;
1.1244 + InstructionPrinter ip(!cfg_only);
1.1245 + BlockPrinter bp(&ip, cfg_only, live_only);
1.1246 + start->iterate_preorder(&bp);
1.1247 + tty->cr();
1.1248 +}
1.1249 +
1.1250 +void IR::print(bool cfg_only, bool live_only) {
1.1251 + if (is_valid()) {
1.1252 + print(start(), cfg_only, live_only);
1.1253 + } else {
1.1254 + tty->print_cr("invalid IR");
1.1255 + }
1.1256 +}
1.1257 +
1.1258 +
1.1259 +define_array(BlockListArray, BlockList*)
1.1260 +define_stack(BlockListList, BlockListArray)
1.1261 +
1.1262 +class PredecessorValidator : public BlockClosure {
1.1263 + private:
1.1264 + BlockListList* _predecessors;
1.1265 + BlockList* _blocks;
1.1266 +
1.1267 + static int cmp(BlockBegin** a, BlockBegin** b) {
1.1268 + return (*a)->block_id() - (*b)->block_id();
1.1269 + }
1.1270 +
1.1271 + public:
1.1272 + PredecessorValidator(IR* hir) {
1.1273 + ResourceMark rm;
1.1274 + _predecessors = new BlockListList(BlockBegin::number_of_blocks(), NULL);
1.1275 + _blocks = new BlockList();
1.1276 +
1.1277 + int i;
1.1278 + hir->start()->iterate_preorder(this);
1.1279 + if (hir->code() != NULL) {
1.1280 + assert(hir->code()->length() == _blocks->length(), "must match");
1.1281 + for (i = 0; i < _blocks->length(); i++) {
1.1282 + assert(hir->code()->contains(_blocks->at(i)), "should be in both lists");
1.1283 + }
1.1284 + }
1.1285 +
1.1286 + for (i = 0; i < _blocks->length(); i++) {
1.1287 + BlockBegin* block = _blocks->at(i);
1.1288 + BlockList* preds = _predecessors->at(block->block_id());
1.1289 + if (preds == NULL) {
1.1290 + assert(block->number_of_preds() == 0, "should be the same");
1.1291 + continue;
1.1292 + }
1.1293 +
1.1294 + // clone the pred list so we can mutate it
1.1295 + BlockList* pred_copy = new BlockList();
1.1296 + int j;
1.1297 + for (j = 0; j < block->number_of_preds(); j++) {
1.1298 + pred_copy->append(block->pred_at(j));
1.1299 + }
1.1300 + // sort them in the same order
1.1301 + preds->sort(cmp);
1.1302 + pred_copy->sort(cmp);
1.1303 + int length = MIN2(preds->length(), block->number_of_preds());
1.1304 + for (j = 0; j < block->number_of_preds(); j++) {
1.1305 + assert(preds->at(j) == pred_copy->at(j), "must match");
1.1306 + }
1.1307 +
1.1308 + assert(preds->length() == block->number_of_preds(), "should be the same");
1.1309 + }
1.1310 + }
1.1311 +
1.1312 + virtual void block_do(BlockBegin* block) {
1.1313 + _blocks->append(block);
1.1314 + BlockEnd* be = block->end();
1.1315 + int n = be->number_of_sux();
1.1316 + int i;
1.1317 + for (i = 0; i < n; i++) {
1.1318 + BlockBegin* sux = be->sux_at(i);
1.1319 + assert(!sux->is_set(BlockBegin::exception_entry_flag), "must not be xhandler");
1.1320 +
1.1321 + BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1.1322 + if (preds == NULL) {
1.1323 + preds = new BlockList();
1.1324 + _predecessors->at_put(sux->block_id(), preds);
1.1325 + }
1.1326 + preds->append(block);
1.1327 + }
1.1328 +
1.1329 + n = block->number_of_exception_handlers();
1.1330 + for (i = 0; i < n; i++) {
1.1331 + BlockBegin* sux = block->exception_handler_at(i);
1.1332 + assert(sux->is_set(BlockBegin::exception_entry_flag), "must be xhandler");
1.1333 +
1.1334 + BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1.1335 + if (preds == NULL) {
1.1336 + preds = new BlockList();
1.1337 + _predecessors->at_put(sux->block_id(), preds);
1.1338 + }
1.1339 + preds->append(block);
1.1340 + }
1.1341 + }
1.1342 +};
1.1343 +
1.1344 +class VerifyBlockBeginField : public BlockClosure {
1.1345 +
1.1346 +public:
1.1347 +
1.1348 + virtual void block_do(BlockBegin *block) {
1.1349 + for ( Instruction *cur = block; cur != NULL; cur = cur->next()) {
1.1350 + assert(cur->block() == block, "Block begin is not correct");
1.1351 + }
1.1352 + }
1.1353 +};
1.1354 +
1.1355 +void IR::verify() {
1.1356 +#ifdef ASSERT
1.1357 + PredecessorValidator pv(this);
1.1358 + VerifyBlockBeginField verifier;
1.1359 + this->iterate_postorder(&verifier);
1.1360 +#endif
1.1361 +}
1.1362 +
1.1363 +#endif // PRODUCT
1.1364 +
1.1365 +void SubstitutionResolver::visit(Value* v) {
1.1366 + Value v0 = *v;
1.1367 + if (v0) {
1.1368 + Value vs = v0->subst();
1.1369 + if (vs != v0) {
1.1370 + *v = v0->subst();
1.1371 + }
1.1372 + }
1.1373 +}
1.1374 +
1.1375 +#ifdef ASSERT
1.1376 +class SubstitutionChecker: public ValueVisitor {
1.1377 + void visit(Value* v) {
1.1378 + Value v0 = *v;
1.1379 + if (v0) {
1.1380 + Value vs = v0->subst();
1.1381 + assert(vs == v0, "missed substitution");
1.1382 + }
1.1383 + }
1.1384 +};
1.1385 +#endif
1.1386 +
1.1387 +
1.1388 +void SubstitutionResolver::block_do(BlockBegin* block) {
1.1389 + Instruction* last = NULL;
1.1390 + for (Instruction* n = block; n != NULL;) {
1.1391 + n->values_do(this);
1.1392 + // need to remove this instruction from the instruction stream
1.1393 + if (n->subst() != n) {
1.1394 + assert(last != NULL, "must have last");
1.1395 + last->set_next(n->next());
1.1396 + } else {
1.1397 + last = n;
1.1398 + }
1.1399 + n = last->next();
1.1400 + }
1.1401 +
1.1402 +#ifdef ASSERT
1.1403 + SubstitutionChecker check_substitute;
1.1404 + if (block->state()) block->state()->values_do(&check_substitute);
1.1405 + block->block_values_do(&check_substitute);
1.1406 + if (block->end() && block->end()->state()) block->end()->state()->values_do(&check_substitute);
1.1407 +#endif
1.1408 +}
