1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/share/vm/opto/chaitin.hpp Wed Apr 27 01:25:04 2016 +0800
1.3 @@ -0,0 +1,625 @@
1.4 +/*
1.5 + * Copyright (c) 1997, 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 +#ifndef SHARE_VM_OPTO_CHAITIN_HPP
1.29 +#define SHARE_VM_OPTO_CHAITIN_HPP
1.30 +
1.31 +#include "code/vmreg.hpp"
1.32 +#include "libadt/port.hpp"
1.33 +#include "memory/resourceArea.hpp"
1.34 +#include "opto/connode.hpp"
1.35 +#include "opto/live.hpp"
1.36 +#include "opto/matcher.hpp"
1.37 +#include "opto/phase.hpp"
1.38 +#include "opto/regalloc.hpp"
1.39 +#include "opto/regmask.hpp"
1.40 +
1.41 +class LoopTree;
1.42 +class MachCallNode;
1.43 +class MachSafePointNode;
1.44 +class Matcher;
1.45 +class PhaseCFG;
1.46 +class PhaseLive;
1.47 +class PhaseRegAlloc;
1.48 +class PhaseChaitin;
1.49 +
1.50 +#define OPTO_DEBUG_SPLIT_FREQ BLOCK_FREQUENCY(0.001)
1.51 +#define OPTO_LRG_HIGH_FREQ BLOCK_FREQUENCY(0.25)
1.52 +
1.53 +//------------------------------LRG--------------------------------------------
1.54 +// Live-RanGe structure.
1.55 +class LRG : public ResourceObj {
1.56 + friend class VMStructs;
1.57 +public:
1.58 + static const uint AllStack_size = 0xFFFFF; // This mask size is used to tell that the mask of this LRG supports stack positions
1.59 + enum { SPILL_REG=29999 }; // Register number of a spilled LRG
1.60 +
1.61 + double _cost; // 2 for loads/1 for stores times block freq
1.62 + double _area; // Sum of all simultaneously live values
1.63 + double score() const; // Compute score from cost and area
1.64 + double _maxfreq; // Maximum frequency of any def or use
1.65 +
1.66 + Node *_def; // Check for multi-def live ranges
1.67 +#ifndef PRODUCT
1.68 + GrowableArray<Node*>* _defs;
1.69 +#endif
1.70 +
1.71 + uint _risk_bias; // Index of LRG which we want to avoid color
1.72 + uint _copy_bias; // Index of LRG which we want to share color
1.73 +
1.74 + uint _next; // Index of next LRG in linked list
1.75 + uint _prev; // Index of prev LRG in linked list
1.76 +private:
1.77 + uint _reg; // Chosen register; undefined if mask is plural
1.78 +public:
1.79 + // Return chosen register for this LRG. Error if the LRG is not bound to
1.80 + // a single register.
1.81 + OptoReg::Name reg() const { return OptoReg::Name(_reg); }
1.82 + void set_reg( OptoReg::Name r ) { _reg = r; }
1.83 +
1.84 +private:
1.85 + uint _eff_degree; // Effective degree: Sum of neighbors _num_regs
1.86 +public:
1.87 + int degree() const { assert( _degree_valid , "" ); return _eff_degree; }
1.88 + // Degree starts not valid and any change to the IFG neighbor
1.89 + // set makes it not valid.
1.90 + void set_degree( uint degree ) {
1.91 + _eff_degree = degree;
1.92 + debug_only(_degree_valid = 1;)
1.93 + assert(!_mask.is_AllStack() || (_mask.is_AllStack() && lo_degree()), "_eff_degree can't be bigger than AllStack_size - _num_regs if the mask supports stack registers");
1.94 + }
1.95 + // Made a change that hammered degree
1.96 + void invalid_degree() { debug_only(_degree_valid=0;) }
1.97 + // Incrementally modify degree. If it was correct, it should remain correct
1.98 + void inc_degree( uint mod ) {
1.99 + _eff_degree += mod;
1.100 + assert(!_mask.is_AllStack() || (_mask.is_AllStack() && lo_degree()), "_eff_degree can't be bigger than AllStack_size - _num_regs if the mask supports stack registers");
1.101 + }
1.102 + // Compute the degree between 2 live ranges
1.103 + int compute_degree( LRG &l ) const;
1.104 +
1.105 +private:
1.106 + RegMask _mask; // Allowed registers for this LRG
1.107 + uint _mask_size; // cache of _mask.Size();
1.108 +public:
1.109 + int compute_mask_size() const { return _mask.is_AllStack() ? AllStack_size : _mask.Size(); }
1.110 + void set_mask_size( int size ) {
1.111 + assert((size == (int)AllStack_size) || (size == (int)_mask.Size()), "");
1.112 + _mask_size = size;
1.113 +#ifdef ASSERT
1.114 + _msize_valid=1;
1.115 + if (_is_vector) {
1.116 + assert(!_fat_proj, "sanity");
1.117 + _mask.verify_sets(_num_regs);
1.118 + } else if (_num_regs == 2 && !_fat_proj) {
1.119 + _mask.verify_pairs();
1.120 + }
1.121 +#endif
1.122 + }
1.123 + void compute_set_mask_size() { set_mask_size(compute_mask_size()); }
1.124 + int mask_size() const { assert( _msize_valid, "mask size not valid" );
1.125 + return _mask_size; }
1.126 + // Get the last mask size computed, even if it does not match the
1.127 + // count of bits in the current mask.
1.128 + int get_invalid_mask_size() const { return _mask_size; }
1.129 + const RegMask &mask() const { return _mask; }
1.130 + void set_mask( const RegMask &rm ) { _mask = rm; debug_only(_msize_valid=0;)}
1.131 + void AND( const RegMask &rm ) { _mask.AND(rm); debug_only(_msize_valid=0;)}
1.132 + void SUBTRACT( const RegMask &rm ) { _mask.SUBTRACT(rm); debug_only(_msize_valid=0;)}
1.133 + void Clear() { _mask.Clear() ; debug_only(_msize_valid=1); _mask_size = 0; }
1.134 + void Set_All() { _mask.Set_All(); debug_only(_msize_valid=1); _mask_size = RegMask::CHUNK_SIZE; }
1.135 + void Insert( OptoReg::Name reg ) { _mask.Insert(reg); debug_only(_msize_valid=0;) }
1.136 + void Remove( OptoReg::Name reg ) { _mask.Remove(reg); debug_only(_msize_valid=0;) }
1.137 + void clear_to_pairs() { _mask.clear_to_pairs(); debug_only(_msize_valid=0;) }
1.138 + void clear_to_sets() { _mask.clear_to_sets(_num_regs); debug_only(_msize_valid=0;) }
1.139 +
1.140 + // Number of registers this live range uses when it colors
1.141 +private:
1.142 + uint8 _num_regs; // 2 for Longs and Doubles, 1 for all else
1.143 + // except _num_regs is kill count for fat_proj
1.144 +public:
1.145 + int num_regs() const { return _num_regs; }
1.146 + void set_num_regs( int reg ) { assert( _num_regs == reg || !_num_regs, "" ); _num_regs = reg; }
1.147 +
1.148 +private:
1.149 + // Number of physical registers this live range uses when it colors
1.150 + // Architecture and register-set dependent
1.151 + uint8 _reg_pressure;
1.152 +public:
1.153 + void set_reg_pressure(int i) { _reg_pressure = i; }
1.154 + int reg_pressure() const { return _reg_pressure; }
1.155 +
1.156 + // How much 'wiggle room' does this live range have?
1.157 + // How many color choices can it make (scaled by _num_regs)?
1.158 + int degrees_of_freedom() const { return mask_size() - _num_regs; }
1.159 + // Bound LRGs have ZERO degrees of freedom. We also count
1.160 + // must_spill as bound.
1.161 + bool is_bound () const { return _is_bound; }
1.162 + // Negative degrees-of-freedom; even with no neighbors this
1.163 + // live range must spill.
1.164 + bool not_free() const { return degrees_of_freedom() < 0; }
1.165 + // Is this live range of "low-degree"? Trivially colorable?
1.166 + bool lo_degree () const { return degree() <= degrees_of_freedom(); }
1.167 + // Is this live range just barely "low-degree"? Trivially colorable?
1.168 + bool just_lo_degree () const { return degree() == degrees_of_freedom(); }
1.169 +
1.170 + uint _is_oop:1, // Live-range holds an oop
1.171 + _is_float:1, // True if in float registers
1.172 + _is_vector:1, // True if in vector registers
1.173 + _was_spilled1:1, // True if prior spilling on def
1.174 + _was_spilled2:1, // True if twice prior spilling on def
1.175 + _is_bound:1, // live range starts life with no
1.176 + // degrees of freedom.
1.177 + _direct_conflict:1, // True if def and use registers in conflict
1.178 + _must_spill:1, // live range has lost all degrees of freedom
1.179 + // If _fat_proj is set, live range does NOT require aligned, adjacent
1.180 + // registers and has NO interferences.
1.181 + // If _fat_proj is clear, live range requires num_regs() to be a power of
1.182 + // 2, and it requires registers to form an aligned, adjacent set.
1.183 + _fat_proj:1, //
1.184 + _was_lo:1, // Was lo-degree prior to coalesce
1.185 + _msize_valid:1, // _mask_size cache valid
1.186 + _degree_valid:1, // _degree cache valid
1.187 + _has_copy:1, // Adjacent to some copy instruction
1.188 + _at_risk:1; // Simplify says this guy is at risk to spill
1.189 +
1.190 +
1.191 + // Alive if non-zero, dead if zero
1.192 + bool alive() const { return _def != NULL; }
1.193 + bool is_multidef() const { return _def == NodeSentinel; }
1.194 + bool is_singledef() const { return _def != NodeSentinel; }
1.195 +
1.196 +#ifndef PRODUCT
1.197 + void dump( ) const;
1.198 +#endif
1.199 +};
1.200 +
1.201 +//------------------------------IFG--------------------------------------------
1.202 +// InterFerence Graph
1.203 +// An undirected graph implementation. Created with a fixed number of
1.204 +// vertices. Edges can be added & tested. Vertices can be removed, then
1.205 +// added back later with all edges intact. Can add edges between one vertex
1.206 +// and a list of other vertices. Can union vertices (and their edges)
1.207 +// together. The IFG needs to be really really fast, and also fairly
1.208 +// abstract! It needs abstraction so I can fiddle with the implementation to
1.209 +// get even more speed.
1.210 +class PhaseIFG : public Phase {
1.211 + friend class VMStructs;
1.212 + // Current implementation: a triangular adjacency list.
1.213 +
1.214 + // Array of adjacency-lists, indexed by live-range number
1.215 + IndexSet *_adjs;
1.216 +
1.217 + // Assertion bit for proper use of Squaring
1.218 + bool _is_square;
1.219 +
1.220 + // Live range structure goes here
1.221 + LRG *_lrgs; // Array of LRG structures
1.222 +
1.223 +public:
1.224 + // Largest live-range number
1.225 + uint _maxlrg;
1.226 +
1.227 + Arena *_arena;
1.228 +
1.229 + // Keep track of inserted and deleted Nodes
1.230 + VectorSet *_yanked;
1.231 +
1.232 + PhaseIFG( Arena *arena );
1.233 + void init( uint maxlrg );
1.234 +
1.235 + // Add edge between a and b. Returns true if actually addded.
1.236 + int add_edge( uint a, uint b );
1.237 +
1.238 + // Add edge between a and everything in the vector
1.239 + void add_vector( uint a, IndexSet *vec );
1.240 +
1.241 + // Test for edge existance
1.242 + int test_edge( uint a, uint b ) const;
1.243 +
1.244 + // Square-up matrix for faster Union
1.245 + void SquareUp();
1.246 +
1.247 + // Return number of LRG neighbors
1.248 + uint neighbor_cnt( uint a ) const { return _adjs[a].count(); }
1.249 + // Union edges of b into a on Squared-up matrix
1.250 + void Union( uint a, uint b );
1.251 + // Test for edge in Squared-up matrix
1.252 + int test_edge_sq( uint a, uint b ) const;
1.253 + // Yank a Node and all connected edges from the IFG. Be prepared to
1.254 + // re-insert the yanked Node in reverse order of yanking. Return a
1.255 + // list of neighbors (edges) yanked.
1.256 + IndexSet *remove_node( uint a );
1.257 + // Reinsert a yanked Node
1.258 + void re_insert( uint a );
1.259 + // Return set of neighbors
1.260 + IndexSet *neighbors( uint a ) const { return &_adjs[a]; }
1.261 +
1.262 +#ifndef PRODUCT
1.263 + // Dump the IFG
1.264 + void dump() const;
1.265 + void stats() const;
1.266 + void verify( const PhaseChaitin * ) const;
1.267 +#endif
1.268 +
1.269 + //--------------- Live Range Accessors
1.270 + LRG &lrgs(uint idx) const { assert(idx < _maxlrg, "oob"); return _lrgs[idx]; }
1.271 +
1.272 + // Compute and set effective degree. Might be folded into SquareUp().
1.273 + void Compute_Effective_Degree();
1.274 +
1.275 + // Compute effective degree as the sum of neighbors' _sizes.
1.276 + int effective_degree( uint lidx ) const;
1.277 +};
1.278 +
1.279 +// The LiveRangeMap class is responsible for storing node to live range id mapping.
1.280 +// Each node is mapped to a live range id (a virtual register). Nodes that are
1.281 +// not considered for register allocation are given live range id 0.
1.282 +class LiveRangeMap VALUE_OBJ_CLASS_SPEC {
1.283 +
1.284 +private:
1.285 +
1.286 + uint _max_lrg_id;
1.287 +
1.288 + // Union-find map. Declared as a short for speed.
1.289 + // Indexed by live-range number, it returns the compacted live-range number
1.290 + LRG_List _uf_map;
1.291 +
1.292 + // Map from Nodes to live ranges
1.293 + LRG_List _names;
1.294 +
1.295 + // Straight out of Tarjan's union-find algorithm
1.296 + uint find_compress(const Node *node) {
1.297 + uint lrg_id = find_compress(_names.at(node->_idx));
1.298 + _names.at_put(node->_idx, lrg_id);
1.299 + return lrg_id;
1.300 + }
1.301 +
1.302 + uint find_compress(uint lrg);
1.303 +
1.304 +public:
1.305 +
1.306 + const LRG_List& names() {
1.307 + return _names;
1.308 + }
1.309 +
1.310 + uint max_lrg_id() const {
1.311 + return _max_lrg_id;
1.312 + }
1.313 +
1.314 + void set_max_lrg_id(uint max_lrg_id) {
1.315 + _max_lrg_id = max_lrg_id;
1.316 + }
1.317 +
1.318 + uint size() const {
1.319 + return _names.length();
1.320 + }
1.321 +
1.322 + uint live_range_id(uint idx) const {
1.323 + return _names.at(idx);
1.324 + }
1.325 +
1.326 + uint live_range_id(const Node *node) const {
1.327 + return _names.at(node->_idx);
1.328 + }
1.329 +
1.330 + uint uf_live_range_id(uint lrg_id) const {
1.331 + return _uf_map.at(lrg_id);
1.332 + }
1.333 +
1.334 + void map(uint idx, uint lrg_id) {
1.335 + _names.at_put(idx, lrg_id);
1.336 + }
1.337 +
1.338 + void uf_map(uint dst_lrg_id, uint src_lrg_id) {
1.339 + _uf_map.at_put(dst_lrg_id, src_lrg_id);
1.340 + }
1.341 +
1.342 + void extend(uint idx, uint lrg_id) {
1.343 + _names.at_put_grow(idx, lrg_id);
1.344 + }
1.345 +
1.346 + void uf_extend(uint dst_lrg_id, uint src_lrg_id) {
1.347 + _uf_map.at_put_grow(dst_lrg_id, src_lrg_id);
1.348 + }
1.349 +
1.350 + LiveRangeMap(Arena* arena, uint unique)
1.351 + : _names(arena, unique, unique, 0)
1.352 + , _uf_map(arena, unique, unique, 0)
1.353 + , _max_lrg_id(0) {}
1.354 +
1.355 + uint find_id( const Node *n ) {
1.356 + uint retval = live_range_id(n);
1.357 + assert(retval == find(n),"Invalid node to lidx mapping");
1.358 + return retval;
1.359 + }
1.360 +
1.361 + // Reset the Union-Find map to identity
1.362 + void reset_uf_map(uint max_lrg_id);
1.363 +
1.364 + // Make all Nodes map directly to their final live range; no need for
1.365 + // the Union-Find mapping after this call.
1.366 + void compress_uf_map_for_nodes();
1.367 +
1.368 + uint find(uint lidx) {
1.369 + uint uf_lidx = _uf_map.at(lidx);
1.370 + return (uf_lidx == lidx) ? uf_lidx : find_compress(lidx);
1.371 + }
1.372 +
1.373 + // Convert a Node into a Live Range Index - a lidx
1.374 + uint find(const Node *node) {
1.375 + uint lidx = live_range_id(node);
1.376 + uint uf_lidx = _uf_map.at(lidx);
1.377 + return (uf_lidx == lidx) ? uf_lidx : find_compress(node);
1.378 + }
1.379 +
1.380 + // Like Find above, but no path compress, so bad asymptotic behavior
1.381 + uint find_const(uint lrg) const;
1.382 +
1.383 + // Like Find above, but no path compress, so bad asymptotic behavior
1.384 + uint find_const(const Node *node) const {
1.385 + if(node->_idx >= (uint)_names.length()) {
1.386 + return 0; // not mapped, usual for debug dump
1.387 + }
1.388 + return find_const(_names.at(node->_idx));
1.389 + }
1.390 +};
1.391 +
1.392 +//------------------------------Chaitin----------------------------------------
1.393 +// Briggs-Chaitin style allocation, mostly.
1.394 +class PhaseChaitin : public PhaseRegAlloc {
1.395 + friend class VMStructs;
1.396 +
1.397 + int _trip_cnt;
1.398 + int _alternate;
1.399 +
1.400 + LRG &lrgs(uint idx) const { return _ifg->lrgs(idx); }
1.401 + PhaseLive *_live; // Liveness, used in the interference graph
1.402 + PhaseIFG *_ifg; // Interference graph (for original chunk)
1.403 + Node_List **_lrg_nodes; // Array of node; lists for lrgs which spill
1.404 + VectorSet _spilled_once; // Nodes that have been spilled
1.405 + VectorSet _spilled_twice; // Nodes that have been spilled twice
1.406 +
1.407 + // Combine the Live Range Indices for these 2 Nodes into a single live
1.408 + // range. Future requests for any Node in either live range will
1.409 + // return the live range index for the combined live range.
1.410 + void Union( const Node *src, const Node *dst );
1.411 +
1.412 + void new_lrg( const Node *x, uint lrg );
1.413 +
1.414 + // Compact live ranges, removing unused ones. Return new maxlrg.
1.415 + void compact();
1.416 +
1.417 + uint _lo_degree; // Head of lo-degree LRGs list
1.418 + uint _lo_stk_degree; // Head of lo-stk-degree LRGs list
1.419 + uint _hi_degree; // Head of hi-degree LRGs list
1.420 + uint _simplified; // Linked list head of simplified LRGs
1.421 +
1.422 + // Helper functions for Split()
1.423 + uint split_DEF( Node *def, Block *b, int loc, uint max, Node **Reachblock, Node **debug_defs, GrowableArray<uint> splits, int slidx );
1.424 + uint split_USE( Node *def, Block *b, Node *use, uint useidx, uint max, bool def_down, bool cisc_sp, GrowableArray<uint> splits, int slidx );
1.425 +
1.426 + //------------------------------clone_projs------------------------------------
1.427 + // After cloning some rematerialized instruction, clone any MachProj's that
1.428 + // follow it. Example: Intel zero is XOR, kills flags. Sparc FP constants
1.429 + // use G3 as an address temp.
1.430 + int clone_projs(Block* b, uint idx, Node* orig, Node* copy, uint& max_lrg_id);
1.431 +
1.432 + int clone_projs(Block* b, uint idx, Node* orig, Node* copy, LiveRangeMap& lrg_map) {
1.433 + uint max_lrg_id = lrg_map.max_lrg_id();
1.434 + int found_projs = clone_projs(b, idx, orig, copy, max_lrg_id);
1.435 + if (found_projs > 0) {
1.436 + // max_lrg_id is updated during call above
1.437 + lrg_map.set_max_lrg_id(max_lrg_id);
1.438 + }
1.439 + return found_projs;
1.440 + }
1.441 +
1.442 + Node *split_Rematerialize(Node *def, Block *b, uint insidx, uint &maxlrg, GrowableArray<uint> splits,
1.443 + int slidx, uint *lrg2reach, Node **Reachblock, bool walkThru);
1.444 + // True if lidx is used before any real register is def'd in the block
1.445 + bool prompt_use( Block *b, uint lidx );
1.446 + Node *get_spillcopy_wide( Node *def, Node *use, uint uidx );
1.447 + // Insert the spill at chosen location. Skip over any intervening Proj's or
1.448 + // Phis. Skip over a CatchNode and projs, inserting in the fall-through block
1.449 + // instead. Update high-pressure indices. Create a new live range.
1.450 + void insert_proj( Block *b, uint i, Node *spill, uint maxlrg );
1.451 +
1.452 + bool is_high_pressure( Block *b, LRG *lrg, uint insidx );
1.453 +
1.454 + uint _oldphi; // Node index which separates pre-allocation nodes
1.455 +
1.456 + Block **_blks; // Array of blocks sorted by frequency for coalescing
1.457 +
1.458 + float _high_frequency_lrg; // Frequency at which LRG will be spilled for debug info
1.459 +
1.460 +#ifndef PRODUCT
1.461 + bool _trace_spilling;
1.462 +#endif
1.463 +
1.464 +public:
1.465 + PhaseChaitin( uint unique, PhaseCFG &cfg, Matcher &matcher );
1.466 + ~PhaseChaitin() {}
1.467 +
1.468 + LiveRangeMap _lrg_map;
1.469 +
1.470 + // Do all the real work of allocate
1.471 + void Register_Allocate();
1.472 +
1.473 + float high_frequency_lrg() const { return _high_frequency_lrg; }
1.474 +
1.475 +#ifndef PRODUCT
1.476 + bool trace_spilling() const { return _trace_spilling; }
1.477 +#endif
1.478 +
1.479 +private:
1.480 + // De-SSA the world. Assign registers to Nodes. Use the same register for
1.481 + // all inputs to a PhiNode, effectively coalescing live ranges. Insert
1.482 + // copies as needed.
1.483 + void de_ssa();
1.484 +
1.485 + // Add edge between reg and everything in the vector.
1.486 + // Same as _ifg->add_vector(reg,live) EXCEPT use the RegMask
1.487 + // information to trim the set of interferences. Return the
1.488 + // count of edges added.
1.489 + void interfere_with_live( uint reg, IndexSet *live );
1.490 + // Count register pressure for asserts
1.491 + uint count_int_pressure( IndexSet *liveout );
1.492 + uint count_float_pressure( IndexSet *liveout );
1.493 +
1.494 + // Build the interference graph using virtual registers only.
1.495 + // Used for aggressive coalescing.
1.496 + void build_ifg_virtual( );
1.497 +
1.498 + // Build the interference graph using physical registers when available.
1.499 + // That is, if 2 live ranges are simultaneously alive but in their
1.500 + // acceptable register sets do not overlap, then they do not interfere.
1.501 + uint build_ifg_physical( ResourceArea *a );
1.502 +
1.503 + // Gather LiveRanGe information, including register masks and base pointer/
1.504 + // derived pointer relationships.
1.505 + void gather_lrg_masks( bool mod_cisc_masks );
1.506 +
1.507 + // Force the bases of derived pointers to be alive at GC points.
1.508 + bool stretch_base_pointer_live_ranges( ResourceArea *a );
1.509 + // Helper to stretch above; recursively discover the base Node for
1.510 + // a given derived Node. Easy for AddP-related machine nodes, but
1.511 + // needs to be recursive for derived Phis.
1.512 + Node *find_base_for_derived( Node **derived_base_map, Node *derived, uint &maxlrg );
1.513 +
1.514 + // Set the was-lo-degree bit. Conservative coalescing should not change the
1.515 + // colorability of the graph. If any live range was of low-degree before
1.516 + // coalescing, it should Simplify. This call sets the was-lo-degree bit.
1.517 + void set_was_low();
1.518 +
1.519 + // Split live-ranges that must spill due to register conflicts (as opposed
1.520 + // to capacity spills). Typically these are things def'd in a register
1.521 + // and used on the stack or vice-versa.
1.522 + void pre_spill();
1.523 +
1.524 + // Init LRG caching of degree, numregs. Init lo_degree list.
1.525 + void cache_lrg_info( );
1.526 +
1.527 + // Simplify the IFG by removing LRGs of low degree with no copies
1.528 + void Pre_Simplify();
1.529 +
1.530 + // Simplify the IFG by removing LRGs of low degree
1.531 + void Simplify();
1.532 +
1.533 + // Select colors by re-inserting edges into the IFG.
1.534 + // Return TRUE if any spills occurred.
1.535 + uint Select( );
1.536 + // Helper function for select which allows biased coloring
1.537 + OptoReg::Name choose_color( LRG &lrg, int chunk );
1.538 + // Helper function which implements biasing heuristic
1.539 + OptoReg::Name bias_color( LRG &lrg, int chunk );
1.540 +
1.541 + // Split uncolorable live ranges
1.542 + // Return new number of live ranges
1.543 + uint Split(uint maxlrg, ResourceArea* split_arena);
1.544 +
1.545 + // Copy 'was_spilled'-edness from one Node to another.
1.546 + void copy_was_spilled( Node *src, Node *dst );
1.547 + // Set the 'spilled_once' or 'spilled_twice' flag on a node.
1.548 + void set_was_spilled( Node *n );
1.549 +
1.550 + // Convert ideal spill-nodes into machine loads & stores
1.551 + // Set C->failing when fixup spills could not complete, node limit exceeded.
1.552 + void fixup_spills();
1.553 +
1.554 + // Post-Allocation peephole copy removal
1.555 + void post_allocate_copy_removal();
1.556 + Node *skip_copies( Node *c );
1.557 + // Replace the old node with the current live version of that value
1.558 + // and yank the old value if it's dead.
1.559 + int replace_and_yank_if_dead( Node *old, OptoReg::Name nreg,
1.560 + Block *current_block, Node_List& value, Node_List& regnd ) {
1.561 + Node* v = regnd[nreg];
1.562 + assert(v->outcnt() != 0, "no dead values");
1.563 + old->replace_by(v);
1.564 + return yank_if_dead(old, current_block, &value, ®nd);
1.565 + }
1.566 +
1.567 + int yank_if_dead( Node *old, Block *current_block, Node_List *value, Node_List *regnd ) {
1.568 + return yank_if_dead_recurse(old, old, current_block, value, regnd);
1.569 + }
1.570 + int yank_if_dead_recurse(Node *old, Node *orig_old, Block *current_block,
1.571 + Node_List *value, Node_List *regnd);
1.572 + int yank( Node *old, Block *current_block, Node_List *value, Node_List *regnd );
1.573 + int elide_copy( Node *n, int k, Block *current_block, Node_List &value, Node_List ®nd, bool can_change_regs );
1.574 + int use_prior_register( Node *copy, uint idx, Node *def, Block *current_block, Node_List &value, Node_List ®nd );
1.575 + bool may_be_copy_of_callee( Node *def ) const;
1.576 +
1.577 + // If nreg already contains the same constant as val then eliminate it
1.578 + bool eliminate_copy_of_constant(Node* val, Node* n,
1.579 + Block *current_block, Node_List& value, Node_List ®nd,
1.580 + OptoReg::Name nreg, OptoReg::Name nreg2);
1.581 + // Extend the node to LRG mapping
1.582 + void add_reference( const Node *node, const Node *old_node);
1.583 +
1.584 +private:
1.585 +
1.586 + static int _final_loads, _final_stores, _final_copies, _final_memoves;
1.587 + static double _final_load_cost, _final_store_cost, _final_copy_cost, _final_memove_cost;
1.588 + static int _conserv_coalesce, _conserv_coalesce_pair;
1.589 + static int _conserv_coalesce_trie, _conserv_coalesce_quad;
1.590 + static int _post_alloc;
1.591 + static int _lost_opp_pp_coalesce, _lost_opp_cflow_coalesce;
1.592 + static int _used_cisc_instructions, _unused_cisc_instructions;
1.593 + static int _allocator_attempts, _allocator_successes;
1.594 +
1.595 +#ifndef PRODUCT
1.596 + static uint _high_pressure, _low_pressure;
1.597 +
1.598 + void dump() const;
1.599 + void dump( const Node *n ) const;
1.600 + void dump( const Block * b ) const;
1.601 + void dump_degree_lists() const;
1.602 + void dump_simplified() const;
1.603 + void dump_lrg( uint lidx, bool defs_only) const;
1.604 + void dump_lrg( uint lidx) const {
1.605 + // dump defs and uses by default
1.606 + dump_lrg(lidx, false);
1.607 + }
1.608 + void dump_bb( uint pre_order ) const;
1.609 +
1.610 + // Verify that base pointers and derived pointers are still sane
1.611 + void verify_base_ptrs( ResourceArea *a ) const;
1.612 +
1.613 + void verify( ResourceArea *a, bool verify_ifg = false ) const;
1.614 +
1.615 + void dump_for_spill_split_recycle() const;
1.616 +
1.617 +public:
1.618 + void dump_frame() const;
1.619 + char *dump_register( const Node *n, char *buf ) const;
1.620 +private:
1.621 + static void print_chaitin_statistics();
1.622 +#endif
1.623 + friend class PhaseCoalesce;
1.624 + friend class PhaseAggressiveCoalesce;
1.625 + friend class PhaseConservativeCoalesce;
1.626 +};
1.627 +
1.628 +#endif // SHARE_VM_OPTO_CHAITIN_HPP
