| 1 | /* |
| 2 | * Copyright (c) 2007, 2019, Oracle and/or its affiliates. All rights reserved. |
| 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 4 | * |
| 5 | * This code is free software; you can redistribute it and/or modify it |
| 6 | * under the terms of the GNU General Public License version 2 only, as |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
| 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 12 | * version 2 for more details (a copy is included in the LICENSE file that |
| 13 | * accompanied this code). |
| 14 | * |
| 15 | * You should have received a copy of the GNU General Public License version |
| 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
| 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
| 18 | * |
| 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
| 20 | * or visit www.oracle.com if you need additional information or have any |
| 21 | * questions. |
| 22 | */ |
| 23 | |
| 24 | #ifndef SHARE_OPTO_SUPERWORD_HPP |
| 25 | #define SHARE_OPTO_SUPERWORD_HPP |
| 26 | |
| 27 | #include "opto/loopnode.hpp" |
| 28 | #include "opto/node.hpp" |
| 29 | #include "opto/phaseX.hpp" |
| 30 | #include "opto/vectornode.hpp" |
| 31 | #include "utilities/growableArray.hpp" |
| 32 | #include "libadt/dict.hpp" |
| 33 | |
| 34 | // |
| 35 | // S U P E R W O R D T R A N S F O R M |
| 36 | // |
| 37 | // SuperWords are short, fixed length vectors. |
| 38 | // |
| 39 | // Algorithm from: |
| 40 | // |
| 41 | // Exploiting SuperWord Level Parallelism with |
| 42 | // Multimedia Instruction Sets |
| 43 | // by |
| 44 | // Samuel Larsen and Saman Amarasinghe |
| 45 | // MIT Laboratory for Computer Science |
| 46 | // date |
| 47 | // May 2000 |
| 48 | // published in |
| 49 | // ACM SIGPLAN Notices |
| 50 | // Proceedings of ACM PLDI '00, Volume 35 Issue 5 |
| 51 | // |
| 52 | // Definition 3.1 A Pack is an n-tuple, <s1, ...,sn>, where |
| 53 | // s1,...,sn are independent isomorphic statements in a basic |
| 54 | // block. |
| 55 | // |
| 56 | // Definition 3.2 A PackSet is a set of Packs. |
| 57 | // |
| 58 | // Definition 3.3 A Pair is a Pack of size two, where the |
| 59 | // first statement is considered the left element, and the |
| 60 | // second statement is considered the right element. |
| 61 | |
| 62 | class SWPointer; |
| 63 | class OrderedPair; |
| 64 | |
| 65 | // ========================= Dependence Graph ===================== |
| 66 | |
| 67 | class DepMem; |
| 68 | |
| 69 | //------------------------------DepEdge--------------------------- |
| 70 | // An edge in the dependence graph. The edges incident to a dependence |
| 71 | // node are threaded through _next_in for incoming edges and _next_out |
| 72 | // for outgoing edges. |
| 73 | class DepEdge : public ResourceObj { |
| 74 | protected: |
| 75 | DepMem* _pred; |
| 76 | DepMem* _succ; |
| 77 | DepEdge* _next_in; // list of in edges, null terminated |
| 78 | DepEdge* _next_out; // list of out edges, null terminated |
| 79 | |
| 80 | public: |
| 81 | DepEdge(DepMem* pred, DepMem* succ, DepEdge* next_in, DepEdge* next_out) : |
| 82 | _pred(pred), _succ(succ), _next_in(next_in), _next_out(next_out) {} |
| 83 | |
| 84 | DepEdge* next_in() { return _next_in; } |
| 85 | DepEdge* next_out() { return _next_out; } |
| 86 | DepMem* pred() { return _pred; } |
| 87 | DepMem* succ() { return _succ; } |
| 88 | |
| 89 | void print(); |
| 90 | }; |
| 91 | |
| 92 | //------------------------------DepMem--------------------------- |
| 93 | // A node in the dependence graph. _in_head starts the threaded list of |
| 94 | // incoming edges, and _out_head starts the list of outgoing edges. |
| 95 | class DepMem : public ResourceObj { |
| 96 | protected: |
| 97 | Node* _node; // Corresponding ideal node |
| 98 | DepEdge* _in_head; // Head of list of in edges, null terminated |
| 99 | DepEdge* _out_head; // Head of list of out edges, null terminated |
| 100 | |
| 101 | public: |
| 102 | DepMem(Node* node) : _node(node), _in_head(NULL), _out_head(NULL) {} |
| 103 | |
| 104 | Node* node() { return _node; } |
| 105 | DepEdge* in_head() { return _in_head; } |
| 106 | DepEdge* out_head() { return _out_head; } |
| 107 | void set_in_head(DepEdge* hd) { _in_head = hd; } |
| 108 | void set_out_head(DepEdge* hd) { _out_head = hd; } |
| 109 | |
| 110 | int in_cnt(); // Incoming edge count |
| 111 | int out_cnt(); // Outgoing edge count |
| 112 | |
| 113 | void print(); |
| 114 | }; |
| 115 | |
| 116 | //------------------------------DepGraph--------------------------- |
| 117 | class DepGraph { |
| 118 | protected: |
| 119 | Arena* _arena; |
| 120 | GrowableArray<DepMem*> _map; |
| 121 | DepMem* _root; |
| 122 | DepMem* _tail; |
| 123 | |
| 124 | public: |
| 125 | DepGraph(Arena* a) : _arena(a), _map(a, 8, 0, NULL) { |
| 126 | _root = new (_arena) DepMem(NULL); |
| 127 | _tail = new (_arena) DepMem(NULL); |
| 128 | } |
| 129 | |
| 130 | DepMem* root() { return _root; } |
| 131 | DepMem* tail() { return _tail; } |
| 132 | |
| 133 | // Return dependence node corresponding to an ideal node |
| 134 | DepMem* dep(Node* node) { return _map.at(node->_idx); } |
| 135 | |
| 136 | // Make a new dependence graph node for an ideal node. |
| 137 | DepMem* make_node(Node* node); |
| 138 | |
| 139 | // Make a new dependence graph edge dprec->dsucc |
| 140 | DepEdge* make_edge(DepMem* dpred, DepMem* dsucc); |
| 141 | |
| 142 | DepEdge* make_edge(Node* pred, Node* succ) { return make_edge(dep(pred), dep(succ)); } |
| 143 | DepEdge* make_edge(DepMem* pred, Node* succ) { return make_edge(pred, dep(succ)); } |
| 144 | DepEdge* make_edge(Node* pred, DepMem* succ) { return make_edge(dep(pred), succ); } |
| 145 | |
| 146 | void init() { _map.clear(); } // initialize |
| 147 | |
| 148 | void print(Node* n) { dep(n)->print(); } |
| 149 | void print(DepMem* d) { d->print(); } |
| 150 | }; |
| 151 | |
| 152 | //------------------------------DepPreds--------------------------- |
| 153 | // Iterator over predecessors in the dependence graph and |
| 154 | // non-memory-graph inputs of ideal nodes. |
| 155 | class DepPreds : public StackObj { |
| 156 | private: |
| 157 | Node* _n; |
| 158 | int _next_idx, _end_idx; |
| 159 | DepEdge* _dep_next; |
| 160 | Node* _current; |
| 161 | bool _done; |
| 162 | |
| 163 | public: |
| 164 | DepPreds(Node* n, DepGraph& dg); |
| 165 | Node* current() { return _current; } |
| 166 | bool done() { return _done; } |
| 167 | void next(); |
| 168 | }; |
| 169 | |
| 170 | //------------------------------DepSuccs--------------------------- |
| 171 | // Iterator over successors in the dependence graph and |
| 172 | // non-memory-graph outputs of ideal nodes. |
| 173 | class DepSuccs : public StackObj { |
| 174 | private: |
| 175 | Node* _n; |
| 176 | int _next_idx, _end_idx; |
| 177 | DepEdge* _dep_next; |
| 178 | Node* _current; |
| 179 | bool _done; |
| 180 | |
| 181 | public: |
| 182 | DepSuccs(Node* n, DepGraph& dg); |
| 183 | Node* current() { return _current; } |
| 184 | bool done() { return _done; } |
| 185 | void next(); |
| 186 | }; |
| 187 | |
| 188 | |
| 189 | // ========================= SuperWord ===================== |
| 190 | |
| 191 | // -----------------------------SWNodeInfo--------------------------------- |
| 192 | // Per node info needed by SuperWord |
| 193 | class SWNodeInfo { |
| 194 | public: |
| 195 | int _alignment; // memory alignment for a node |
| 196 | int _depth; // Max expression (DAG) depth from block start |
| 197 | const Type* _velt_type; // vector element type |
| 198 | Node_List* _my_pack; // pack containing this node |
| 199 | |
| 200 | SWNodeInfo() : _alignment(-1), _depth(0), _velt_type(NULL), _my_pack(NULL) {} |
| 201 | static const SWNodeInfo initial; |
| 202 | }; |
| 203 | |
| 204 | class SuperWord; |
| 205 | class CMoveKit { |
| 206 | friend class SuperWord; |
| 207 | private: |
| 208 | SuperWord* _sw; |
| 209 | Dict* _dict; |
| 210 | CMoveKit(Arena* a, SuperWord* sw) : _sw(sw) {_dict = new Dict(cmpkey, hashkey, a);} |
| 211 | void* _2p(Node* key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy |
| 212 | Dict* dict() const { return _dict; } |
| 213 | void map(Node* key, Node_List* val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed" ); _dict->Insert(_2p(key), (void*)val); } |
| 214 | void unmap(Node* key) { _dict->Delete(_2p(key)); } |
| 215 | Node_List* pack(Node* key) const { return (Node_List*)_dict->operator[](_2p(key)); } |
| 216 | Node* is_Bool_candidate(Node* nd) const; // if it is the right candidate return corresponding CMove* , |
| 217 | Node* is_CmpD_candidate(Node* nd) const; // otherwise return NULL |
| 218 | Node_List* make_cmovevd_pack(Node_List* cmovd_pk); |
| 219 | bool test_cmpd_pack(Node_List* cmpd_pk, Node_List* cmovd_pk); |
| 220 | };//class CMoveKit |
| 221 | |
| 222 | // JVMCI: OrderedPair is moved up to deal with compilation issues on Windows |
| 223 | //------------------------------OrderedPair--------------------------- |
| 224 | // Ordered pair of Node*. |
| 225 | class OrderedPair { |
| 226 | protected: |
| 227 | Node* _p1; |
| 228 | Node* _p2; |
| 229 | public: |
| 230 | OrderedPair() : _p1(NULL), _p2(NULL) {} |
| 231 | OrderedPair(Node* p1, Node* p2) { |
| 232 | if (p1->_idx < p2->_idx) { |
| 233 | _p1 = p1; _p2 = p2; |
| 234 | } else { |
| 235 | _p1 = p2; _p2 = p1; |
| 236 | } |
| 237 | } |
| 238 | |
| 239 | bool operator==(const OrderedPair &rhs) { |
| 240 | return _p1 == rhs._p1 && _p2 == rhs._p2; |
| 241 | } |
| 242 | void print() { tty->print(" (%d, %d)" , _p1->_idx, _p2->_idx); } |
| 243 | |
| 244 | static const OrderedPair initial; |
| 245 | }; |
| 246 | |
| 247 | // -----------------------------SuperWord--------------------------------- |
| 248 | // Transforms scalar operations into packed (superword) operations. |
| 249 | class SuperWord : public ResourceObj { |
| 250 | friend class SWPointer; |
| 251 | friend class CMoveKit; |
| 252 | private: |
| 253 | PhaseIdealLoop* _phase; |
| 254 | Arena* _arena; |
| 255 | PhaseIterGVN &_igvn; |
| 256 | |
| 257 | enum consts { top_align = -1, bottom_align = -666 }; |
| 258 | |
| 259 | GrowableArray<Node_List*> _packset; // Packs for the current block |
| 260 | |
| 261 | GrowableArray<int> _bb_idx; // Map from Node _idx to index within block |
| 262 | |
| 263 | GrowableArray<Node*> _block; // Nodes in current block |
| 264 | GrowableArray<Node*> _post_block; // Nodes in post loop block |
| 265 | GrowableArray<Node*> _data_entry; // Nodes with all inputs from outside |
| 266 | GrowableArray<Node*> _mem_slice_head; // Memory slice head nodes |
| 267 | GrowableArray<Node*> _mem_slice_tail; // Memory slice tail nodes |
| 268 | GrowableArray<Node*> _iteration_first; // nodes in the generation that has deps from phi |
| 269 | GrowableArray<Node*> _iteration_last; // nodes in the generation that has deps to phi |
| 270 | GrowableArray<SWNodeInfo> _node_info; // Info needed per node |
| 271 | CloneMap& _clone_map; // map of nodes created in cloning |
| 272 | CMoveKit _cmovev_kit; // support for vectorization of CMov |
| 273 | MemNode* _align_to_ref; // Memory reference that pre-loop will align to |
| 274 | |
| 275 | GrowableArray<OrderedPair> _disjoint_ptrs; // runtime disambiguated pointer pairs |
| 276 | |
| 277 | DepGraph _dg; // Dependence graph |
| 278 | |
| 279 | // Scratch pads |
| 280 | VectorSet _visited; // Visited set |
| 281 | VectorSet _post_visited; // Post-visited set |
| 282 | Node_Stack _n_idx_list; // List of (node,index) pairs |
| 283 | GrowableArray<Node*> _nlist; // List of nodes |
| 284 | GrowableArray<Node*> _stk; // Stack of nodes |
| 285 | |
| 286 | public: |
| 287 | SuperWord(PhaseIdealLoop* phase); |
| 288 | |
| 289 | void transform_loop(IdealLoopTree* lpt, bool do_optimization); |
| 290 | |
| 291 | void unrolling_analysis(int &local_loop_unroll_factor); |
| 292 | |
| 293 | // Accessors for SWPointer |
| 294 | PhaseIdealLoop* phase() { return _phase; } |
| 295 | IdealLoopTree* lpt() { return _lpt; } |
| 296 | PhiNode* iv() { return _iv; } |
| 297 | |
| 298 | bool early_return() { return _early_return; } |
| 299 | |
| 300 | #ifndef PRODUCT |
| 301 | bool is_debug() { return _vector_loop_debug > 0; } |
| 302 | bool is_trace_alignment() { return (_vector_loop_debug & 2) > 0; } |
| 303 | bool is_trace_mem_slice() { return (_vector_loop_debug & 4) > 0; } |
| 304 | bool is_trace_loop() { return (_vector_loop_debug & 8) > 0; } |
| 305 | bool is_trace_adjacent() { return (_vector_loop_debug & 16) > 0; } |
| 306 | bool is_trace_cmov() { return (_vector_loop_debug & 32) > 0; } |
| 307 | bool is_trace_loop_reverse() { return (_vector_loop_debug & 64) > 0; } |
| 308 | #endif |
| 309 | bool do_vector_loop() { return _do_vector_loop; } |
| 310 | bool do_reserve_copy() { return _do_reserve_copy; } |
| 311 | private: |
| 312 | IdealLoopTree* _lpt; // Current loop tree node |
| 313 | LoopNode* _lp; // Current LoopNode |
| 314 | Node* _bb; // Current basic block |
| 315 | PhiNode* _iv; // Induction var |
| 316 | bool _race_possible; // In cases where SDMU is true |
| 317 | bool _early_return; // True if we do not initialize |
| 318 | bool _do_vector_loop; // whether to do vectorization/simd style |
| 319 | bool _do_reserve_copy; // do reserve copy of the graph(loop) before final modification in output |
| 320 | int _num_work_vecs; // Number of non memory vector operations |
| 321 | int _num_reductions; // Number of reduction expressions applied |
| 322 | int _ii_first; // generation with direct deps from mem phi |
| 323 | int _ii_last; // generation with direct deps to mem phi |
| 324 | GrowableArray<int> _ii_order; |
| 325 | #ifndef PRODUCT |
| 326 | uintx _vector_loop_debug; // provide more printing in debug mode |
| 327 | #endif |
| 328 | |
| 329 | // Accessors |
| 330 | Arena* arena() { return _arena; } |
| 331 | |
| 332 | Node* bb() { return _bb; } |
| 333 | void set_bb(Node* bb) { _bb = bb; } |
| 334 | |
| 335 | void set_lpt(IdealLoopTree* lpt) { _lpt = lpt; } |
| 336 | |
| 337 | LoopNode* lp() { return _lp; } |
| 338 | void set_lp(LoopNode* lp) { _lp = lp; |
| 339 | _iv = lp->as_CountedLoop()->phi()->as_Phi(); } |
| 340 | int iv_stride() { return lp()->as_CountedLoop()->stride_con(); } |
| 341 | |
| 342 | int vector_width(Node* n) { |
| 343 | BasicType bt = velt_basic_type(n); |
| 344 | return MIN2(ABS(iv_stride()), Matcher::max_vector_size(bt)); |
| 345 | } |
| 346 | int vector_width_in_bytes(Node* n) { |
| 347 | BasicType bt = velt_basic_type(n); |
| 348 | return vector_width(n)*type2aelembytes(bt); |
| 349 | } |
| 350 | int get_vw_bytes_special(MemNode* s); |
| 351 | MemNode* align_to_ref() { return _align_to_ref; } |
| 352 | void set_align_to_ref(MemNode* m) { _align_to_ref = m; } |
| 353 | |
| 354 | Node* ctrl(Node* n) const { return _phase->has_ctrl(n) ? _phase->get_ctrl(n) : n; } |
| 355 | |
| 356 | // block accessors |
| 357 | bool in_bb(Node* n) { return n != NULL && n->outcnt() > 0 && ctrl(n) == _bb; } |
| 358 | int bb_idx(Node* n) { assert(in_bb(n), "must be" ); return _bb_idx.at(n->_idx); } |
| 359 | void set_bb_idx(Node* n, int i) { _bb_idx.at_put_grow(n->_idx, i); } |
| 360 | |
| 361 | // visited set accessors |
| 362 | void visited_clear() { _visited.Clear(); } |
| 363 | void visited_set(Node* n) { return _visited.set(bb_idx(n)); } |
| 364 | int visited_test(Node* n) { return _visited.test(bb_idx(n)); } |
| 365 | int visited_test_set(Node* n) { return _visited.test_set(bb_idx(n)); } |
| 366 | void post_visited_clear() { _post_visited.Clear(); } |
| 367 | void post_visited_set(Node* n) { return _post_visited.set(bb_idx(n)); } |
| 368 | int post_visited_test(Node* n) { return _post_visited.test(bb_idx(n)); } |
| 369 | |
| 370 | // Ensure node_info contains element "i" |
| 371 | void grow_node_info(int i) { if (i >= _node_info.length()) _node_info.at_put_grow(i, SWNodeInfo::initial); } |
| 372 | |
| 373 | // memory alignment for a node |
| 374 | int alignment(Node* n) { return _node_info.adr_at(bb_idx(n))->_alignment; } |
| 375 | void set_alignment(Node* n, int a) { int i = bb_idx(n); grow_node_info(i); _node_info.adr_at(i)->_alignment = a; } |
| 376 | |
| 377 | // Max expression (DAG) depth from beginning of the block for each node |
| 378 | int depth(Node* n) { return _node_info.adr_at(bb_idx(n))->_depth; } |
| 379 | void set_depth(Node* n, int d) { int i = bb_idx(n); grow_node_info(i); _node_info.adr_at(i)->_depth = d; } |
| 380 | |
| 381 | // vector element type |
| 382 | const Type* velt_type(Node* n) { return _node_info.adr_at(bb_idx(n))->_velt_type; } |
| 383 | BasicType velt_basic_type(Node* n) { return velt_type(n)->array_element_basic_type(); } |
| 384 | 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; } |
| 385 | bool same_velt_type(Node* n1, Node* n2); |
| 386 | |
| 387 | // my_pack |
| 388 | Node_List* my_pack(Node* n) { return !in_bb(n) ? NULL : _node_info.adr_at(bb_idx(n))->_my_pack; } |
| 389 | 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; } |
| 390 | // is pack good for converting into one vector node replacing 12 nodes of Cmp, Bool, CMov |
| 391 | bool is_cmov_pack(Node_List* p); |
| 392 | bool is_cmov_pack_internal_node(Node_List* p, Node* nd) { return is_cmov_pack(p) && !nd->is_CMove(); } |
| 393 | // For pack p, are all idx operands the same? |
| 394 | bool same_inputs(Node_List* p, int idx); |
| 395 | // CloneMap utilities |
| 396 | bool same_origin_idx(Node* a, Node* b) const; |
| 397 | bool same_generation(Node* a, Node* b) const; |
| 398 | |
| 399 | // methods |
| 400 | |
| 401 | // Extract the superword level parallelism |
| 402 | void (); |
| 403 | // Find the adjacent memory references and create pack pairs for them. |
| 404 | void find_adjacent_refs(); |
| 405 | // Tracing support |
| 406 | #ifndef PRODUCT |
| 407 | void find_adjacent_refs_trace_1(Node* best_align_to_mem_ref, int best_iv_adjustment); |
| 408 | void print_loop(bool whole); |
| 409 | #endif |
| 410 | // Find a memory reference to align the loop induction variable to. |
| 411 | MemNode* find_align_to_ref(Node_List &memops); |
| 412 | // Calculate loop's iv adjustment for this memory ops. |
| 413 | int get_iv_adjustment(MemNode* mem); |
| 414 | // Can the preloop align the reference to position zero in the vector? |
| 415 | bool ref_is_alignable(SWPointer& p); |
| 416 | // rebuild the graph so all loads in different iterations of cloned loop become dependant on phi node (in _do_vector_loop only) |
| 417 | bool hoist_loads_in_graph(); |
| 418 | // Test whether MemNode::Memory dependency to the same load but in the first iteration of this loop is coming from memory phi |
| 419 | // Return false if failed |
| 420 | Node* find_phi_for_mem_dep(LoadNode* ld); |
| 421 | // Return same node but from the first generation. Return 0, if not found |
| 422 | Node* first_node(Node* nd); |
| 423 | // Return same node as this but from the last generation. Return 0, if not found |
| 424 | Node* last_node(Node* n); |
| 425 | // Mark nodes belonging to first and last generation |
| 426 | // returns first generation index or -1 if vectorization/simd is impossible |
| 427 | int mark_generations(); |
| 428 | // swapping inputs of commutative instruction (Add or Mul) |
| 429 | bool fix_commutative_inputs(Node* gold, Node* fix); |
| 430 | // make packs forcefully (in _do_vector_loop only) |
| 431 | bool pack_parallel(); |
| 432 | // Construct dependency graph. |
| 433 | void dependence_graph(); |
| 434 | // Return a memory slice (node list) in predecessor order starting at "start" |
| 435 | void mem_slice_preds(Node* start, Node* stop, GrowableArray<Node*> &preds); |
| 436 | // Can s1 and s2 be in a pack with s1 immediately preceding s2 and s1 aligned at "align" |
| 437 | bool stmts_can_pack(Node* s1, Node* s2, int align); |
| 438 | // Does s exist in a pack at position pos? |
| 439 | bool exists_at(Node* s, uint pos); |
| 440 | // Is s1 immediately before s2 in memory? |
| 441 | bool are_adjacent_refs(Node* s1, Node* s2); |
| 442 | // Are s1 and s2 similar? |
| 443 | bool isomorphic(Node* s1, Node* s2); |
| 444 | // Is there no data path from s1 to s2 or s2 to s1? |
| 445 | bool independent(Node* s1, Node* s2); |
| 446 | // For a node pair (s1, s2) which is isomorphic and independent, |
| 447 | // do s1 and s2 have similar input edges? |
| 448 | bool have_similar_inputs(Node* s1, Node* s2); |
| 449 | // Is there a data path between s1 and s2 and both are reductions? |
| 450 | bool reduction(Node* s1, Node* s2); |
| 451 | // Helper for independent |
| 452 | bool independent_path(Node* shallow, Node* deep, uint dp=0); |
| 453 | void set_alignment(Node* s1, Node* s2, int align); |
| 454 | int data_size(Node* s); |
| 455 | // Extend packset by following use->def and def->use links from pack members. |
| 456 | void extend_packlist(); |
| 457 | // Extend the packset by visiting operand definitions of nodes in pack p |
| 458 | bool follow_use_defs(Node_List* p); |
| 459 | // Extend the packset by visiting uses of nodes in pack p |
| 460 | bool follow_def_uses(Node_List* p); |
| 461 | // For extended packsets, ordinally arrange uses packset by major component |
| 462 | void order_def_uses(Node_List* p); |
| 463 | // Estimate the savings from executing s1 and s2 as a pack |
| 464 | int est_savings(Node* s1, Node* s2); |
| 465 | int adjacent_profit(Node* s1, Node* s2); |
| 466 | int pack_cost(int ct); |
| 467 | int unpack_cost(int ct); |
| 468 | // Combine packs A and B with A.last == B.first into A.first..,A.last,B.second,..B.last |
| 469 | void combine_packs(); |
| 470 | // Construct the map from nodes to packs. |
| 471 | void construct_my_pack_map(); |
| 472 | // Remove packs that are not implemented or not profitable. |
| 473 | void filter_packs(); |
| 474 | // Merge CMoveD into new vector-nodes |
| 475 | void merge_packs_to_cmovd(); |
| 476 | // Adjust the memory graph for the packed operations |
| 477 | void schedule(); |
| 478 | // Remove "current" from its current position in the memory graph and insert |
| 479 | // it after the appropriate insert points (lip or uip); |
| 480 | void remove_and_insert(MemNode *current, MemNode *prev, MemNode *lip, Node *uip, Unique_Node_List &schd_before); |
| 481 | // Within a store pack, schedule stores together by moving out the sandwiched memory ops according |
| 482 | // to dependence info; and within a load pack, move loads down to the last executed load. |
| 483 | void co_locate_pack(Node_List* p); |
| 484 | // Convert packs into vector node operations |
| 485 | void output(); |
| 486 | // Create a vector operand for the nodes in pack p for operand: in(opd_idx) |
| 487 | Node* vector_opd(Node_List* p, int opd_idx); |
| 488 | // Can code be generated for pack p? |
| 489 | bool implemented(Node_List* p); |
| 490 | // For pack p, are all operands and all uses (with in the block) vector? |
| 491 | bool profitable(Node_List* p); |
| 492 | // If a use of pack p is not a vector use, then replace the use with an extract operation. |
| 493 | void (Node_List* p); |
| 494 | // Is use->in(u_idx) a vector use? |
| 495 | bool is_vector_use(Node* use, int u_idx); |
| 496 | // Construct reverse postorder list of block members |
| 497 | bool construct_bb(); |
| 498 | // Initialize per node info |
| 499 | void initialize_bb(); |
| 500 | // Insert n into block after pos |
| 501 | void bb_insert_after(Node* n, int pos); |
| 502 | // Compute max depth for expressions from beginning of block |
| 503 | void compute_max_depth(); |
| 504 | // Compute necessary vector element type for expressions |
| 505 | void compute_vector_element_type(); |
| 506 | // Are s1 and s2 in a pack pair and ordered as s1,s2? |
| 507 | bool in_packset(Node* s1, Node* s2); |
| 508 | // Is s in pack p? |
| 509 | Node_List* in_pack(Node* s, Node_List* p); |
| 510 | // Remove the pack at position pos in the packset |
| 511 | void remove_pack_at(int pos); |
| 512 | // Return the node executed first in pack p. |
| 513 | Node* executed_first(Node_List* p); |
| 514 | // Return the node executed last in pack p. |
| 515 | Node* executed_last(Node_List* p); |
| 516 | static LoadNode::ControlDependency control_dependency(Node_List* p); |
| 517 | // Alignment within a vector memory reference |
| 518 | int memory_alignment(MemNode* s, int iv_adjust); |
| 519 | // (Start, end] half-open range defining which operands are vector |
| 520 | void vector_opd_range(Node* n, uint* start, uint* end); |
| 521 | // Smallest type containing range of values |
| 522 | const Type* container_type(Node* n); |
| 523 | // Adjust pre-loop limit so that in main loop, a load/store reference |
| 524 | // to align_to_ref will be a position zero in the vector. |
| 525 | void align_initial_loop_index(MemNode* align_to_ref); |
| 526 | // Find pre loop end from main loop. Returns null if none. |
| 527 | CountedLoopEndNode* get_pre_loop_end(CountedLoopNode *cl); |
| 528 | // Is the use of d1 in u1 at the same operand position as d2 in u2? |
| 529 | bool opnd_positions_match(Node* d1, Node* u1, Node* d2, Node* u2); |
| 530 | void init(); |
| 531 | // clean up some basic structures - used if the ideal graph was rebuilt |
| 532 | void restart(); |
| 533 | |
| 534 | // print methods |
| 535 | void print_packset(); |
| 536 | void print_pack(Node_List* p); |
| 537 | void print_bb(); |
| 538 | void print_stmt(Node* s); |
| 539 | char* blank(uint depth); |
| 540 | |
| 541 | void packset_sort(int n); |
| 542 | }; |
| 543 | |
| 544 | |
| 545 | |
| 546 | //------------------------------SWPointer--------------------------- |
| 547 | // Information about an address for dependence checking and vector alignment |
| 548 | class SWPointer { |
| 549 | protected: |
| 550 | MemNode* _mem; // My memory reference node |
| 551 | SuperWord* _slp; // SuperWord class |
| 552 | |
| 553 | Node* _base; // NULL if unsafe nonheap reference |
| 554 | Node* _adr; // address pointer |
| 555 | jint _scale; // multiplier for iv (in bytes), 0 if no loop iv |
| 556 | jint _offset; // constant offset (in bytes) |
| 557 | Node* _invar; // invariant offset (in bytes), NULL if none |
| 558 | bool _negate_invar; // if true then use: (0 - _invar) |
| 559 | Node_Stack* _nstack; // stack used to record a swpointer trace of variants |
| 560 | bool _analyze_only; // Used in loop unrolling only for swpointer trace |
| 561 | uint _stack_idx; // Used in loop unrolling only for swpointer trace |
| 562 | |
| 563 | PhaseIdealLoop* phase() { return _slp->phase(); } |
| 564 | IdealLoopTree* lpt() { return _slp->lpt(); } |
| 565 | PhiNode* iv() { return _slp->iv(); } // Induction var |
| 566 | |
| 567 | bool invariant(Node* n); |
| 568 | |
| 569 | // Match: k*iv + offset |
| 570 | bool scaled_iv_plus_offset(Node* n); |
| 571 | // Match: k*iv where k is a constant that's not zero |
| 572 | bool scaled_iv(Node* n); |
| 573 | // Match: offset is (k [+/- invariant]) |
| 574 | bool offset_plus_k(Node* n, bool negate = false); |
| 575 | |
| 576 | public: |
| 577 | enum CMP { |
| 578 | Less = 1, |
| 579 | Greater = 2, |
| 580 | Equal = 4, |
| 581 | NotEqual = (Less | Greater), |
| 582 | NotComparable = (Less | Greater | Equal) |
| 583 | }; |
| 584 | |
| 585 | SWPointer(MemNode* mem, SuperWord* slp, Node_Stack *nstack, bool analyze_only); |
| 586 | // Following is used to create a temporary object during |
| 587 | // the pattern match of an address expression. |
| 588 | SWPointer(SWPointer* p); |
| 589 | |
| 590 | bool valid() { return _adr != NULL; } |
| 591 | bool has_iv() { return _scale != 0; } |
| 592 | |
| 593 | Node* base() { return _base; } |
| 594 | Node* adr() { return _adr; } |
| 595 | MemNode* mem() { return _mem; } |
| 596 | int scale_in_bytes() { return _scale; } |
| 597 | Node* invar() { return _invar; } |
| 598 | bool negate_invar() { return _negate_invar; } |
| 599 | int offset_in_bytes() { return _offset; } |
| 600 | int memory_size() { return _mem->memory_size(); } |
| 601 | Node_Stack* node_stack() { return _nstack; } |
| 602 | |
| 603 | // Comparable? |
| 604 | int cmp(SWPointer& q) { |
| 605 | if (valid() && q.valid() && |
| 606 | (_adr == q._adr || (_base == _adr && q._base == q._adr)) && |
| 607 | _scale == q._scale && |
| 608 | _invar == q._invar && |
| 609 | _negate_invar == q._negate_invar) { |
| 610 | bool overlap = q._offset < _offset + memory_size() && |
| 611 | _offset < q._offset + q.memory_size(); |
| 612 | return overlap ? Equal : (_offset < q._offset ? Less : Greater); |
| 613 | } else { |
| 614 | return NotComparable; |
| 615 | } |
| 616 | } |
| 617 | |
| 618 | bool not_equal(SWPointer& q) { return not_equal(cmp(q)); } |
| 619 | bool equal(SWPointer& q) { return equal(cmp(q)); } |
| 620 | bool comparable(SWPointer& q) { return comparable(cmp(q)); } |
| 621 | static bool not_equal(int cmp) { return cmp <= NotEqual; } |
| 622 | static bool equal(int cmp) { return cmp == Equal; } |
| 623 | static bool comparable(int cmp) { return cmp < NotComparable; } |
| 624 | |
| 625 | void print(); |
| 626 | |
| 627 | #ifndef PRODUCT |
| 628 | class Tracer { |
| 629 | friend class SuperWord; |
| 630 | friend class SWPointer; |
| 631 | SuperWord* _slp; |
| 632 | static int _depth; |
| 633 | int _depth_save; |
| 634 | void print_depth(); |
| 635 | int depth() const { return _depth; } |
| 636 | void set_depth(int d) { _depth = d; } |
| 637 | void inc_depth() { _depth++;} |
| 638 | void dec_depth() { if (_depth > 0) _depth--;} |
| 639 | void store_depth() {_depth_save = _depth;} |
| 640 | void restore_depth() {_depth = _depth_save;} |
| 641 | |
| 642 | class Depth { |
| 643 | friend class Tracer; |
| 644 | friend class SWPointer; |
| 645 | friend class SuperWord; |
| 646 | Depth() { ++_depth; } |
| 647 | Depth(int x) { _depth = 0; } |
| 648 | ~Depth() { if (_depth > 0) --_depth;} |
| 649 | }; |
| 650 | Tracer (SuperWord* slp) : _slp(slp) {} |
| 651 | |
| 652 | // tracing functions |
| 653 | void ctor_1(Node* mem); |
| 654 | void ctor_2(Node* adr); |
| 655 | void ctor_3(Node* adr, int i); |
| 656 | void ctor_4(Node* adr, int i); |
| 657 | void ctor_5(Node* adr, Node* base, int i); |
| 658 | void ctor_6(Node* mem); |
| 659 | |
| 660 | void invariant_1(Node *n, Node *n_c); |
| 661 | |
| 662 | void scaled_iv_plus_offset_1(Node* n); |
| 663 | void scaled_iv_plus_offset_2(Node* n); |
| 664 | void scaled_iv_plus_offset_3(Node* n); |
| 665 | void scaled_iv_plus_offset_4(Node* n); |
| 666 | void scaled_iv_plus_offset_5(Node* n); |
| 667 | void scaled_iv_plus_offset_6(Node* n); |
| 668 | void scaled_iv_plus_offset_7(Node* n); |
| 669 | void scaled_iv_plus_offset_8(Node* n); |
| 670 | |
| 671 | void scaled_iv_1(Node* n); |
| 672 | void scaled_iv_2(Node* n, int scale); |
| 673 | void scaled_iv_3(Node* n, int scale); |
| 674 | void scaled_iv_4(Node* n, int scale); |
| 675 | void scaled_iv_5(Node* n, int scale); |
| 676 | void scaled_iv_6(Node* n, int scale); |
| 677 | void scaled_iv_7(Node* n); |
| 678 | void scaled_iv_8(Node* n, SWPointer* tmp); |
| 679 | void scaled_iv_9(Node* n, int _scale, int _offset, int mult); |
| 680 | void scaled_iv_10(Node* n); |
| 681 | |
| 682 | void offset_plus_k_1(Node* n); |
| 683 | void offset_plus_k_2(Node* n, int _offset); |
| 684 | void offset_plus_k_3(Node* n, int _offset); |
| 685 | void offset_plus_k_4(Node* n); |
| 686 | void offset_plus_k_5(Node* n, Node* _invar); |
| 687 | void offset_plus_k_6(Node* n, Node* _invar, bool _negate_invar, int _offset); |
| 688 | void offset_plus_k_7(Node* n, Node* _invar, bool _negate_invar, int _offset); |
| 689 | void offset_plus_k_8(Node* n, Node* _invar, bool _negate_invar, int _offset); |
| 690 | void offset_plus_k_9(Node* n, Node* _invar, bool _negate_invar, int _offset); |
| 691 | void offset_plus_k_10(Node* n, Node* _invar, bool _negate_invar, int _offset); |
| 692 | void offset_plus_k_11(Node* n); |
| 693 | |
| 694 | } _tracer;//TRacer; |
| 695 | #endif |
| 696 | }; |
| 697 | |
| 698 | #endif // SHARE_OPTO_SUPERWORD_HPP |
| 699 | |