superword.hpp 19.7 KB
Newer Older
D
duke 已提交
1
/*
2
 * Copyright (c) 2007, 2013, Oracle and/or its affiliates. All rights reserved.
D
duke 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
19 20 21
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
D
duke 已提交
22 23
 */

24 25 26 27 28 29 30 31 32 33
#ifndef SHARE_VM_OPTO_SUPERWORD_HPP
#define SHARE_VM_OPTO_SUPERWORD_HPP

#include "opto/connode.hpp"
#include "opto/loopnode.hpp"
#include "opto/node.hpp"
#include "opto/phaseX.hpp"
#include "opto/vectornode.hpp"
#include "utilities/growableArray.hpp"

D
duke 已提交
34 35 36 37 38 39 40 41 42 43
//
//                  S U P E R W O R D   T R A N S F O R M
//
// SuperWords are short, fixed length vectors.
//
// Algorithm from:
//
// Exploiting SuperWord Level Parallelism with
//   Multimedia Instruction Sets
// by
44
//   Samuel Larsen and Saman Amarasinghe
D
duke 已提交
45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203
//   MIT Laboratory for Computer Science
// date
//   May 2000
// published in
//   ACM SIGPLAN Notices
//   Proceedings of ACM PLDI '00,  Volume 35 Issue 5
//
// Definition 3.1 A Pack is an n-tuple, <s1, ...,sn>, where
// s1,...,sn are independent isomorphic statements in a basic
// block.
//
// Definition 3.2 A PackSet is a set of Packs.
//
// Definition 3.3 A Pair is a Pack of size two, where the
// first statement is considered the left element, and the
// second statement is considered the right element.

class SWPointer;
class OrderedPair;

// ========================= Dependence Graph =====================

class DepMem;

//------------------------------DepEdge---------------------------
// An edge in the dependence graph.  The edges incident to a dependence
// node are threaded through _next_in for incoming edges and _next_out
// for outgoing edges.
class DepEdge : public ResourceObj {
 protected:
  DepMem* _pred;
  DepMem* _succ;
  DepEdge* _next_in;   // list of in edges, null terminated
  DepEdge* _next_out;  // list of out edges, null terminated

 public:
  DepEdge(DepMem* pred, DepMem* succ, DepEdge* next_in, DepEdge* next_out) :
    _pred(pred), _succ(succ), _next_in(next_in), _next_out(next_out) {}

  DepEdge* next_in()  { return _next_in; }
  DepEdge* next_out() { return _next_out; }
  DepMem*  pred()     { return _pred; }
  DepMem*  succ()     { return _succ; }

  void print();
};

//------------------------------DepMem---------------------------
// A node in the dependence graph.  _in_head starts the threaded list of
// incoming edges, and _out_head starts the list of outgoing edges.
class DepMem : public ResourceObj {
 protected:
  Node*    _node;     // Corresponding ideal node
  DepEdge* _in_head;  // Head of list of in edges, null terminated
  DepEdge* _out_head; // Head of list of out edges, null terminated

 public:
  DepMem(Node* node) : _node(node), _in_head(NULL), _out_head(NULL) {}

  Node*    node()                { return _node;     }
  DepEdge* in_head()             { return _in_head;  }
  DepEdge* out_head()            { return _out_head; }
  void set_in_head(DepEdge* hd)  { _in_head = hd;    }
  void set_out_head(DepEdge* hd) { _out_head = hd;   }

  int in_cnt();  // Incoming edge count
  int out_cnt(); // Outgoing edge count

  void print();
};

//------------------------------DepGraph---------------------------
class DepGraph VALUE_OBJ_CLASS_SPEC {
 protected:
  Arena* _arena;
  GrowableArray<DepMem*> _map;
  DepMem* _root;
  DepMem* _tail;

 public:
  DepGraph(Arena* a) : _arena(a), _map(a, 8,  0, NULL) {
    _root = new (_arena) DepMem(NULL);
    _tail = new (_arena) DepMem(NULL);
  }

  DepMem* root() { return _root; }
  DepMem* tail() { return _tail; }

  // Return dependence node corresponding to an ideal node
  DepMem* dep(Node* node) { return _map.at(node->_idx); }

  // Make a new dependence graph node for an ideal node.
  DepMem* make_node(Node* node);

  // Make a new dependence graph edge dprec->dsucc
  DepEdge* make_edge(DepMem* dpred, DepMem* dsucc);

  DepEdge* make_edge(Node* pred,   Node* succ)   { return make_edge(dep(pred), dep(succ)); }
  DepEdge* make_edge(DepMem* pred, Node* succ)   { return make_edge(pred,      dep(succ)); }
  DepEdge* make_edge(Node* pred,   DepMem* succ) { return make_edge(dep(pred), succ);      }

  void init() { _map.clear(); } // initialize

  void print(Node* n)   { dep(n)->print(); }
  void print(DepMem* d) { d->print(); }
};

//------------------------------DepPreds---------------------------
// Iterator over predecessors in the dependence graph and
// non-memory-graph inputs of ideal nodes.
class DepPreds : public StackObj {
private:
  Node*    _n;
  int      _next_idx, _end_idx;
  DepEdge* _dep_next;
  Node*    _current;
  bool     _done;

public:
  DepPreds(Node* n, DepGraph& dg);
  Node* current() { return _current; }
  bool  done()    { return _done; }
  void  next();
};

//------------------------------DepSuccs---------------------------
// Iterator over successors in the dependence graph and
// non-memory-graph outputs of ideal nodes.
class DepSuccs : public StackObj {
private:
  Node*    _n;
  int      _next_idx, _end_idx;
  DepEdge* _dep_next;
  Node*    _current;
  bool     _done;

public:
  DepSuccs(Node* n, DepGraph& dg);
  Node* current() { return _current; }
  bool  done()    { return _done; }
  void  next();
};


// ========================= SuperWord =====================

// -----------------------------SWNodeInfo---------------------------------
// Per node info needed by SuperWord
class SWNodeInfo VALUE_OBJ_CLASS_SPEC {
 public:
  int         _alignment; // memory alignment for a node
  int         _depth;     // Max expression (DAG) depth from block start
  const Type* _velt_type; // vector element type
  Node_List*  _my_pack;   // pack containing this node

  SWNodeInfo() : _alignment(-1), _depth(0), _velt_type(NULL), _my_pack(NULL) {}
  static const SWNodeInfo initial;
};

204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228
// JVMCI: OrderedPair is moved up to deal with compilation issues on Windows
//------------------------------OrderedPair---------------------------
// Ordered pair of Node*.
class OrderedPair VALUE_OBJ_CLASS_SPEC {
 protected:
  Node* _p1;
  Node* _p2;
 public:
  OrderedPair() : _p1(NULL), _p2(NULL) {}
  OrderedPair(Node* p1, Node* p2) {
    if (p1->_idx < p2->_idx) {
      _p1 = p1; _p2 = p2;
    } else {
      _p1 = p2; _p2 = p1;
    }
  }

  bool operator==(const OrderedPair &rhs) {
    return _p1 == rhs._p1 && _p2 == rhs._p2;
  }
  void print() { tty->print("  (%d, %d)", _p1->_idx, _p2->_idx); }

  static const OrderedPair initial;
};

D
duke 已提交
229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291
// -----------------------------SuperWord---------------------------------
// Transforms scalar operations into packed (superword) operations.
class SuperWord : public ResourceObj {
 private:
  PhaseIdealLoop* _phase;
  Arena*          _arena;
  PhaseIterGVN   &_igvn;

  enum consts { top_align = -1, bottom_align = -666 };

  GrowableArray<Node_List*> _packset;    // Packs for the current block

  GrowableArray<int> _bb_idx;            // Map from Node _idx to index within block

  GrowableArray<Node*> _block;           // Nodes in current block
  GrowableArray<Node*> _data_entry;      // Nodes with all inputs from outside
  GrowableArray<Node*> _mem_slice_head;  // Memory slice head nodes
  GrowableArray<Node*> _mem_slice_tail;  // Memory slice tail nodes

  GrowableArray<SWNodeInfo> _node_info;  // Info needed per node

  MemNode* _align_to_ref;                // Memory reference that pre-loop will align to

  GrowableArray<OrderedPair> _disjoint_ptrs; // runtime disambiguated pointer pairs

  DepGraph _dg; // Dependence graph

  // Scratch pads
  VectorSet    _visited;       // Visited set
  VectorSet    _post_visited;  // Post-visited set
  Node_Stack   _n_idx_list;    // List of (node,index) pairs
  GrowableArray<Node*> _nlist; // List of nodes
  GrowableArray<Node*> _stk;   // Stack of nodes

 public:
  SuperWord(PhaseIdealLoop* phase);

  void transform_loop(IdealLoopTree* lpt);

  // Accessors for SWPointer
  PhaseIdealLoop* phase()          { return _phase; }
  IdealLoopTree* lpt()             { return _lpt; }
  PhiNode* iv()                    { return _iv; }

 private:
  IdealLoopTree* _lpt;             // Current loop tree node
  LoopNode*      _lp;              // Current LoopNode
  Node*          _bb;              // Current basic block
  PhiNode*       _iv;              // Induction var

  // Accessors
  Arena* arena()                   { return _arena; }

  Node* bb()                       { return _bb; }
  void  set_bb(Node* bb)           { _bb = bb; }

  void set_lpt(IdealLoopTree* lpt) { _lpt = lpt; }

  LoopNode* lp()                   { return _lp; }
  void      set_lp(LoopNode* lp)   { _lp = lp;
                                     _iv = lp->as_CountedLoop()->phi()->as_Phi(); }
  int      iv_stride()             { return lp()->as_CountedLoop()->stride_con(); }

292 293 294 295 296 297 298
  int vector_width(Node* n) {
    BasicType bt = velt_basic_type(n);
    return MIN2(ABS(iv_stride()), Matcher::max_vector_size(bt));
  }
  int vector_width_in_bytes(Node* n) {
    BasicType bt = velt_basic_type(n);
    return vector_width(n)*type2aelembytes(bt);
299
  }
D
duke 已提交
300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331
  MemNode* align_to_ref()            { return _align_to_ref; }
  void  set_align_to_ref(MemNode* m) { _align_to_ref = m; }

  Node* ctrl(Node* n) const { return _phase->has_ctrl(n) ? _phase->get_ctrl(n) : n; }

  // block accessors
  bool in_bb(Node* n)      { return n != NULL && n->outcnt() > 0 && ctrl(n) == _bb; }
  int  bb_idx(Node* n)     { assert(in_bb(n), "must be"); return _bb_idx.at(n->_idx); }
  void set_bb_idx(Node* n, int i) { _bb_idx.at_put_grow(n->_idx, i); }

  // visited set accessors
  void visited_clear()           { _visited.Clear(); }
  void visited_set(Node* n)      { return _visited.set(bb_idx(n)); }
  int visited_test(Node* n)      { return _visited.test(bb_idx(n)); }
  int visited_test_set(Node* n)  { return _visited.test_set(bb_idx(n)); }
  void post_visited_clear()      { _post_visited.Clear(); }
  void post_visited_set(Node* n) { return _post_visited.set(bb_idx(n)); }
  int post_visited_test(Node* n) { return _post_visited.test(bb_idx(n)); }

  // Ensure node_info contains element "i"
  void grow_node_info(int i) { if (i >= _node_info.length()) _node_info.at_put_grow(i, SWNodeInfo::initial); }

  // memory alignment for a node
  int alignment(Node* n)                     { return _node_info.adr_at(bb_idx(n))->_alignment; }
  void set_alignment(Node* n, int a)         { int i = bb_idx(n); grow_node_info(i); _node_info.adr_at(i)->_alignment = a; }

  // Max expression (DAG) depth from beginning of the block for each node
  int depth(Node* n)                         { return _node_info.adr_at(bb_idx(n))->_depth; }
  void set_depth(Node* n, int d)             { int i = bb_idx(n); grow_node_info(i); _node_info.adr_at(i)->_depth = d; }

  // vector element type
  const Type* velt_type(Node* n)             { return _node_info.adr_at(bb_idx(n))->_velt_type; }
332
  BasicType velt_basic_type(Node* n)         { return velt_type(n)->array_element_basic_type(); }
D
duke 已提交
333
  void set_velt_type(Node* n, const Type* t) { int i = bb_idx(n); grow_node_info(i); _node_info.adr_at(i)->_velt_type = t; }
334
  bool same_velt_type(Node* n1, Node* n2);
D
duke 已提交
335 336 337 338 339 340 341 342 343 344 345 346

  // my_pack
  Node_List* my_pack(Node* n)                { return !in_bb(n) ? NULL : _node_info.adr_at(bb_idx(n))->_my_pack; }
  void set_my_pack(Node* n, Node_List* p)    { int i = bb_idx(n); grow_node_info(i); _node_info.adr_at(i)->_my_pack = p; }

  // methods

  // Extract the superword level parallelism
  void SLP_extract();
  // Find the adjacent memory references and create pack pairs for them.
  void find_adjacent_refs();
  // Find a memory reference to align the loop induction variable to.
347 348 349
  MemNode* find_align_to_ref(Node_List &memops);
  // Calculate loop's iv adjustment for this memory ops.
  int get_iv_adjustment(MemNode* mem);
D
duke 已提交
350 351 352 353 354 355
  // Can the preloop align the reference to position zero in the vector?
  bool ref_is_alignable(SWPointer& p);
  // Construct dependency graph.
  void dependence_graph();
  // Return a memory slice (node list) in predecessor order starting at "start"
  void mem_slice_preds(Node* start, Node* stop, GrowableArray<Node*> &preds);
T
twisti 已提交
356
  // Can s1 and s2 be in a pack with s1 immediately preceding s2 and  s1 aligned at "align"
D
duke 已提交
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
  bool stmts_can_pack(Node* s1, Node* s2, int align);
  // Does s exist in a pack at position pos?
  bool exists_at(Node* s, uint pos);
  // Is s1 immediately before s2 in memory?
  bool are_adjacent_refs(Node* s1, Node* s2);
  // Are s1 and s2 similar?
  bool isomorphic(Node* s1, Node* s2);
  // Is there no data path from s1 to s2 or s2 to s1?
  bool independent(Node* s1, Node* s2);
  // Helper for independent
  bool independent_path(Node* shallow, Node* deep, uint dp=0);
  void set_alignment(Node* s1, Node* s2, int align);
  int data_size(Node* s);
  // Extend packset by following use->def and def->use links from pack members.
  void extend_packlist();
  // Extend the packset by visiting operand definitions of nodes in pack p
  bool follow_use_defs(Node_List* p);
  // Extend the packset by visiting uses of nodes in pack p
  bool follow_def_uses(Node_List* p);
  // Estimate the savings from executing s1 and s2 as a pack
  int est_savings(Node* s1, Node* s2);
  int adjacent_profit(Node* s1, Node* s2);
  int pack_cost(int ct);
  int unpack_cost(int ct);
  // Combine packs A and B with A.last == B.first into A.first..,A.last,B.second,..B.last
  void combine_packs();
  // Construct the map from nodes to packs.
  void construct_my_pack_map();
  // Remove packs that are not implemented or not profitable.
  void filter_packs();
  // Adjust the memory graph for the packed operations
  void schedule();
C
cfang 已提交
389 390 391 392 393
  // Remove "current" from its current position in the memory graph and insert
  // it after the appropriate insert points (lip or uip);
  void remove_and_insert(MemNode *current, MemNode *prev, MemNode *lip, Node *uip, Unique_Node_List &schd_before);
  // Within a store pack, schedule stores together by moving out the sandwiched memory ops according
  // to dependence info; and within a load pack, move loads down to the last executed load.
D
duke 已提交
394 395 396 397
  void co_locate_pack(Node_List* p);
  // Convert packs into vector node operations
  void output();
  // Create a vector operand for the nodes in pack p for operand: in(opd_idx)
K
kvn 已提交
398
  Node* vector_opd(Node_List* p, int opd_idx);
D
duke 已提交
399 400 401 402 403 404 405 406 407
  // Can code be generated for pack p?
  bool implemented(Node_List* p);
  // For pack p, are all operands and all uses (with in the block) vector?
  bool profitable(Node_List* p);
  // If a use of pack p is not a vector use, then replace the use with an extract operation.
  void insert_extracts(Node_List* p);
  // Is use->in(u_idx) a vector use?
  bool is_vector_use(Node* use, int u_idx);
  // Construct reverse postorder list of block members
408
  bool construct_bb();
D
duke 已提交
409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426
  // Initialize per node info
  void initialize_bb();
  // Insert n into block after pos
  void bb_insert_after(Node* n, int pos);
  // Compute max depth for expressions from beginning of block
  void compute_max_depth();
  // Compute necessary vector element type for expressions
  void compute_vector_element_type();
  // Are s1 and s2 in a pack pair and ordered as s1,s2?
  bool in_packset(Node* s1, Node* s2);
  // Is s in pack p?
  Node_List* in_pack(Node* s, Node_List* p);
  // Remove the pack at position pos in the packset
  void remove_pack_at(int pos);
  // Return the node executed first in pack p.
  Node* executed_first(Node_List* p);
  // Return the node executed last in pack p.
  Node* executed_last(Node_List* p);
427
  static LoadNode::ControlDependency control_dependency(Node_List* p);
D
duke 已提交
428
  // Alignment within a vector memory reference
K
kvn 已提交
429
  int memory_alignment(MemNode* s, int iv_adjust);
D
duke 已提交
430 431 432
  // (Start, end] half-open range defining which operands are vector
  void vector_opd_range(Node* n, uint* start, uint* end);
  // Smallest type containing range of values
433
  const Type* container_type(Node* n);
D
duke 已提交
434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451
  // Adjust pre-loop limit so that in main loop, a load/store reference
  // to align_to_ref will be a position zero in the vector.
  void align_initial_loop_index(MemNode* align_to_ref);
  // Find pre loop end from main loop.  Returns null if none.
  CountedLoopEndNode* get_pre_loop_end(CountedLoopNode *cl);
  // Is the use of d1 in u1 at the same operand position as d2 in u2?
  bool opnd_positions_match(Node* d1, Node* u1, Node* d2, Node* u2);
  void init();

  // print methods
  void print_packset();
  void print_pack(Node_List* p);
  void print_bb();
  void print_stmt(Node* s);
  char* blank(uint depth);
};


452

D
duke 已提交
453 454 455 456 457 458 459 460 461
//------------------------------SWPointer---------------------------
// Information about an address for dependence checking and vector alignment
class SWPointer VALUE_OBJ_CLASS_SPEC {
 protected:
  MemNode*   _mem;     // My memory reference node
  SuperWord* _slp;     // SuperWord class

  Node* _base;         // NULL if unsafe nonheap reference
  Node* _adr;          // address pointer
462
  jint  _scale;        // multiplier for iv (in bytes), 0 if no loop iv
D
duke 已提交
463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
  jint  _offset;       // constant offset (in bytes)
  Node* _invar;        // invariant offset (in bytes), NULL if none
  bool  _negate_invar; // if true then use: (0 - _invar)

  PhaseIdealLoop* phase() { return _slp->phase(); }
  IdealLoopTree*  lpt()   { return _slp->lpt(); }
  PhiNode*        iv()    { return _slp->iv();  } // Induction var

  bool invariant(Node* n) {
    Node *n_c = phase()->get_ctrl(n);
    return !lpt()->is_member(phase()->get_loop(n_c));
  }

  // Match: k*iv + offset
  bool scaled_iv_plus_offset(Node* n);
  // Match: k*iv where k is a constant that's not zero
  bool scaled_iv(Node* n);
  // Match: offset is (k [+/- invariant])
  bool offset_plus_k(Node* n, bool negate = false);

 public:
  enum CMP {
    Less          = 1,
    Greater       = 2,
    Equal         = 4,
    NotEqual      = (Less | Greater),
    NotComparable = (Less | Greater | Equal)
  };

  SWPointer(MemNode* mem, SuperWord* slp);
  // Following is used to create a temporary object during
  // the pattern match of an address expression.
  SWPointer(SWPointer* p);

  bool valid()  { return _adr != NULL; }
  bool has_iv() { return _scale != 0; }

  Node* base()            { return _base; }
  Node* adr()             { return _adr; }
502
  MemNode* mem()          { return _mem; }
D
duke 已提交
503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533
  int   scale_in_bytes()  { return _scale; }
  Node* invar()           { return _invar; }
  bool  negate_invar()    { return _negate_invar; }
  int   offset_in_bytes() { return _offset; }
  int   memory_size()     { return _mem->memory_size(); }

  // Comparable?
  int cmp(SWPointer& q) {
    if (valid() && q.valid() &&
        (_adr == q._adr || _base == _adr && q._base == q._adr) &&
        _scale == q._scale   &&
        _invar == q._invar   &&
        _negate_invar == q._negate_invar) {
      bool overlap = q._offset <   _offset +   memory_size() &&
                       _offset < q._offset + q.memory_size();
      return overlap ? Equal : (_offset < q._offset ? Less : Greater);
    } else {
      return NotComparable;
    }
  }

  bool not_equal(SWPointer& q)    { return not_equal(cmp(q)); }
  bool equal(SWPointer& q)        { return equal(cmp(q)); }
  bool comparable(SWPointer& q)   { return comparable(cmp(q)); }
  static bool not_equal(int cmp)  { return cmp <= NotEqual; }
  static bool equal(int cmp)      { return cmp == Equal; }
  static bool comparable(int cmp) { return cmp < NotComparable; }

  void print();
};

534
#endif // SHARE_VM_OPTO_SUPERWORD_HPP