1 /*
2 * Copyright (c) 2007, 2015, 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 #include "precompiled.hpp"
25 #include "compiler/compileLog.hpp"
26 #include "libadt/vectset.hpp"
27 #include "memory/allocation.inline.hpp"
28 #include "opto/addnode.hpp"
29 #include "opto/callnode.hpp"
30 #include "opto/castnode.hpp"
31 #include "opto/convertnode.hpp"
32 #include "opto/divnode.hpp"
33 #include "opto/matcher.hpp"
34 #include "opto/memnode.hpp"
35 #include "opto/mulnode.hpp"
36 #include "opto/opcodes.hpp"
37 #include "opto/opaquenode.hpp"
38 #include "opto/superword.hpp"
39 #include "opto/vectornode.hpp"
40 #include "opto/movenode.hpp"
41
42 //
43 // S U P E R W O R D T R A N S F O R M
44 //=============================================================================
45
46 //------------------------------SuperWord---------------------------
47 SuperWord::SuperWord(PhaseIdealLoop* phase) :
48 _phase(phase),
49 _igvn(phase->_igvn),
50 _arena(phase->C->comp_arena()),
51 _packset(arena(), 8, 0, NULL), // packs for the current block
52 _bb_idx(arena(), (int)(1.10 * phase->C->unique()), 0, 0), // node idx to index in bb
53 _block(arena(), 8, 0, NULL), // nodes in current block
54 _post_block(arena(), 8, 0, NULL), // nodes common to current block which are marked as post loop vectorizable
55 _data_entry(arena(), 8, 0, NULL), // nodes with all inputs from outside
56 _mem_slice_head(arena(), 8, 0, NULL), // memory slice heads
57 _mem_slice_tail(arena(), 8, 0, NULL), // memory slice tails
58 _node_info(arena(), 8, 0, SWNodeInfo::initial), // info needed per node
59 _clone_map(phase->C->clone_map()), // map of nodes created in cloning
60 _cmovev_kit(_arena, this), // map to facilitate CMoveVD creation
61 _align_to_ref(NULL), // memory reference to align vectors to
62 _disjoint_ptrs(arena(), 8, 0, OrderedPair::initial), // runtime disambiguated pointer pairs
63 _dg(_arena), // dependence graph
64 _visited(arena()), // visited node set
65 _post_visited(arena()), // post visited node set
66 _n_idx_list(arena(), 8), // scratch list of (node,index) pairs
67 _stk(arena(), 8, 0, NULL), // scratch stack of nodes
68 _nlist(arena(), 8, 0, NULL), // scratch list of nodes
69 _lpt(NULL), // loop tree node
70 _lp(NULL), // LoopNode
71 _bb(NULL), // basic block
72 _iv(NULL), // induction var
73 _race_possible(false), // cases where SDMU is true
74 _early_return(true), // analysis evaluations routine
75 _num_work_vecs(0), // amount of vector work we have
76 _num_reductions(0), // amount of reduction work we have
77 _do_vector_loop(phase->C->do_vector_loop()), // whether to do vectorization/simd style
78 _do_reserve_copy(DoReserveCopyInSuperWord),
79 _ii_first(-1), // first loop generation index - only if do_vector_loop()
80 _ii_last(-1), // last loop generation index - only if do_vector_loop()
81 _ii_order(arena(), 8, 0, 0)
82 {
83 #ifndef PRODUCT
84 _vector_loop_debug = 0;
85 if (_phase->C->method() != NULL) {
86 _vector_loop_debug = phase->C->directive()->VectorizeDebugOption;
87 }
88
89 #endif
90 }
91
92 //------------------------------transform_loop---------------------------
93 void SuperWord::transform_loop(IdealLoopTree* lpt, bool do_optimization) {
94 assert(UseSuperWord, "should be");
95 // Do vectors exist on this architecture?
96 if (Matcher::vector_width_in_bytes(T_BYTE) < 2) return;
97
98 assert(lpt->_head->is_CountedLoop(), "must be");
99 CountedLoopNode *cl = lpt->_head->as_CountedLoop();
100
101 if (!cl->is_valid_counted_loop()) return; // skip malformed counted loop
102
103 bool post_loop_allowed = (PostLoopMultiversioning && Matcher::has_predicated_vectors() && cl->is_post_loop());
104 if (post_loop_allowed) {
105 if (cl->is_reduction_loop()) return; // no predication mapping
106 Node *limit = cl->limit();
107 if (limit->is_Con()) return; // non constant limits only
108 // Now check the limit for expressions we do not handle
109 if (limit->is_Add()) {
110 Node *in2 = limit->in(2);
111 if (in2->is_Con()) {
112 int val = in2->get_int();
113 // should not try to program these cases
114 if (val < 0) return;
115 }
116 }
117 }
118
119 // skip any loop that has not been assigned max unroll by analysis
120 if (do_optimization) {
121 if (cl->slp_max_unroll() == 0) return;
122 }
123
124 // Check for no control flow in body (other than exit)
125 Node *cl_exit = cl->loopexit();
126 if (cl->is_main_loop() && (cl_exit->in(0) != lpt->_head)) {
127 #ifndef PRODUCT
128 if (TraceSuperWord) {
129 tty->print_cr("SuperWord::transform_loop: loop too complicated, cl_exit->in(0) != lpt->_head");
130 tty->print("cl_exit %d", cl_exit->_idx); cl_exit->dump();
131 tty->print("cl_exit->in(0) %d", cl_exit->in(0)->_idx); cl_exit->in(0)->dump();
132 tty->print("lpt->_head %d", lpt->_head->_idx); lpt->_head->dump();
133 lpt->dump_head();
134 }
135 #endif
136 return;
137 }
138
139 // Make sure the are no extra control users of the loop backedge
140 if (cl->back_control()->outcnt() != 1) {
141 return;
142 }
143
144 // Skip any loops already optimized by slp
145 if (cl->is_vectorized_loop()) return;
146
147 if (cl->is_main_loop()) {
148 // Check for pre-loop ending with CountedLoopEnd(Bool(Cmp(x,Opaque1(limit))))
149 CountedLoopEndNode* pre_end = get_pre_loop_end(cl);
150 if (pre_end == NULL) return;
151 Node *pre_opaq1 = pre_end->limit();
152 if (pre_opaq1->Opcode() != Op_Opaque1) return;
153 }
154
155 init(); // initialize data structures
156
157 set_lpt(lpt);
158 set_lp(cl);
159
160 // For now, define one block which is the entire loop body
161 set_bb(cl);
162
163 if (do_optimization) {
164 assert(_packset.length() == 0, "packset must be empty");
165 SLP_extract();
166 if (PostLoopMultiversioning && Matcher::has_predicated_vectors()) {
167 if (cl->is_vectorized_loop() && cl->is_main_loop() && !cl->is_reduction_loop()) {
168 IdealLoopTree *lpt_next = lpt->_next;
169 CountedLoopNode *cl_next = lpt_next->_head->as_CountedLoop();
170 _phase->has_range_checks(lpt_next);
171 if (cl_next->is_post_loop() && !cl_next->range_checks_present()) {
172 if (!cl_next->is_vectorized_loop()) {
173 int slp_max_unroll_factor = cl->slp_max_unroll();
174 cl_next->set_slp_max_unroll(slp_max_unroll_factor);
175 }
176 }
177 }
178 }
179 }
180 }
181
182 //------------------------------early unrolling analysis------------------------------
183 void SuperWord::unrolling_analysis(int &local_loop_unroll_factor) {
184 bool is_slp = true;
185 ResourceMark rm;
186 size_t ignored_size = lpt()->_body.size();
187 int *ignored_loop_nodes = NEW_RESOURCE_ARRAY(int, ignored_size);
188 Node_Stack nstack((int)ignored_size);
189 CountedLoopNode *cl = lpt()->_head->as_CountedLoop();
190 Node *cl_exit = cl->loopexit();
191 int rpo_idx = _post_block.length();
192
193 assert(rpo_idx == 0, "post loop block is empty");
194
195 // First clear the entries
196 for (uint i = 0; i < lpt()->_body.size(); i++) {
197 ignored_loop_nodes[i] = -1;
198 }
199
200 int max_vector = Matcher::max_vector_size(T_INT);
201 bool post_loop_allowed = (PostLoopMultiversioning && Matcher::has_predicated_vectors() && cl->is_post_loop());
202
203 // Process the loop, some/all of the stack entries will not be in order, ergo
204 // need to preprocess the ignored initial state before we process the loop
205 for (uint i = 0; i < lpt()->_body.size(); i++) {
206 Node* n = lpt()->_body.at(i);
207 if (n == cl->incr() ||
208 n->is_reduction() ||
209 n->is_AddP() ||
210 n->is_Cmp() ||
211 n->is_IfTrue() ||
212 n->is_CountedLoop() ||
213 (n == cl_exit)) {
214 ignored_loop_nodes[i] = n->_idx;
215 continue;
216 }
217
218 if (n->is_If()) {
219 IfNode *iff = n->as_If();
220 if (iff->_fcnt != COUNT_UNKNOWN && iff->_prob != PROB_UNKNOWN) {
221 if (lpt()->is_loop_exit(iff)) {
222 ignored_loop_nodes[i] = n->_idx;
223 continue;
224 }
225 }
226 }
227
228 if (n->is_Phi() && (n->bottom_type() == Type::MEMORY)) {
229 Node* n_tail = n->in(LoopNode::LoopBackControl);
230 if (n_tail != n->in(LoopNode::EntryControl)) {
231 if (!n_tail->is_Mem()) {
232 is_slp = false;
233 break;
234 }
235 }
236 }
237
238 // This must happen after check of phi/if
239 if (n->is_Phi() || n->is_If()) {
240 ignored_loop_nodes[i] = n->_idx;
241 continue;
242 }
243
244 if (n->is_LoadStore() || n->is_MergeMem() ||
245 (n->is_Proj() && !n->as_Proj()->is_CFG())) {
246 is_slp = false;
247 break;
248 }
249
250 // Ignore nodes with non-primitive type.
251 BasicType bt;
252 if (n->is_Mem()) {
253 bt = n->as_Mem()->memory_type();
254 } else {
255 bt = n->bottom_type()->basic_type();
256 }
257 if (is_java_primitive(bt) == false) {
258 ignored_loop_nodes[i] = n->_idx;
259 continue;
260 }
261
262 if (n->is_Mem()) {
263 MemNode* current = n->as_Mem();
264 Node* adr = n->in(MemNode::Address);
265 Node* n_ctrl = _phase->get_ctrl(adr);
266
267 // save a queue of post process nodes
268 if (n_ctrl != NULL && lpt()->is_member(_phase->get_loop(n_ctrl))) {
269 // Process the memory expression
270 int stack_idx = 0;
271 bool have_side_effects = true;
272 if (adr->is_AddP() == false) {
273 nstack.push(adr, stack_idx++);
274 } else {
275 // Mark the components of the memory operation in nstack
276 SWPointer p1(current, this, &nstack, true);
277 have_side_effects = p1.node_stack()->is_nonempty();
278 }
279
280 // Process the pointer stack
281 while (have_side_effects) {
282 Node* pointer_node = nstack.node();
283 for (uint j = 0; j < lpt()->_body.size(); j++) {
284 Node* cur_node = lpt()->_body.at(j);
285 if (cur_node == pointer_node) {
286 ignored_loop_nodes[j] = cur_node->_idx;
287 break;
288 }
289 }
290 nstack.pop();
291 have_side_effects = nstack.is_nonempty();
292 }
293 }
294 }
295 }
296
297 if (is_slp) {
298 // Now we try to find the maximum supported consistent vector which the machine
299 // description can use
300 bool small_basic_type = false;
301 for (uint i = 0; i < lpt()->_body.size(); i++) {
302 if (ignored_loop_nodes[i] != -1) continue;
303
304 BasicType bt;
305 Node* n = lpt()->_body.at(i);
306 if (n->is_Mem()) {
307 bt = n->as_Mem()->memory_type();
308 } else {
309 bt = n->bottom_type()->basic_type();
310 }
311
312 if (post_loop_allowed) {
313 if (!small_basic_type) {
314 switch (bt) {
315 case T_CHAR:
316 case T_BYTE:
317 case T_SHORT:
318 small_basic_type = true;
319 break;
320 case T_LONG:
321 // TODO: Remove when support completed for mask context with LONG.
322 // The ad files need to be augmented.
323 small_basic_type = true;
324 break;
325 }
326 }
327 }
328
329 if (is_java_primitive(bt) == false) continue;
330
331 int cur_max_vector = Matcher::max_vector_size(bt);
332
333 // If a max vector exists which is not larger than _local_loop_unroll_factor
334 // stop looking, we already have the max vector to map to.
335 if (cur_max_vector < local_loop_unroll_factor) {
336 is_slp = false;
337 if (TraceSuperWordLoopUnrollAnalysis) {
338 tty->print_cr("slp analysis fails: unroll limit greater than max vector\n");
339 }
340 break;
341 }
342
343 // Map the maximal common vector
344 if (VectorNode::implemented(n->Opcode(), cur_max_vector, bt)) {
345 if (cur_max_vector < max_vector) {
346 max_vector = cur_max_vector;
347 }
348
349 // We only process post loops on predicated targets where we want to
350 // mask map the loop to a single iteration
351 if (post_loop_allowed) {
352 _post_block.at_put_grow(rpo_idx++, n);
353 }
354 }
355 }
356 if (is_slp) {
357 local_loop_unroll_factor = max_vector;
358 cl->mark_passed_slp();
359 }
360 cl->mark_was_slp();
361 if (cl->is_main_loop()) {
362 cl->set_slp_max_unroll(local_loop_unroll_factor);
363 } else if (post_loop_allowed) {
364 if (!small_basic_type) {
365 // avoid replication context for small basic types in programmable masked loops
366 cl->set_slp_max_unroll(local_loop_unroll_factor);
367 }
368 }
369 }
370 }
371
372 //------------------------------SLP_extract---------------------------
373 // Extract the superword level parallelism
374 //
375 // 1) A reverse post-order of nodes in the block is constructed. By scanning
376 // this list from first to last, all definitions are visited before their uses.
377 //
378 // 2) A point-to-point dependence graph is constructed between memory references.
379 // This simplies the upcoming "independence" checker.
380 //
381 // 3) The maximum depth in the node graph from the beginning of the block
382 // to each node is computed. This is used to prune the graph search
383 // in the independence checker.
384 //
385 // 4) For integer types, the necessary bit width is propagated backwards
386 // from stores to allow packed operations on byte, char, and short
387 // integers. This reverses the promotion to type "int" that javac
388 // did for operations like: char c1,c2,c3; c1 = c2 + c3.
389 //
390 // 5) One of the memory references is picked to be an aligned vector reference.
391 // The pre-loop trip count is adjusted to align this reference in the
392 // unrolled body.
393 //
394 // 6) The initial set of pack pairs is seeded with memory references.
395 //
396 // 7) The set of pack pairs is extended by following use->def and def->use links.
397 //
398 // 8) The pairs are combined into vector sized packs.
399 //
400 // 9) Reorder the memory slices to co-locate members of the memory packs.
401 //
402 // 10) Generate ideal vector nodes for the final set of packs and where necessary,
403 // inserting scalar promotion, vector creation from multiple scalars, and
404 // extraction of scalar values from vectors.
405 //
406 void SuperWord::SLP_extract() {
407
408 #ifndef PRODUCT
409 if (_do_vector_loop && TraceSuperWord) {
410 tty->print("SuperWord::SLP_extract\n");
411 tty->print("input loop\n");
412 _lpt->dump_head();
413 _lpt->dump();
414 for (uint i = 0; i < _lpt->_body.size(); i++) {
415 _lpt->_body.at(i)->dump();
416 }
417 }
418 #endif
419 // Ready the block
420 if (!construct_bb()) {
421 return; // Exit if no interesting nodes or complex graph.
422 }
423
424 // build _dg, _disjoint_ptrs
425 dependence_graph();
426
427 // compute function depth(Node*)
428 compute_max_depth();
429
430 CountedLoopNode *cl = lpt()->_head->as_CountedLoop();
431 bool post_loop_allowed = (PostLoopMultiversioning && Matcher::has_predicated_vectors() && cl->is_post_loop());
432 if (cl->is_main_loop()) {
433 if (_do_vector_loop) {
434 if (mark_generations() != -1) {
435 hoist_loads_in_graph(); // this only rebuild the graph; all basic structs need rebuild explicitly
436
437 if (!construct_bb()) {
438 return; // Exit if no interesting nodes or complex graph.
439 }
440 dependence_graph();
441 compute_max_depth();
442 }
443
444 #ifndef PRODUCT
445 if (TraceSuperWord) {
446 tty->print_cr("\nSuperWord::_do_vector_loop: graph after hoist_loads_in_graph");
447 _lpt->dump_head();
448 for (int j = 0; j < _block.length(); j++) {
449 Node* n = _block.at(j);
450 int d = depth(n);
451 for (int i = 0; i < d; i++) tty->print("%s", " ");
452 tty->print("%d :", d);
453 n->dump();
454 }
455 }
456 #endif
457 }
458
459 compute_vector_element_type();
460
461 // Attempt vectorization
462
463 find_adjacent_refs();
464
465 extend_packlist();
466
467 if (_do_vector_loop) {
468 if (_packset.length() == 0) {
469 if (TraceSuperWord) {
470 tty->print_cr("\nSuperWord::_do_vector_loop DFA could not build packset, now trying to build anyway");
471 }
472 pack_parallel();
473 }
474 }
475
476 combine_packs();
477
478 construct_my_pack_map();
479
480 if (_do_vector_loop) {
481 merge_packs_to_cmovd();
482 }
483
484 filter_packs();
485
486 schedule();
487 } else if (post_loop_allowed) {
488 int saved_mapped_unroll_factor = cl->slp_max_unroll();
489 if (saved_mapped_unroll_factor) {
490 int vector_mapped_unroll_factor = saved_mapped_unroll_factor;
491
492 // now reset the slp_unroll_factor so that we can check the analysis mapped
493 // what the vector loop was mapped to
494 cl->set_slp_max_unroll(0);
495
496 // do the analysis on the post loop
497 unrolling_analysis(vector_mapped_unroll_factor);
498
499 // if our analyzed loop is a canonical fit, start processing it
500 if (vector_mapped_unroll_factor == saved_mapped_unroll_factor) {
501 // now add the vector nodes to packsets
502 for (int i = 0; i < _post_block.length(); i++) {
503 Node* n = _post_block.at(i);
504 Node_List* singleton = new Node_List();
505 singleton->push(n);
506 _packset.append(singleton);
507 set_my_pack(n, singleton);
508 }
509
510 // map base types for vector usage
511 compute_vector_element_type();
512 } else {
513 return;
514 }
515 } else {
516 // for some reason we could not map the slp analysis state of the vectorized loop
517 return;
518 }
519 }
520
521 output();
522 }
523
524 //------------------------------find_adjacent_refs---------------------------
525 // Find the adjacent memory references and create pack pairs for them.
526 // This is the initial set of packs that will then be extended by
527 // following use->def and def->use links. The align positions are
528 // assigned relative to the reference "align_to_ref"
529 void SuperWord::find_adjacent_refs() {
530 // Get list of memory operations
531 Node_List memops;
532 for (int i = 0; i < _block.length(); i++) {
533 Node* n = _block.at(i);
534 if (n->is_Mem() && !n->is_LoadStore() && in_bb(n) &&
535 is_java_primitive(n->as_Mem()->memory_type())) {
536 int align = memory_alignment(n->as_Mem(), 0);
537 if (align != bottom_align) {
538 memops.push(n);
539 }
540 }
541 }
542
543 Node_List align_to_refs;
544 int best_iv_adjustment = 0;
545 MemNode* best_align_to_mem_ref = NULL;
546
547 while (memops.size() != 0) {
548 // Find a memory reference to align to.
549 MemNode* mem_ref = find_align_to_ref(memops);
550 if (mem_ref == NULL) break;
551 align_to_refs.push(mem_ref);
552 int iv_adjustment = get_iv_adjustment(mem_ref);
553
554 if (best_align_to_mem_ref == NULL) {
555 // Set memory reference which is the best from all memory operations
556 // to be used for alignment. The pre-loop trip count is modified to align
557 // this reference to a vector-aligned address.
558 best_align_to_mem_ref = mem_ref;
559 best_iv_adjustment = iv_adjustment;
560 NOT_PRODUCT(find_adjacent_refs_trace_1(best_align_to_mem_ref, best_iv_adjustment);)
561 }
562
563 SWPointer align_to_ref_p(mem_ref, this, NULL, false);
564 // Set alignment relative to "align_to_ref" for all related memory operations.
565 for (int i = memops.size() - 1; i >= 0; i--) {
566 MemNode* s = memops.at(i)->as_Mem();
567 if (isomorphic(s, mem_ref) &&
568 (!_do_vector_loop || same_origin_idx(s, mem_ref))) {
569 SWPointer p2(s, this, NULL, false);
570 if (p2.comparable(align_to_ref_p)) {
571 int align = memory_alignment(s, iv_adjustment);
572 set_alignment(s, align);
573 }
574 }
575 }
576
577 // Create initial pack pairs of memory operations for which
578 // alignment is set and vectors will be aligned.
579 bool create_pack = true;
580 if (memory_alignment(mem_ref, best_iv_adjustment) == 0 || _do_vector_loop) {
581 if (!Matcher::misaligned_vectors_ok()) {
582 int vw = vector_width(mem_ref);
583 int vw_best = vector_width(best_align_to_mem_ref);
584 if (vw > vw_best) {
585 // Do not vectorize a memory access with more elements per vector
586 // if unaligned memory access is not allowed because number of
587 // iterations in pre-loop will be not enough to align it.
588 create_pack = false;
589 } else {
590 SWPointer p2(best_align_to_mem_ref, this, NULL, false);
591 if (align_to_ref_p.invar() != p2.invar()) {
592 // Do not vectorize memory accesses with different invariants
593 // if unaligned memory accesses are not allowed.
594 create_pack = false;
595 }
596 }
597 }
598 } else {
599 if (same_velt_type(mem_ref, best_align_to_mem_ref)) {
600 // Can't allow vectorization of unaligned memory accesses with the
601 // same type since it could be overlapped accesses to the same array.
602 create_pack = false;
603 } else {
604 // Allow independent (different type) unaligned memory operations
605 // if HW supports them.
606 if (!Matcher::misaligned_vectors_ok()) {
607 create_pack = false;
608 } else {
609 // Check if packs of the same memory type but
610 // with a different alignment were created before.
611 for (uint i = 0; i < align_to_refs.size(); i++) {
612 MemNode* mr = align_to_refs.at(i)->as_Mem();
613 if (same_velt_type(mr, mem_ref) &&
614 memory_alignment(mr, iv_adjustment) != 0)
615 create_pack = false;
616 }
617 }
618 }
619 }
620 if (create_pack) {
621 for (uint i = 0; i < memops.size(); i++) {
622 Node* s1 = memops.at(i);
623 int align = alignment(s1);
624 if (align == top_align) continue;
625 for (uint j = 0; j < memops.size(); j++) {
626 Node* s2 = memops.at(j);
627 if (alignment(s2) == top_align) continue;
628 if (s1 != s2 && are_adjacent_refs(s1, s2)) {
629 if (stmts_can_pack(s1, s2, align)) {
630 Node_List* pair = new Node_List();
631 pair->push(s1);
632 pair->push(s2);
633 if (!_do_vector_loop || same_origin_idx(s1, s2)) {
634 _packset.append(pair);
635 }
636 }
637 }
638 }
639 }
640 } else { // Don't create unaligned pack
641 // First, remove remaining memory ops of the same type from the list.
642 for (int i = memops.size() - 1; i >= 0; i--) {
643 MemNode* s = memops.at(i)->as_Mem();
644 if (same_velt_type(s, mem_ref)) {
645 memops.remove(i);
646 }
647 }
648
649 // Second, remove already constructed packs of the same type.
650 for (int i = _packset.length() - 1; i >= 0; i--) {
651 Node_List* p = _packset.at(i);
652 MemNode* s = p->at(0)->as_Mem();
653 if (same_velt_type(s, mem_ref)) {
654 remove_pack_at(i);
655 }
656 }
657
658 // If needed find the best memory reference for loop alignment again.
659 if (same_velt_type(mem_ref, best_align_to_mem_ref)) {
660 // Put memory ops from remaining packs back on memops list for
661 // the best alignment search.
662 uint orig_msize = memops.size();
663 for (int i = 0; i < _packset.length(); i++) {
664 Node_List* p = _packset.at(i);
665 MemNode* s = p->at(0)->as_Mem();
666 assert(!same_velt_type(s, mem_ref), "sanity");
667 memops.push(s);
668 }
669 best_align_to_mem_ref = find_align_to_ref(memops);
670 if (best_align_to_mem_ref == NULL) {
671 if (TraceSuperWord) {
672 tty->print_cr("SuperWord::find_adjacent_refs(): best_align_to_mem_ref == NULL");
673 }
674 break;
675 }
676 best_iv_adjustment = get_iv_adjustment(best_align_to_mem_ref);
677 NOT_PRODUCT(find_adjacent_refs_trace_1(best_align_to_mem_ref, best_iv_adjustment);)
678 // Restore list.
679 while (memops.size() > orig_msize)
680 (void)memops.pop();
681 }
682 } // unaligned memory accesses
683
684 // Remove used mem nodes.
685 for (int i = memops.size() - 1; i >= 0; i--) {
686 MemNode* m = memops.at(i)->as_Mem();
687 if (alignment(m) != top_align) {
688 memops.remove(i);
689 }
690 }
691
692 } // while (memops.size() != 0
693 set_align_to_ref(best_align_to_mem_ref);
694
695 if (TraceSuperWord) {
696 tty->print_cr("\nAfter find_adjacent_refs");
697 print_packset();
698 }
699 }
700
701 #ifndef PRODUCT
702 void SuperWord::find_adjacent_refs_trace_1(Node* best_align_to_mem_ref, int best_iv_adjustment) {
703 if (is_trace_adjacent()) {
704 tty->print("SuperWord::find_adjacent_refs best_align_to_mem_ref = %d, best_iv_adjustment = %d",
705 best_align_to_mem_ref->_idx, best_iv_adjustment);
706 best_align_to_mem_ref->dump();
707 }
708 }
709 #endif
710
711 //------------------------------find_align_to_ref---------------------------
712 // Find a memory reference to align the loop induction variable to.
713 // Looks first at stores then at loads, looking for a memory reference
714 // with the largest number of references similar to it.
715 MemNode* SuperWord::find_align_to_ref(Node_List &memops) {
716 GrowableArray<int> cmp_ct(arena(), memops.size(), memops.size(), 0);
717
718 // Count number of comparable memory ops
719 for (uint i = 0; i < memops.size(); i++) {
720 MemNode* s1 = memops.at(i)->as_Mem();
721 SWPointer p1(s1, this, NULL, false);
722 // Discard if pre loop can't align this reference
723 if (!ref_is_alignable(p1)) {
724 *cmp_ct.adr_at(i) = 0;
725 continue;
726 }
727 for (uint j = i+1; j < memops.size(); j++) {
728 MemNode* s2 = memops.at(j)->as_Mem();
729 if (isomorphic(s1, s2)) {
730 SWPointer p2(s2, this, NULL, false);
731 if (p1.comparable(p2)) {
732 (*cmp_ct.adr_at(i))++;
733 (*cmp_ct.adr_at(j))++;
734 }
735 }
736 }
737 }
738
739 // Find Store (or Load) with the greatest number of "comparable" references,
740 // biggest vector size, smallest data size and smallest iv offset.
741 int max_ct = 0;
742 int max_vw = 0;
743 int max_idx = -1;
744 int min_size = max_jint;
745 int min_iv_offset = max_jint;
746 for (uint j = 0; j < memops.size(); j++) {
747 MemNode* s = memops.at(j)->as_Mem();
748 if (s->is_Store()) {
749 int vw = vector_width_in_bytes(s);
750 assert(vw > 1, "sanity");
751 SWPointer p(s, this, NULL, false);
752 if (cmp_ct.at(j) > max_ct ||
753 cmp_ct.at(j) == max_ct &&
754 (vw > max_vw ||
755 vw == max_vw &&
756 (data_size(s) < min_size ||
757 data_size(s) == min_size &&
758 (p.offset_in_bytes() < min_iv_offset)))) {
759 max_ct = cmp_ct.at(j);
760 max_vw = vw;
761 max_idx = j;
762 min_size = data_size(s);
763 min_iv_offset = p.offset_in_bytes();
764 }
765 }
766 }
767 // If no stores, look at loads
768 if (max_ct == 0) {
769 for (uint j = 0; j < memops.size(); j++) {
770 MemNode* s = memops.at(j)->as_Mem();
771 if (s->is_Load()) {
772 int vw = vector_width_in_bytes(s);
773 assert(vw > 1, "sanity");
774 SWPointer p(s, this, NULL, false);
775 if (cmp_ct.at(j) > max_ct ||
776 cmp_ct.at(j) == max_ct &&
777 (vw > max_vw ||
778 vw == max_vw &&
779 (data_size(s) < min_size ||
780 data_size(s) == min_size &&
781 (p.offset_in_bytes() < min_iv_offset)))) {
782 max_ct = cmp_ct.at(j);
783 max_vw = vw;
784 max_idx = j;
785 min_size = data_size(s);
786 min_iv_offset = p.offset_in_bytes();
787 }
788 }
789 }
790 }
791
792 #ifdef ASSERT
793 if (TraceSuperWord && Verbose) {
794 tty->print_cr("\nVector memops after find_align_to_ref");
795 for (uint i = 0; i < memops.size(); i++) {
796 MemNode* s = memops.at(i)->as_Mem();
797 s->dump();
798 }
799 }
800 #endif
801
802 if (max_ct > 0) {
803 #ifdef ASSERT
804 if (TraceSuperWord) {
805 tty->print("\nVector align to node: ");
806 memops.at(max_idx)->as_Mem()->dump();
807 }
808 #endif
809 return memops.at(max_idx)->as_Mem();
810 }
811 return NULL;
812 }
813
814 //------------------------------ref_is_alignable---------------------------
815 // Can the preloop align the reference to position zero in the vector?
816 bool SuperWord::ref_is_alignable(SWPointer& p) {
817 if (!p.has_iv()) {
818 return true; // no induction variable
819 }
820 CountedLoopEndNode* pre_end = get_pre_loop_end(lp()->as_CountedLoop());
821 assert(pre_end != NULL, "we must have a correct pre-loop");
822 assert(pre_end->stride_is_con(), "pre loop stride is constant");
823 int preloop_stride = pre_end->stride_con();
824
825 int span = preloop_stride * p.scale_in_bytes();
826 int mem_size = p.memory_size();
827 int offset = p.offset_in_bytes();
828 // Stride one accesses are alignable if offset is aligned to memory operation size.
829 // Offset can be unaligned when UseUnalignedAccesses is used.
830 if (ABS(span) == mem_size && (ABS(offset) % mem_size) == 0) {
831 return true;
832 }
833 // If the initial offset from start of the object is computable,
834 // check if the pre-loop can align the final offset accordingly.
835 //
836 // In other words: Can we find an i such that the offset
837 // after i pre-loop iterations is aligned to vw?
838 // (init_offset + pre_loop) % vw == 0 (1)
839 // where
840 // pre_loop = i * span
841 // is the number of bytes added to the offset by i pre-loop iterations.
842 //
843 // For this to hold we need pre_loop to increase init_offset by
844 // pre_loop = vw - (init_offset % vw)
845 //
846 // This is only possible if pre_loop is divisible by span because each
847 // pre-loop iteration increases the initial offset by 'span' bytes:
848 // (vw - (init_offset % vw)) % span == 0
849 //
850 int vw = vector_width_in_bytes(p.mem());
851 assert(vw > 1, "sanity");
852 Node* init_nd = pre_end->init_trip();
853 if (init_nd->is_Con() && p.invar() == NULL) {
854 int init = init_nd->bottom_type()->is_int()->get_con();
855 int init_offset = init * p.scale_in_bytes() + offset;
856 assert(init_offset >= 0, "positive offset from object start");
857 if (vw % span == 0) {
858 // If vm is a multiple of span, we use formula (1).
859 if (span > 0) {
860 return (vw - (init_offset % vw)) % span == 0;
861 } else {
862 assert(span < 0, "nonzero stride * scale");
863 return (init_offset % vw) % -span == 0;
864 }
865 } else if (span % vw == 0) {
866 // If span is a multiple of vw, we can simplify formula (1) to:
867 // (init_offset + i * span) % vw == 0
868 // =>
869 // (init_offset % vw) + ((i * span) % vw) == 0
870 // =>
871 // init_offset % vw == 0
872 //
873 // Because we add a multiple of vw to the initial offset, the final
874 // offset is a multiple of vw if and only if init_offset is a multiple.
875 //
876 return (init_offset % vw) == 0;
877 }
878 }
879 return false;
880 }
881
882 //---------------------------get_iv_adjustment---------------------------
883 // Calculate loop's iv adjustment for this memory ops.
884 int SuperWord::get_iv_adjustment(MemNode* mem_ref) {
885 SWPointer align_to_ref_p(mem_ref, this, NULL, false);
886 int offset = align_to_ref_p.offset_in_bytes();
887 int scale = align_to_ref_p.scale_in_bytes();
888 int elt_size = align_to_ref_p.memory_size();
889 int vw = vector_width_in_bytes(mem_ref);
890 assert(vw > 1, "sanity");
891 int iv_adjustment;
892 if (scale != 0) {
893 int stride_sign = (scale * iv_stride()) > 0 ? 1 : -1;
894 // At least one iteration is executed in pre-loop by default. As result
895 // several iterations are needed to align memory operations in main-loop even
896 // if offset is 0.
897 int iv_adjustment_in_bytes = (stride_sign * vw - (offset % vw));
898 assert(((ABS(iv_adjustment_in_bytes) % elt_size) == 0),
899 "(%d) should be divisible by (%d)", iv_adjustment_in_bytes, elt_size);
900 iv_adjustment = iv_adjustment_in_bytes/elt_size;
901 } else {
902 // This memory op is not dependent on iv (scale == 0)
903 iv_adjustment = 0;
904 }
905
906 #ifndef PRODUCT
907 if (TraceSuperWord) {
908 tty->print("SuperWord::get_iv_adjustment: n = %d, noffset = %d iv_adjust = %d elt_size = %d scale = %d iv_stride = %d vect_size %d: ",
909 mem_ref->_idx, offset, iv_adjustment, elt_size, scale, iv_stride(), vw);
910 mem_ref->dump();
911 }
912 #endif
913 return iv_adjustment;
914 }
915
916 //---------------------------dependence_graph---------------------------
917 // Construct dependency graph.
918 // Add dependence edges to load/store nodes for memory dependence
919 // A.out()->DependNode.in(1) and DependNode.out()->B.prec(x)
920 void SuperWord::dependence_graph() {
921 CountedLoopNode *cl = lpt()->_head->as_CountedLoop();
922 // First, assign a dependence node to each memory node
923 for (int i = 0; i < _block.length(); i++ ) {
924 Node *n = _block.at(i);
925 if (n->is_Mem() || n->is_Phi() && n->bottom_type() == Type::MEMORY) {
926 _dg.make_node(n);
927 }
928 }
929
930 // For each memory slice, create the dependences
931 for (int i = 0; i < _mem_slice_head.length(); i++) {
932 Node* n = _mem_slice_head.at(i);
933 Node* n_tail = _mem_slice_tail.at(i);
934
935 // Get slice in predecessor order (last is first)
936 if (cl->is_main_loop()) {
937 mem_slice_preds(n_tail, n, _nlist);
938 }
939
940 #ifndef PRODUCT
941 if(TraceSuperWord && Verbose) {
942 tty->print_cr("SuperWord::dependence_graph: built a new mem slice");
943 for (int j = _nlist.length() - 1; j >= 0 ; j--) {
944 _nlist.at(j)->dump();
945 }
946 }
947 #endif
948 // Make the slice dependent on the root
949 DepMem* slice = _dg.dep(n);
950 _dg.make_edge(_dg.root(), slice);
951
952 // Create a sink for the slice
953 DepMem* slice_sink = _dg.make_node(NULL);
954 _dg.make_edge(slice_sink, _dg.tail());
955
956 // Now visit each pair of memory ops, creating the edges
957 for (int j = _nlist.length() - 1; j >= 0 ; j--) {
958 Node* s1 = _nlist.at(j);
959
960 // If no dependency yet, use slice
961 if (_dg.dep(s1)->in_cnt() == 0) {
962 _dg.make_edge(slice, s1);
963 }
964 SWPointer p1(s1->as_Mem(), this, NULL, false);
965 bool sink_dependent = true;
966 for (int k = j - 1; k >= 0; k--) {
967 Node* s2 = _nlist.at(k);
968 if (s1->is_Load() && s2->is_Load())
969 continue;
970 SWPointer p2(s2->as_Mem(), this, NULL, false);
971
972 int cmp = p1.cmp(p2);
973 if (SuperWordRTDepCheck &&
974 p1.base() != p2.base() && p1.valid() && p2.valid()) {
975 // Create a runtime check to disambiguate
976 OrderedPair pp(p1.base(), p2.base());
977 _disjoint_ptrs.append_if_missing(pp);
978 } else if (!SWPointer::not_equal(cmp)) {
979 // Possibly same address
980 _dg.make_edge(s1, s2);
981 sink_dependent = false;
982 }
983 }
984 if (sink_dependent) {
985 _dg.make_edge(s1, slice_sink);
986 }
987 }
988
989 if (TraceSuperWord) {
990 tty->print_cr("\nDependence graph for slice: %d", n->_idx);
991 for (int q = 0; q < _nlist.length(); q++) {
992 _dg.print(_nlist.at(q));
993 }
994 tty->cr();
995 }
996
997 _nlist.clear();
998 }
999
1000 if (TraceSuperWord) {
1001 tty->print_cr("\ndisjoint_ptrs: %s", _disjoint_ptrs.length() > 0 ? "" : "NONE");
1002 for (int r = 0; r < _disjoint_ptrs.length(); r++) {
1003 _disjoint_ptrs.at(r).print();
1004 tty->cr();
1005 }
1006 tty->cr();
1007 }
1008
1009 }
1010
1011 //---------------------------mem_slice_preds---------------------------
1012 // Return a memory slice (node list) in predecessor order starting at "start"
1013 void SuperWord::mem_slice_preds(Node* start, Node* stop, GrowableArray<Node*> &preds) {
1014 assert(preds.length() == 0, "start empty");
1015 Node* n = start;
1016 Node* prev = NULL;
1017 while (true) {
1018 NOT_PRODUCT( if(is_trace_mem_slice()) tty->print_cr("SuperWord::mem_slice_preds: n %d", n->_idx);)
1019 assert(in_bb(n), "must be in block");
1020 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1021 Node* out = n->fast_out(i);
1022 if (out->is_Load()) {
1023 if (in_bb(out)) {
1024 preds.push(out);
1025 if (TraceSuperWord && Verbose) {
1026 tty->print_cr("SuperWord::mem_slice_preds: added pred(%d)", out->_idx);
1027 }
1028 }
1029 } else {
1030 // FIXME
1031 if (out->is_MergeMem() && !in_bb(out)) {
1032 // Either unrolling is causing a memory edge not to disappear,
1033 // or need to run igvn.optimize() again before SLP
1034 } else if (out->is_Phi() && out->bottom_type() == Type::MEMORY && !in_bb(out)) {
1035 // Ditto. Not sure what else to check further.
1036 } else if (out->Opcode() == Op_StoreCM && out->in(MemNode::OopStore) == n) {
1037 // StoreCM has an input edge used as a precedence edge.
1038 // Maybe an issue when oop stores are vectorized.
1039 } else {
1040 assert(out == prev || prev == NULL, "no branches off of store slice");
1041 }
1042 }//else
1043 }//for
1044 if (n == stop) break;
1045 preds.push(n);
1046 if (TraceSuperWord && Verbose) {
1047 tty->print_cr("SuperWord::mem_slice_preds: added pred(%d)", n->_idx);
1048 }
1049 prev = n;
1050 assert(n->is_Mem(), "unexpected node %s", n->Name());
1051 n = n->in(MemNode::Memory);
1052 }
1053 }
1054
1055 //------------------------------stmts_can_pack---------------------------
1056 // Can s1 and s2 be in a pack with s1 immediately preceding s2 and
1057 // s1 aligned at "align"
1058 bool SuperWord::stmts_can_pack(Node* s1, Node* s2, int align) {
1059
1060 // Do not use superword for non-primitives
1061 BasicType bt1 = velt_basic_type(s1);
1062 BasicType bt2 = velt_basic_type(s2);
1063 if(!is_java_primitive(bt1) || !is_java_primitive(bt2))
1064 return false;
1065 if (Matcher::max_vector_size(bt1) < 2) {
1066 return false; // No vectors for this type
1067 }
1068
1069 if (isomorphic(s1, s2)) {
1070 if (independent(s1, s2) || reduction(s1, s2)) {
1071 if (!exists_at(s1, 0) && !exists_at(s2, 1)) {
1072 if (!s1->is_Mem() || are_adjacent_refs(s1, s2)) {
1073 int s1_align = alignment(s1);
1074 int s2_align = alignment(s2);
1075 if (s1_align == top_align || s1_align == align) {
1076 if (s2_align == top_align || s2_align == align + data_size(s1)) {
1077 return true;
1078 }
1079 }
1080 }
1081 }
1082 }
1083 }
1084 return false;
1085 }
1086
1087 //------------------------------exists_at---------------------------
1088 // Does s exist in a pack at position pos?
1089 bool SuperWord::exists_at(Node* s, uint pos) {
1090 for (int i = 0; i < _packset.length(); i++) {
1091 Node_List* p = _packset.at(i);
1092 if (p->at(pos) == s) {
1093 return true;
1094 }
1095 }
1096 return false;
1097 }
1098
1099 //------------------------------are_adjacent_refs---------------------------
1100 // Is s1 immediately before s2 in memory?
1101 bool SuperWord::are_adjacent_refs(Node* s1, Node* s2) {
1102 if (!s1->is_Mem() || !s2->is_Mem()) return false;
1103 if (!in_bb(s1) || !in_bb(s2)) return false;
1104
1105 // Do not use superword for non-primitives
1106 if (!is_java_primitive(s1->as_Mem()->memory_type()) ||
1107 !is_java_primitive(s2->as_Mem()->memory_type())) {
1108 return false;
1109 }
1110
1111 // FIXME - co_locate_pack fails on Stores in different mem-slices, so
1112 // only pack memops that are in the same alias set until that's fixed.
1113 if (_phase->C->get_alias_index(s1->as_Mem()->adr_type()) !=
1114 _phase->C->get_alias_index(s2->as_Mem()->adr_type()))
1115 return false;
1116 SWPointer p1(s1->as_Mem(), this, NULL, false);
1117 SWPointer p2(s2->as_Mem(), this, NULL, false);
1118 if (p1.base() != p2.base() || !p1.comparable(p2)) return false;
1119 int diff = p2.offset_in_bytes() - p1.offset_in_bytes();
1120 return diff == data_size(s1);
1121 }
1122
1123 //------------------------------isomorphic---------------------------
1124 // Are s1 and s2 similar?
1125 bool SuperWord::isomorphic(Node* s1, Node* s2) {
1126 if (s1->Opcode() != s2->Opcode()) return false;
1127 if (s1->req() != s2->req()) return false;
1128 if (s1->in(0) != s2->in(0)) return false;
1129 if (!same_velt_type(s1, s2)) return false;
1130 return true;
1131 }
1132
1133 //------------------------------independent---------------------------
1134 // Is there no data path from s1 to s2 or s2 to s1?
1135 bool SuperWord::independent(Node* s1, Node* s2) {
1136 // assert(s1->Opcode() == s2->Opcode(), "check isomorphic first");
1137 int d1 = depth(s1);
1138 int d2 = depth(s2);
1139 if (d1 == d2) return s1 != s2;
1140 Node* deep = d1 > d2 ? s1 : s2;
1141 Node* shallow = d1 > d2 ? s2 : s1;
1142
1143 visited_clear();
1144
1145 return independent_path(shallow, deep);
1146 }
1147
1148 //------------------------------reduction---------------------------
1149 // Is there a data path between s1 and s2 and the nodes reductions?
1150 bool SuperWord::reduction(Node* s1, Node* s2) {
1151 bool retValue = false;
1152 int d1 = depth(s1);
1153 int d2 = depth(s2);
1154 if (d1 + 1 == d2) {
1155 if (s1->is_reduction() && s2->is_reduction()) {
1156 // This is an ordered set, so s1 should define s2
1157 for (DUIterator_Fast imax, i = s1->fast_outs(imax); i < imax; i++) {
1158 Node* t1 = s1->fast_out(i);
1159 if (t1 == s2) {
1160 // both nodes are reductions and connected
1161 retValue = true;
1162 }
1163 }
1164 }
1165 }
1166
1167 return retValue;
1168 }
1169
1170 //------------------------------independent_path------------------------------
1171 // Helper for independent
1172 bool SuperWord::independent_path(Node* shallow, Node* deep, uint dp) {
1173 if (dp >= 1000) return false; // stop deep recursion
1174 visited_set(deep);
1175 int shal_depth = depth(shallow);
1176 assert(shal_depth <= depth(deep), "must be");
1177 for (DepPreds preds(deep, _dg); !preds.done(); preds.next()) {
1178 Node* pred = preds.current();
1179 if (in_bb(pred) && !visited_test(pred)) {
1180 if (shallow == pred) {
1181 return false;
1182 }
1183 if (shal_depth < depth(pred) && !independent_path(shallow, pred, dp+1)) {
1184 return false;
1185 }
1186 }
1187 }
1188 return true;
1189 }
1190
1191 //------------------------------set_alignment---------------------------
1192 void SuperWord::set_alignment(Node* s1, Node* s2, int align) {
1193 set_alignment(s1, align);
1194 if (align == top_align || align == bottom_align) {
1195 set_alignment(s2, align);
1196 } else {
1197 set_alignment(s2, align + data_size(s1));
1198 }
1199 }
1200
1201 //------------------------------data_size---------------------------
1202 int SuperWord::data_size(Node* s) {
1203 Node* use = NULL; //test if the node is a candidate for CMoveVD optimization, then return the size of CMov
1204 if (_do_vector_loop) {
1205 use = _cmovev_kit.is_Bool_candidate(s);
1206 if (use != NULL) {
1207 return data_size(use);
1208 }
1209 use = _cmovev_kit.is_CmpD_candidate(s);
1210 if (use != NULL) {
1211 return data_size(use);
1212 }
1213 }
1214 int bsize = type2aelembytes(velt_basic_type(s));
1215 assert(bsize != 0, "valid size");
1216 return bsize;
1217 }
1218
1219 //------------------------------extend_packlist---------------------------
1220 // Extend packset by following use->def and def->use links from pack members.
1221 void SuperWord::extend_packlist() {
1222 bool changed;
1223 do {
1224 packset_sort(_packset.length());
1225 changed = false;
1226 for (int i = 0; i < _packset.length(); i++) {
1227 Node_List* p = _packset.at(i);
1228 changed |= follow_use_defs(p);
1229 changed |= follow_def_uses(p);
1230 }
1231 } while (changed);
1232
1233 if (_race_possible) {
1234 for (int i = 0; i < _packset.length(); i++) {
1235 Node_List* p = _packset.at(i);
1236 order_def_uses(p);
1237 }
1238 }
1239
1240 if (TraceSuperWord) {
1241 tty->print_cr("\nAfter extend_packlist");
1242 print_packset();
1243 }
1244 }
1245
1246 //------------------------------follow_use_defs---------------------------
1247 // Extend the packset by visiting operand definitions of nodes in pack p
1248 bool SuperWord::follow_use_defs(Node_List* p) {
1249 assert(p->size() == 2, "just checking");
1250 Node* s1 = p->at(0);
1251 Node* s2 = p->at(1);
1252 assert(s1->req() == s2->req(), "just checking");
1253 assert(alignment(s1) + data_size(s1) == alignment(s2), "just checking");
1254
1255 if (s1->is_Load()) return false;
1256
1257 int align = alignment(s1);
1258 NOT_PRODUCT(if(is_trace_alignment()) tty->print_cr("SuperWord::follow_use_defs: s1 %d, align %d", s1->_idx, align);)
1259 bool changed = false;
1260 int start = s1->is_Store() ? MemNode::ValueIn : 1;
1261 int end = s1->is_Store() ? MemNode::ValueIn+1 : s1->req();
1262 for (int j = start; j < end; j++) {
1263 Node* t1 = s1->in(j);
1264 Node* t2 = s2->in(j);
1265 if (!in_bb(t1) || !in_bb(t2))
1266 continue;
1267 if (stmts_can_pack(t1, t2, align)) {
1268 if (est_savings(t1, t2) >= 0) {
1269 Node_List* pair = new Node_List();
1270 pair->push(t1);
1271 pair->push(t2);
1272 _packset.append(pair);
1273 NOT_PRODUCT(if(is_trace_alignment()) tty->print_cr("SuperWord::follow_use_defs: set_alignment(%d, %d, %d)", t1->_idx, t2->_idx, align);)
1274 set_alignment(t1, t2, align);
1275 changed = true;
1276 }
1277 }
1278 }
1279 return changed;
1280 }
1281
1282 //------------------------------follow_def_uses---------------------------
1283 // Extend the packset by visiting uses of nodes in pack p
1284 bool SuperWord::follow_def_uses(Node_List* p) {
1285 bool changed = false;
1286 Node* s1 = p->at(0);
1287 Node* s2 = p->at(1);
1288 assert(p->size() == 2, "just checking");
1289 assert(s1->req() == s2->req(), "just checking");
1290 assert(alignment(s1) + data_size(s1) == alignment(s2), "just checking");
1291
1292 if (s1->is_Store()) return false;
1293
1294 int align = alignment(s1);
1295 NOT_PRODUCT(if(is_trace_alignment()) tty->print_cr("SuperWord::follow_def_uses: s1 %d, align %d", s1->_idx, align);)
1296 int savings = -1;
1297 int num_s1_uses = 0;
1298 Node* u1 = NULL;
1299 Node* u2 = NULL;
1300 for (DUIterator_Fast imax, i = s1->fast_outs(imax); i < imax; i++) {
1301 Node* t1 = s1->fast_out(i);
1302 num_s1_uses++;
1303 if (!in_bb(t1)) continue;
1304 for (DUIterator_Fast jmax, j = s2->fast_outs(jmax); j < jmax; j++) {
1305 Node* t2 = s2->fast_out(j);
1306 if (!in_bb(t2)) continue;
1307 if (!opnd_positions_match(s1, t1, s2, t2))
1308 continue;
1309 if (stmts_can_pack(t1, t2, align)) {
1310 int my_savings = est_savings(t1, t2);
1311 if (my_savings > savings) {
1312 savings = my_savings;
1313 u1 = t1;
1314 u2 = t2;
1315 }
1316 }
1317 }
1318 }
1319 if (num_s1_uses > 1) {
1320 _race_possible = true;
1321 }
1322 if (savings >= 0) {
1323 Node_List* pair = new Node_List();
1324 pair->push(u1);
1325 pair->push(u2);
1326 _packset.append(pair);
1327 NOT_PRODUCT(if(is_trace_alignment()) tty->print_cr("SuperWord::follow_def_uses: set_alignment(%d, %d, %d)", u1->_idx, u2->_idx, align);)
1328 set_alignment(u1, u2, align);
1329 changed = true;
1330 }
1331 return changed;
1332 }
1333
1334 //------------------------------order_def_uses---------------------------
1335 // For extended packsets, ordinally arrange uses packset by major component
1336 void SuperWord::order_def_uses(Node_List* p) {
1337 Node* s1 = p->at(0);
1338
1339 if (s1->is_Store()) return;
1340
1341 // reductions are always managed beforehand
1342 if (s1->is_reduction()) return;
1343
1344 for (DUIterator_Fast imax, i = s1->fast_outs(imax); i < imax; i++) {
1345 Node* t1 = s1->fast_out(i);
1346
1347 // Only allow operand swap on commuting operations
1348 if (!t1->is_Add() && !t1->is_Mul()) {
1349 break;
1350 }
1351
1352 // Now find t1's packset
1353 Node_List* p2 = NULL;
1354 for (int j = 0; j < _packset.length(); j++) {
1355 p2 = _packset.at(j);
1356 Node* first = p2->at(0);
1357 if (t1 == first) {
1358 break;
1359 }
1360 p2 = NULL;
1361 }
1362 // Arrange all sub components by the major component
1363 if (p2 != NULL) {
1364 for (uint j = 1; j < p->size(); j++) {
1365 Node* d1 = p->at(j);
1366 Node* u1 = p2->at(j);
1367 opnd_positions_match(s1, t1, d1, u1);
1368 }
1369 }
1370 }
1371 }
1372
1373 //---------------------------opnd_positions_match-------------------------
1374 // Is the use of d1 in u1 at the same operand position as d2 in u2?
1375 bool SuperWord::opnd_positions_match(Node* d1, Node* u1, Node* d2, Node* u2) {
1376 // check reductions to see if they are marshalled to represent the reduction
1377 // operator in a specified opnd
1378 if (u1->is_reduction() && u2->is_reduction()) {
1379 // ensure reductions have phis and reduction definitions feeding the 1st operand
1380 Node* first = u1->in(2);
1381 if (first->is_Phi() || first->is_reduction()) {
1382 u1->swap_edges(1, 2);
1383 }
1384 // ensure reductions have phis and reduction definitions feeding the 1st operand
1385 first = u2->in(2);
1386 if (first->is_Phi() || first->is_reduction()) {
1387 u2->swap_edges(1, 2);
1388 }
1389 return true;
1390 }
1391
1392 uint ct = u1->req();
1393 if (ct != u2->req()) return false;
1394 uint i1 = 0;
1395 uint i2 = 0;
1396 do {
1397 for (i1++; i1 < ct; i1++) if (u1->in(i1) == d1) break;
1398 for (i2++; i2 < ct; i2++) if (u2->in(i2) == d2) break;
1399 if (i1 != i2) {
1400 if ((i1 == (3-i2)) && (u2->is_Add() || u2->is_Mul())) {
1401 // Further analysis relies on operands position matching.
1402 u2->swap_edges(i1, i2);
1403 } else {
1404 return false;
1405 }
1406 }
1407 } while (i1 < ct);
1408 return true;
1409 }
1410
1411 //------------------------------est_savings---------------------------
1412 // Estimate the savings from executing s1 and s2 as a pack
1413 int SuperWord::est_savings(Node* s1, Node* s2) {
1414 int save_in = 2 - 1; // 2 operations per instruction in packed form
1415
1416 // inputs
1417 for (uint i = 1; i < s1->req(); i++) {
1418 Node* x1 = s1->in(i);
1419 Node* x2 = s2->in(i);
1420 if (x1 != x2) {
1421 if (are_adjacent_refs(x1, x2)) {
1422 save_in += adjacent_profit(x1, x2);
1423 } else if (!in_packset(x1, x2)) {
1424 save_in -= pack_cost(2);
1425 } else {
1426 save_in += unpack_cost(2);
1427 }
1428 }
1429 }
1430
1431 // uses of result
1432 uint ct = 0;
1433 int save_use = 0;
1434 for (DUIterator_Fast imax, i = s1->fast_outs(imax); i < imax; i++) {
1435 Node* s1_use = s1->fast_out(i);
1436 for (int j = 0; j < _packset.length(); j++) {
1437 Node_List* p = _packset.at(j);
1438 if (p->at(0) == s1_use) {
1439 for (DUIterator_Fast kmax, k = s2->fast_outs(kmax); k < kmax; k++) {
1440 Node* s2_use = s2->fast_out(k);
1441 if (p->at(p->size()-1) == s2_use) {
1442 ct++;
1443 if (are_adjacent_refs(s1_use, s2_use)) {
1444 save_use += adjacent_profit(s1_use, s2_use);
1445 }
1446 }
1447 }
1448 }
1449 }
1450 }
1451
1452 if (ct < s1->outcnt()) save_use += unpack_cost(1);
1453 if (ct < s2->outcnt()) save_use += unpack_cost(1);
1454
1455 return MAX2(save_in, save_use);
1456 }
1457
1458 //------------------------------costs---------------------------
1459 int SuperWord::adjacent_profit(Node* s1, Node* s2) { return 2; }
1460 int SuperWord::pack_cost(int ct) { return ct; }
1461 int SuperWord::unpack_cost(int ct) { return ct; }
1462
1463 //------------------------------combine_packs---------------------------
1464 // Combine packs A and B with A.last == B.first into A.first..,A.last,B.second,..B.last
1465 void SuperWord::combine_packs() {
1466 bool changed = true;
1467 // Combine packs regardless max vector size.
1468 while (changed) {
1469 changed = false;
1470 for (int i = 0; i < _packset.length(); i++) {
1471 Node_List* p1 = _packset.at(i);
1472 if (p1 == NULL) continue;
1473 // Because of sorting we can start at i + 1
1474 for (int j = i + 1; j < _packset.length(); j++) {
1475 Node_List* p2 = _packset.at(j);
1476 if (p2 == NULL) continue;
1477 if (i == j) continue;
1478 if (p1->at(p1->size()-1) == p2->at(0)) {
1479 for (uint k = 1; k < p2->size(); k++) {
1480 p1->push(p2->at(k));
1481 }
1482 _packset.at_put(j, NULL);
1483 changed = true;
1484 }
1485 }
1486 }
1487 }
1488
1489 // Split packs which have size greater then max vector size.
1490 for (int i = 0; i < _packset.length(); i++) {
1491 Node_List* p1 = _packset.at(i);
1492 if (p1 != NULL) {
1493 BasicType bt = velt_basic_type(p1->at(0));
1494 uint max_vlen = Matcher::max_vector_size(bt); // Max elements in vector
1495 assert(is_power_of_2(max_vlen), "sanity");
1496 uint psize = p1->size();
1497 if (!is_power_of_2(psize)) {
1498 // Skip pack which can't be vector.
1499 // case1: for(...) { a[i] = i; } elements values are different (i+x)
1500 // case2: for(...) { a[i] = b[i+1]; } can't align both, load and store
1501 _packset.at_put(i, NULL);
1502 continue;
1503 }
1504 if (psize > max_vlen) {
1505 Node_List* pack = new Node_List();
1506 for (uint j = 0; j < psize; j++) {
1507 pack->push(p1->at(j));
1508 if (pack->size() >= max_vlen) {
1509 assert(is_power_of_2(pack->size()), "sanity");
1510 _packset.append(pack);
1511 pack = new Node_List();
1512 }
1513 }
1514 _packset.at_put(i, NULL);
1515 }
1516 }
1517 }
1518
1519 // Compress list.
1520 for (int i = _packset.length() - 1; i >= 0; i--) {
1521 Node_List* p1 = _packset.at(i);
1522 if (p1 == NULL) {
1523 _packset.remove_at(i);
1524 }
1525 }
1526
1527 if (TraceSuperWord) {
1528 tty->print_cr("\nAfter combine_packs");
1529 print_packset();
1530 }
1531 }
1532
1533 //-----------------------------construct_my_pack_map--------------------------
1534 // Construct the map from nodes to packs. Only valid after the
1535 // point where a node is only in one pack (after combine_packs).
1536 void SuperWord::construct_my_pack_map() {
1537 Node_List* rslt = NULL;
1538 for (int i = 0; i < _packset.length(); i++) {
1539 Node_List* p = _packset.at(i);
1540 for (uint j = 0; j < p->size(); j++) {
1541 Node* s = p->at(j);
1542 assert(my_pack(s) == NULL, "only in one pack");
1543 set_my_pack(s, p);
1544 }
1545 }
1546 }
1547
1548 //------------------------------filter_packs---------------------------
1549 // Remove packs that are not implemented or not profitable.
1550 void SuperWord::filter_packs() {
1551 // Remove packs that are not implemented
1552 for (int i = _packset.length() - 1; i >= 0; i--) {
1553 Node_List* pk = _packset.at(i);
1554 bool impl = implemented(pk);
1555 if (!impl) {
1556 #ifndef PRODUCT
1557 if (TraceSuperWord && Verbose) {
1558 tty->print_cr("Unimplemented");
1559 pk->at(0)->dump();
1560 }
1561 #endif
1562 remove_pack_at(i);
1563 }
1564 Node *n = pk->at(0);
1565 if (n->is_reduction()) {
1566 _num_reductions++;
1567 } else {
1568 _num_work_vecs++;
1569 }
1570 }
1571
1572 // Remove packs that are not profitable
1573 bool changed;
1574 do {
1575 changed = false;
1576 for (int i = _packset.length() - 1; i >= 0; i--) {
1577 Node_List* pk = _packset.at(i);
1578 bool prof = profitable(pk);
1579 if (!prof) {
1580 #ifndef PRODUCT
1581 if (TraceSuperWord && Verbose) {
1582 tty->print_cr("Unprofitable");
1583 pk->at(0)->dump();
1584 }
1585 #endif
1586 remove_pack_at(i);
1587 changed = true;
1588 }
1589 }
1590 } while (changed);
1591
1592 #ifndef PRODUCT
1593 if (TraceSuperWord) {
1594 tty->print_cr("\nAfter filter_packs");
1595 print_packset();
1596 tty->cr();
1597 }
1598 #endif
1599 }
1600
1601 //------------------------------merge_packs_to_cmovd---------------------------
1602 // Merge CMoveD into new vector-nodes
1603 // We want to catch this pattern and subsume CmpD and Bool into CMoveD
1604 //
1605 // SubD ConD
1606 // / | /
1607 // / | / /
1608 // / | / /
1609 // / | / /
1610 // / / /
1611 // / / | /
1612 // v / | /
1613 // CmpD | /
1614 // | | /
1615 // v | /
1616 // Bool | /
1617 // \ | /
1618 // \ | /
1619 // \ | /
1620 // \ | /
1621 // \ v /
1622 // CMoveD
1623 //
1624
1625 void SuperWord::merge_packs_to_cmovd() {
1626 for (int i = _packset.length() - 1; i >= 0; i--) {
1627 _cmovev_kit.make_cmovevd_pack(_packset.at(i));
1628 }
1629 #ifndef PRODUCT
1630 if (TraceSuperWord) {
1631 tty->print_cr("\nSuperWord::merge_packs_to_cmovd(): After merge");
1632 print_packset();
1633 tty->cr();
1634 }
1635 #endif
1636 }
1637
1638 Node* CMoveKit::is_Bool_candidate(Node* def) const {
1639 Node* use = NULL;
1640 if (!def->is_Bool() || def->in(0) != NULL || def->outcnt() != 1) {
1641 return NULL;
1642 }
1643 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
1644 use = def->fast_out(j);
1645 if (!_sw->same_generation(def, use) || !use->is_CMove()) {
1646 return NULL;
1647 }
1648 }
1649 return use;
1650 }
1651
1652 Node* CMoveKit::is_CmpD_candidate(Node* def) const {
1653 Node* use = NULL;
1654 if (!def->is_Cmp() || def->in(0) != NULL || def->outcnt() != 1) {
1655 return NULL;
1656 }
1657 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
1658 use = def->fast_out(j);
1659 if (!_sw->same_generation(def, use) || (use = is_Bool_candidate(use)) == NULL || !_sw->same_generation(def, use)) {
1660 return NULL;
1661 }
1662 }
1663 return use;
1664 }
1665
1666 Node_List* CMoveKit::make_cmovevd_pack(Node_List* cmovd_pk) {
1667 Node *cmovd = cmovd_pk->at(0);
1668 if (!cmovd->is_CMove()) {
1669 return NULL;
1670 }
1671 if (pack(cmovd) != NULL) { // already in the cmov pack
1672 return NULL;
1673 }
1674 if (cmovd->in(0) != NULL) {
1675 NOT_PRODUCT(if(_sw->is_trace_cmov()) {tty->print("CMoveKit::make_cmovevd_pack: CMoveD %d has control flow, escaping...", cmovd->_idx); cmovd->dump();})
1676 return NULL;
1677 }
1678
1679 Node* bol = cmovd->as_CMove()->in(CMoveNode::Condition);
1680 if (!bol->is_Bool()
1681 || bol->outcnt() != 1
1682 || !_sw->same_generation(bol, cmovd)
1683 || bol->in(0) != NULL // BoolNode has control flow!!
1684 || _sw->my_pack(bol) == NULL) {
1685 NOT_PRODUCT(if(_sw->is_trace_cmov()) {tty->print("CMoveKit::make_cmovevd_pack: Bool %d does not fit CMoveD %d for building vector, escaping...", bol->_idx, cmovd->_idx); bol->dump();})
1686 return NULL;
1687 }
1688 Node_List* bool_pk = _sw->my_pack(bol);
1689 if (bool_pk->size() != cmovd_pk->size() ) {
1690 return NULL;
1691 }
1692
1693 Node* cmpd = bol->in(1);
1694 if (!cmpd->is_Cmp()
1695 || cmpd->outcnt() != 1
1696 || !_sw->same_generation(cmpd, cmovd)
1697 || cmpd->in(0) != NULL // CmpDNode has control flow!!
1698 || _sw->my_pack(cmpd) == NULL) {
1699 NOT_PRODUCT(if(_sw->is_trace_cmov()) {tty->print("CMoveKit::make_cmovevd_pack: CmpD %d does not fit CMoveD %d for building vector, escaping...", cmpd->_idx, cmovd->_idx); cmpd->dump();})
1700 return NULL;
1701 }
1702 Node_List* cmpd_pk = _sw->my_pack(cmpd);
1703 if (cmpd_pk->size() != cmovd_pk->size() ) {
1704 return NULL;
1705 }
1706
1707 if (!test_cmpd_pack(cmpd_pk, cmovd_pk)) {
1708 NOT_PRODUCT(if(_sw->is_trace_cmov()) {tty->print("CMoveKit::make_cmovevd_pack: cmpd pack for CmpD %d failed vectorization test", cmpd->_idx); cmpd->dump();})
1709 return NULL;
1710 }
1711
1712 Node_List* new_cmpd_pk = new Node_List();
1713 uint sz = cmovd_pk->size() - 1;
1714 for (uint i = 0; i <= sz; ++i) {
1715 Node* cmov = cmovd_pk->at(i);
1716 Node* bol = bool_pk->at(i);
1717 Node* cmp = cmpd_pk->at(i);
1718
1719 new_cmpd_pk->insert(i, cmov);
1720
1721 map(cmov, new_cmpd_pk);
1722 map(bol, new_cmpd_pk);
1723 map(cmp, new_cmpd_pk);
1724
1725 _sw->set_my_pack(cmov, new_cmpd_pk); // and keep old packs for cmp and bool
1726 }
1727 _sw->_packset.remove(cmovd_pk);
1728 _sw->_packset.remove(bool_pk);
1729 _sw->_packset.remove(cmpd_pk);
1730 _sw->_packset.append(new_cmpd_pk);
1731 NOT_PRODUCT(if(_sw->is_trace_cmov()) {tty->print_cr("CMoveKit::make_cmovevd_pack: added syntactic CMoveD pack"); _sw->print_pack(new_cmpd_pk);})
1732 return new_cmpd_pk;
1733 }
1734
1735 bool CMoveKit::test_cmpd_pack(Node_List* cmpd_pk, Node_List* cmovd_pk) {
1736 Node* cmpd0 = cmpd_pk->at(0);
1737 assert(cmpd0->is_Cmp(), "CMoveKit::test_cmpd_pack: should be CmpDNode");
1738 assert(cmovd_pk->at(0)->is_CMove(), "CMoveKit::test_cmpd_pack: should be CMoveD");
1739 assert(cmpd_pk->size() == cmovd_pk->size(), "CMoveKit::test_cmpd_pack: should be same size");
1740 Node* in1 = cmpd0->in(1);
1741 Node* in2 = cmpd0->in(2);
1742 Node_List* in1_pk = _sw->my_pack(in1);
1743 Node_List* in2_pk = _sw->my_pack(in2);
1744
1745 if (in1_pk != NULL && in1_pk->size() != cmpd_pk->size()
1746 || in2_pk != NULL && in2_pk->size() != cmpd_pk->size() ) {
1747 return false;
1748 }
1749
1750 // test if "all" in1 are in the same pack or the same node
1751 if (in1_pk == NULL) {
1752 for (uint j = 1; j < cmpd_pk->size(); j++) {
1753 if (cmpd_pk->at(j)->in(1) != in1) {
1754 return false;
1755 }
1756 }//for: in1_pk is not pack but all CmpD nodes in the pack have the same in(1)
1757 }
1758 // test if "all" in2 are in the same pack or the same node
1759 if (in2_pk == NULL) {
1760 for (uint j = 1; j < cmpd_pk->size(); j++) {
1761 if (cmpd_pk->at(j)->in(2) != in2) {
1762 return false;
1763 }
1764 }//for: in2_pk is not pack but all CmpD nodes in the pack have the same in(2)
1765 }
1766 //now check if cmpd_pk may be subsumed in vector built for cmovd_pk
1767 int cmovd_ind1, cmovd_ind2;
1768 if (cmpd_pk->at(0)->in(1) == cmovd_pk->at(0)->as_CMove()->in(CMoveNode::IfFalse)
1769 && cmpd_pk->at(0)->in(2) == cmovd_pk->at(0)->as_CMove()->in(CMoveNode::IfTrue)) {
1770 cmovd_ind1 = CMoveNode::IfFalse;
1771 cmovd_ind2 = CMoveNode::IfTrue;
1772 } else if (cmpd_pk->at(0)->in(2) == cmovd_pk->at(0)->as_CMove()->in(CMoveNode::IfFalse)
1773 && cmpd_pk->at(0)->in(1) == cmovd_pk->at(0)->as_CMove()->in(CMoveNode::IfTrue)) {
1774 cmovd_ind2 = CMoveNode::IfFalse;
1775 cmovd_ind1 = CMoveNode::IfTrue;
1776 }
1777 else {
1778 return false;
1779 }
1780
1781 for (uint j = 1; j < cmpd_pk->size(); j++) {
1782 if (cmpd_pk->at(j)->in(1) != cmovd_pk->at(j)->as_CMove()->in(cmovd_ind1)
1783 || cmpd_pk->at(j)->in(2) != cmovd_pk->at(j)->as_CMove()->in(cmovd_ind2)) {
1784 return false;
1785 }//if
1786 }
1787 NOT_PRODUCT(if(_sw->is_trace_cmov()) { tty->print("CMoveKit::test_cmpd_pack: cmpd pack for 1st CmpD %d is OK for vectorization: ", cmpd0->_idx); cmpd0->dump(); })
1788 return true;
1789 }
1790
1791 //------------------------------implemented---------------------------
1792 // Can code be generated for pack p?
1793 bool SuperWord::implemented(Node_List* p) {
1794 bool retValue = false;
1795 Node* p0 = p->at(0);
1796 if (p0 != NULL) {
1797 int opc = p0->Opcode();
1798 uint size = p->size();
1799 if (p0->is_reduction()) {
1800 const Type *arith_type = p0->bottom_type();
1801 // Length 2 reductions of INT/LONG do not offer performance benefits
1802 if (((arith_type->basic_type() == T_INT) || (arith_type->basic_type() == T_LONG)) && (size == 2)) {
1803 retValue = false;
1804 } else {
1805 retValue = ReductionNode::implemented(opc, size, arith_type->basic_type());
1806 }
1807 } else {
1808 retValue = VectorNode::implemented(opc, size, velt_basic_type(p0));
1809 }
1810 if (!retValue) {
1811 if (is_cmov_pack(p)) {
1812 NOT_PRODUCT(if(is_trace_cmov()) {tty->print_cr("SWPointer::implemented: found cmpd pack"); print_pack(p);})
1813 return true;
1814 }
1815 }
1816 }
1817 return retValue;
1818 }
1819
1820 bool SuperWord::is_cmov_pack(Node_List* p) {
1821 return _cmovev_kit.pack(p->at(0)) != NULL;
1822 }
1823 //------------------------------same_inputs--------------------------
1824 // For pack p, are all idx operands the same?
1825 bool SuperWord::same_inputs(Node_List* p, int idx) {
1826 Node* p0 = p->at(0);
1827 uint vlen = p->size();
1828 Node* p0_def = p0->in(idx);
1829 for (uint i = 1; i < vlen; i++) {
1830 Node* pi = p->at(i);
1831 Node* pi_def = pi->in(idx);
1832 if (p0_def != pi_def) {
1833 return false;
1834 }
1835 }
1836 return true;
1837 }
1838
1839 //------------------------------profitable---------------------------
1840 // For pack p, are all operands and all uses (with in the block) vector?
1841 bool SuperWord::profitable(Node_List* p) {
1842 Node* p0 = p->at(0);
1843 uint start, end;
1844 VectorNode::vector_operands(p0, &start, &end);
1845
1846 // Return false if some inputs are not vectors or vectors with different
1847 // size or alignment.
1848 // Also, for now, return false if not scalar promotion case when inputs are
1849 // the same. Later, implement PackNode and allow differing, non-vector inputs
1850 // (maybe just the ones from outside the block.)
1851 for (uint i = start; i < end; i++) {
1852 if (!is_vector_use(p0, i)) {
1853 return false;
1854 }
1855 }
1856 // Check if reductions are connected
1857 if (p0->is_reduction()) {
1858 Node* second_in = p0->in(2);
1859 Node_List* second_pk = my_pack(second_in);
1860 if ((second_pk == NULL) || (_num_work_vecs == _num_reductions)) {
1861 // Remove reduction flag if no parent pack or if not enough work
1862 // to cover reduction expansion overhead
1863 p0->remove_flag(Node::Flag_is_reduction);
1864 return false;
1865 } else if (second_pk->size() != p->size()) {
1866 return false;
1867 }
1868 }
1869 if (VectorNode::is_shift(p0)) {
1870 // For now, return false if shift count is vector or not scalar promotion
1871 // case (different shift counts) because it is not supported yet.
1872 Node* cnt = p0->in(2);
1873 Node_List* cnt_pk = my_pack(cnt);
1874 if (cnt_pk != NULL)
1875 return false;
1876 if (!same_inputs(p, 2))
1877 return false;
1878 }
1879 if (!p0->is_Store()) {
1880 // For now, return false if not all uses are vector.
1881 // Later, implement ExtractNode and allow non-vector uses (maybe
1882 // just the ones outside the block.)
1883 for (uint i = 0; i < p->size(); i++) {
1884 Node* def = p->at(i);
1885 if (is_cmov_pack_internal_node(p, def)) {
1886 continue;
1887 }
1888 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
1889 Node* use = def->fast_out(j);
1890 for (uint k = 0; k < use->req(); k++) {
1891 Node* n = use->in(k);
1892 if (def == n) {
1893 // reductions can be loop carried dependences
1894 if (def->is_reduction() && use->is_Phi())
1895 continue;
1896 if (!is_vector_use(use, k)) {
1897 return false;
1898 }
1899 }
1900 }
1901 }
1902 }
1903 }
1904 return true;
1905 }
1906
1907 //------------------------------schedule---------------------------
1908 // Adjust the memory graph for the packed operations
1909 void SuperWord::schedule() {
1910
1911 // Co-locate in the memory graph the members of each memory pack
1912 for (int i = 0; i < _packset.length(); i++) {
1913 co_locate_pack(_packset.at(i));
1914 }
1915 }
1916
1917 //-------------------------------remove_and_insert-------------------
1918 // Remove "current" from its current position in the memory graph and insert
1919 // it after the appropriate insertion point (lip or uip).
1920 void SuperWord::remove_and_insert(MemNode *current, MemNode *prev, MemNode *lip,
1921 Node *uip, Unique_Node_List &sched_before) {
1922 Node* my_mem = current->in(MemNode::Memory);
1923 bool sched_up = sched_before.member(current);
1924
1925 // remove current_store from its current position in the memmory graph
1926 for (DUIterator i = current->outs(); current->has_out(i); i++) {
1927 Node* use = current->out(i);
1928 if (use->is_Mem()) {
1929 assert(use->in(MemNode::Memory) == current, "must be");
1930 if (use == prev) { // connect prev to my_mem
1931 _igvn.replace_input_of(use, MemNode::Memory, my_mem);
1932 --i; //deleted this edge; rescan position
1933 } else if (sched_before.member(use)) {
1934 if (!sched_up) { // Will be moved together with current
1935 _igvn.replace_input_of(use, MemNode::Memory, uip);
1936 --i; //deleted this edge; rescan position
1937 }
1938 } else {
1939 if (sched_up) { // Will be moved together with current
1940 _igvn.replace_input_of(use, MemNode::Memory, lip);
1941 --i; //deleted this edge; rescan position
1942 }
1943 }
1944 }
1945 }
1946
1947 Node *insert_pt = sched_up ? uip : lip;
1948
1949 // all uses of insert_pt's memory state should use current's instead
1950 for (DUIterator i = insert_pt->outs(); insert_pt->has_out(i); i++) {
1951 Node* use = insert_pt->out(i);
1952 if (use->is_Mem()) {
1953 assert(use->in(MemNode::Memory) == insert_pt, "must be");
1954 _igvn.replace_input_of(use, MemNode::Memory, current);
1955 --i; //deleted this edge; rescan position
1956 } else if (!sched_up && use->is_Phi() && use->bottom_type() == Type::MEMORY) {
1957 uint pos; //lip (lower insert point) must be the last one in the memory slice
1958 for (pos=1; pos < use->req(); pos++) {
1959 if (use->in(pos) == insert_pt) break;
1960 }
1961 _igvn.replace_input_of(use, pos, current);
1962 --i;
1963 }
1964 }
1965
1966 //connect current to insert_pt
1967 _igvn.replace_input_of(current, MemNode::Memory, insert_pt);
1968 }
1969
1970 //------------------------------co_locate_pack----------------------------------
1971 // To schedule a store pack, we need to move any sandwiched memory ops either before
1972 // or after the pack, based upon dependence information:
1973 // (1) If any store in the pack depends on the sandwiched memory op, the
1974 // sandwiched memory op must be scheduled BEFORE the pack;
1975 // (2) If a sandwiched memory op depends on any store in the pack, the
1976 // sandwiched memory op must be scheduled AFTER the pack;
1977 // (3) If a sandwiched memory op (say, memA) depends on another sandwiched
1978 // memory op (say memB), memB must be scheduled before memA. So, if memA is
1979 // scheduled before the pack, memB must also be scheduled before the pack;
1980 // (4) If there is no dependence restriction for a sandwiched memory op, we simply
1981 // schedule this store AFTER the pack
1982 // (5) We know there is no dependence cycle, so there in no other case;
1983 // (6) Finally, all memory ops in another single pack should be moved in the same direction.
1984 //
1985 // To schedule a load pack, we use the memory state of either the first or the last load in
1986 // the pack, based on the dependence constraint.
1987 void SuperWord::co_locate_pack(Node_List* pk) {
1988 if (pk->at(0)->is_Store()) {
1989 MemNode* first = executed_first(pk)->as_Mem();
1990 MemNode* last = executed_last(pk)->as_Mem();
1991 Unique_Node_List schedule_before_pack;
1992 Unique_Node_List memops;
1993
1994 MemNode* current = last->in(MemNode::Memory)->as_Mem();
1995 MemNode* previous = last;
1996 while (true) {
1997 assert(in_bb(current), "stay in block");
1998 memops.push(previous);
1999 for (DUIterator i = current->outs(); current->has_out(i); i++) {
2000 Node* use = current->out(i);
2001 if (use->is_Mem() && use != previous)
2002 memops.push(use);
2003 }
2004 if (current == first) break;
2005 previous = current;
2006 current = current->in(MemNode::Memory)->as_Mem();
2007 }
2008
2009 // determine which memory operations should be scheduled before the pack
2010 for (uint i = 1; i < memops.size(); i++) {
2011 Node *s1 = memops.at(i);
2012 if (!in_pack(s1, pk) && !schedule_before_pack.member(s1)) {
2013 for (uint j = 0; j< i; j++) {
2014 Node *s2 = memops.at(j);
2015 if (!independent(s1, s2)) {
2016 if (in_pack(s2, pk) || schedule_before_pack.member(s2)) {
2017 schedule_before_pack.push(s1); // s1 must be scheduled before
2018 Node_List* mem_pk = my_pack(s1);
2019 if (mem_pk != NULL) {
2020 for (uint ii = 0; ii < mem_pk->size(); ii++) {
2021 Node* s = mem_pk->at(ii); // follow partner
2022 if (memops.member(s) && !schedule_before_pack.member(s))
2023 schedule_before_pack.push(s);
2024 }
2025 }
2026 break;
2027 }
2028 }
2029 }
2030 }
2031 }
2032
2033 Node* upper_insert_pt = first->in(MemNode::Memory);
2034 // Following code moves loads connected to upper_insert_pt below aliased stores.
2035 // Collect such loads here and reconnect them back to upper_insert_pt later.
2036 memops.clear();
2037 for (DUIterator i = upper_insert_pt->outs(); upper_insert_pt->has_out(i); i++) {
2038 Node* use = upper_insert_pt->out(i);
2039 if (use->is_Mem() && !use->is_Store()) {
2040 memops.push(use);
2041 }
2042 }
2043
2044 MemNode* lower_insert_pt = last;
2045 previous = last; //previous store in pk
2046 current = last->in(MemNode::Memory)->as_Mem();
2047
2048 // start scheduling from "last" to "first"
2049 while (true) {
2050 assert(in_bb(current), "stay in block");
2051 assert(in_pack(previous, pk), "previous stays in pack");
2052 Node* my_mem = current->in(MemNode::Memory);
2053
2054 if (in_pack(current, pk)) {
2055 // Forward users of my memory state (except "previous) to my input memory state
2056 for (DUIterator i = current->outs(); current->has_out(i); i++) {
2057 Node* use = current->out(i);
2058 if (use->is_Mem() && use != previous) {
2059 assert(use->in(MemNode::Memory) == current, "must be");
2060 if (schedule_before_pack.member(use)) {
2061 _igvn.replace_input_of(use, MemNode::Memory, upper_insert_pt);
2062 } else {
2063 _igvn.replace_input_of(use, MemNode::Memory, lower_insert_pt);
2064 }
2065 --i; // deleted this edge; rescan position
2066 }
2067 }
2068 previous = current;
2069 } else { // !in_pack(current, pk) ==> a sandwiched store
2070 remove_and_insert(current, previous, lower_insert_pt, upper_insert_pt, schedule_before_pack);
2071 }
2072
2073 if (current == first) break;
2074 current = my_mem->as_Mem();
2075 } // end while
2076
2077 // Reconnect loads back to upper_insert_pt.
2078 for (uint i = 0; i < memops.size(); i++) {
2079 Node *ld = memops.at(i);
2080 if (ld->in(MemNode::Memory) != upper_insert_pt) {
2081 _igvn.replace_input_of(ld, MemNode::Memory, upper_insert_pt);
2082 }
2083 }
2084 } else if (pk->at(0)->is_Load()) { //load
2085 // all loads in the pack should have the same memory state. By default,
2086 // we use the memory state of the last load. However, if any load could
2087 // not be moved down due to the dependence constraint, we use the memory
2088 // state of the first load.
2089 Node* last_mem = executed_last(pk)->in(MemNode::Memory);
2090 Node* first_mem = executed_first(pk)->in(MemNode::Memory);
2091 bool schedule_last = true;
2092 for (uint i = 0; i < pk->size(); i++) {
2093 Node* ld = pk->at(i);
2094 for (Node* current = last_mem; current != ld->in(MemNode::Memory);
2095 current=current->in(MemNode::Memory)) {
2096 assert(current != first_mem, "corrupted memory graph");
2097 if(current->is_Mem() && !independent(current, ld)){
2098 schedule_last = false; // a later store depends on this load
2099 break;
2100 }
2101 }
2102 }
2103
2104 Node* mem_input = schedule_last ? last_mem : first_mem;
2105 _igvn.hash_delete(mem_input);
2106 // Give each load the same memory state
2107 for (uint i = 0; i < pk->size(); i++) {
2108 LoadNode* ld = pk->at(i)->as_Load();
2109 _igvn.replace_input_of(ld, MemNode::Memory, mem_input);
2110 }
2111 }
2112 }
2113
2114 #ifndef PRODUCT
2115 void SuperWord::print_loop(bool whole) {
2116 Node_Stack stack(_arena, _phase->C->unique() >> 2);
2117 Node_List rpo_list;
2118 VectorSet visited(_arena);
2119 visited.set(lpt()->_head->_idx);
2120 _phase->rpo(lpt()->_head, stack, visited, rpo_list);
2121 _phase->dump(lpt(), rpo_list.size(), rpo_list );
2122 if(whole) {
2123 tty->print_cr("\n Whole loop tree");
2124 _phase->dump();
2125 tty->print_cr(" End of whole loop tree\n");
2126 }
2127 }
2128 #endif
2129
2130 //------------------------------output---------------------------
2131 // Convert packs into vector node operations
2132 void SuperWord::output() {
2133 if (_packset.length() == 0) return;
2134
2135 #ifndef PRODUCT
2136 if (TraceLoopOpts) {
2137 tty->print("SuperWord::output ");
2138 lpt()->dump_head();
2139 }
2140 #endif
2141
2142 CountedLoopNode *cl = lpt()->_head->as_CountedLoop();
2143 if (cl->is_main_loop()) {
2144 // MUST ENSURE main loop's initial value is properly aligned:
2145 // (iv_initial_value + min_iv_offset) % vector_width_in_bytes() == 0
2146
2147 align_initial_loop_index(align_to_ref());
2148
2149 // Insert extract (unpack) operations for scalar uses
2150 for (int i = 0; i < _packset.length(); i++) {
2151 insert_extracts(_packset.at(i));
2152 }
2153 }
2154
2155 Compile* C = _phase->C;
2156 uint max_vlen_in_bytes = 0;
2157 uint max_vlen = 0;
2158 bool can_process_post_loop = (PostLoopMultiversioning && Matcher::has_predicated_vectors() && cl->is_post_loop());
2159
2160 NOT_PRODUCT(if(is_trace_loop_reverse()) {tty->print_cr("SWPointer::output: print loop before create_reserve_version_of_loop"); print_loop(true);})
2161
2162 CountedLoopReserveKit make_reversable(_phase, _lpt, do_reserve_copy());
2163
2164 NOT_PRODUCT(if(is_trace_loop_reverse()) {tty->print_cr("SWPointer::output: print loop after create_reserve_version_of_loop"); print_loop(true);})
2165
2166 if (do_reserve_copy() && !make_reversable.has_reserved()) {
2167 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: loop was not reserved correctly, exiting SuperWord");})
2168 return;
2169 }
2170
2171 for (int i = 0; i < _block.length(); i++) {
2172 Node* n = _block.at(i);
2173 Node_List* p = my_pack(n);
2174 if (p && n == executed_last(p)) {
2175 uint vlen = p->size();
2176 uint vlen_in_bytes = 0;
2177 Node* vn = NULL;
2178 Node* low_adr = p->at(0);
2179 Node* first = executed_first(p);
2180 if (can_process_post_loop) {
2181 // override vlen with the main loops vector length
2182 vlen = cl->slp_max_unroll();
2183 }
2184 NOT_PRODUCT(if(is_trace_cmov()) {tty->print_cr("SWPointer::output: %d executed first, %d executed last in pack", first->_idx, n->_idx); print_pack(p);})
2185 int opc = n->Opcode();
2186 if (n->is_Load()) {
2187 Node* ctl = n->in(MemNode::Control);
2188 Node* mem = first->in(MemNode::Memory);
2189 SWPointer p1(n->as_Mem(), this, NULL, false);
2190 // Identify the memory dependency for the new loadVector node by
2191 // walking up through memory chain.
2192 // This is done to give flexibility to the new loadVector node so that
2193 // it can move above independent storeVector nodes.
2194 while (mem->is_StoreVector()) {
2195 SWPointer p2(mem->as_Mem(), this, NULL, false);
2196 int cmp = p1.cmp(p2);
2197 if (SWPointer::not_equal(cmp) || !SWPointer::comparable(cmp)) {
2198 mem = mem->in(MemNode::Memory);
2199 } else {
2200 break; // dependent memory
2201 }
2202 }
2203 Node* adr = low_adr->in(MemNode::Address);
2204 const TypePtr* atyp = n->adr_type();
2205 vn = LoadVectorNode::make(opc, ctl, mem, adr, atyp, vlen, velt_basic_type(n), control_dependency(p));
2206 vlen_in_bytes = vn->as_LoadVector()->memory_size();
2207 } else if (n->is_Store()) {
2208 // Promote value to be stored to vector
2209 Node* val = vector_opd(p, MemNode::ValueIn);
2210 if (val == NULL) {
2211 if (do_reserve_copy()) {
2212 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: val should not be NULL, exiting SuperWord");})
2213 return; //and reverse to backup IG
2214 }
2215 ShouldNotReachHere();
2216 }
2217
2218 Node* ctl = n->in(MemNode::Control);
2219 Node* mem = first->in(MemNode::Memory);
2220 Node* adr = low_adr->in(MemNode::Address);
2221 const TypePtr* atyp = n->adr_type();
2222 vn = StoreVectorNode::make(opc, ctl, mem, adr, atyp, val, vlen);
2223 vlen_in_bytes = vn->as_StoreVector()->memory_size();
2224 } else if (n->req() == 3 && !is_cmov_pack(p)) {
2225 // Promote operands to vector
2226 Node* in1 = NULL;
2227 bool node_isa_reduction = n->is_reduction();
2228 if (node_isa_reduction) {
2229 // the input to the first reduction operation is retained
2230 in1 = low_adr->in(1);
2231 } else {
2232 in1 = vector_opd(p, 1);
2233 if (in1 == NULL) {
2234 if (do_reserve_copy()) {
2235 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: in1 should not be NULL, exiting SuperWord");})
2236 return; //and reverse to backup IG
2237 }
2238 ShouldNotReachHere();
2239 }
2240 }
2241 Node* in2 = vector_opd(p, 2);
2242 if (in2 == NULL) {
2243 if (do_reserve_copy()) {
2244 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: in2 should not be NULL, exiting SuperWord");})
2245 return; //and reverse to backup IG
2246 }
2247 ShouldNotReachHere();
2248 }
2249 if (VectorNode::is_invariant_vector(in1) && (node_isa_reduction == false) && (n->is_Add() || n->is_Mul())) {
2250 // Move invariant vector input into second position to avoid register spilling.
2251 Node* tmp = in1;
2252 in1 = in2;
2253 in2 = tmp;
2254 }
2255 if (node_isa_reduction) {
2256 const Type *arith_type = n->bottom_type();
2257 vn = ReductionNode::make(opc, NULL, in1, in2, arith_type->basic_type());
2258 if (in2->is_Load()) {
2259 vlen_in_bytes = in2->as_LoadVector()->memory_size();
2260 } else {
2261 vlen_in_bytes = in2->as_Vector()->length_in_bytes();
2262 }
2263 } else {
2264 vn = VectorNode::make(opc, in1, in2, vlen, velt_basic_type(n));
2265 vlen_in_bytes = vn->as_Vector()->length_in_bytes();
2266 }
2267 } else if (opc == Op_SqrtD || opc == Op_AbsF || opc == Op_AbsD || opc == Op_NegF || opc == Op_NegD) {
2268 // Promote operand to vector (Sqrt/Abs/Neg are 2 address instructions)
2269 Node* in = vector_opd(p, 1);
2270 vn = VectorNode::make(opc, in, NULL, vlen, velt_basic_type(n));
2271 vlen_in_bytes = vn->as_Vector()->length_in_bytes();
2272 } else if (is_cmov_pack(p)) {
2273 if (can_process_post_loop) {
2274 // do not refactor of flow in post loop context
2275 return;
2276 }
2277 if (!n->is_CMove()) {
2278 continue;
2279 }
2280 // place here CMoveVDNode
2281 NOT_PRODUCT(if(is_trace_cmov()) {tty->print_cr("SWPointer::output: print before CMove vectorization"); print_loop(false);})
2282 Node* bol = n->in(CMoveNode::Condition);
2283 if (!bol->is_Bool() && bol->Opcode() == Op_ExtractI && bol->req() > 1 ) {
2284 NOT_PRODUCT(if(is_trace_cmov()) {tty->print_cr("SWPointer::output: %d is not Bool node, trying its in(1) node %d", bol->_idx, bol->in(1)->_idx); bol->dump(); bol->in(1)->dump();})
2285 bol = bol->in(1); //may be ExtractNode
2286 }
2287
2288 assert(bol->is_Bool(), "should be BoolNode - too late to bail out!");
2289 if (!bol->is_Bool()) {
2290 if (do_reserve_copy()) {
2291 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: expected %d bool node, exiting SuperWord", bol->_idx); bol->dump();})
2292 return; //and reverse to backup IG
2293 }
2294 ShouldNotReachHere();
2295 }
2296
2297 int cond = (int)bol->as_Bool()->_test._test;
2298 Node* in_cc = _igvn.intcon(cond);
2299 NOT_PRODUCT(if(is_trace_cmov()) {tty->print("SWPointer::output: created intcon in_cc node %d", in_cc->_idx); in_cc->dump();})
2300 Node* cc = bol->clone();
2301 cc->set_req(1, in_cc);
2302 NOT_PRODUCT(if(is_trace_cmov()) {tty->print("SWPointer::output: created bool cc node %d", cc->_idx); cc->dump();})
2303
2304 Node* src1 = vector_opd(p, 2); //2=CMoveNode::IfFalse
2305 if (src1 == NULL) {
2306 if (do_reserve_copy()) {
2307 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: src1 should not be NULL, exiting SuperWord");})
2308 return; //and reverse to backup IG
2309 }
2310 ShouldNotReachHere();
2311 }
2312 Node* src2 = vector_opd(p, 3); //3=CMoveNode::IfTrue
2313 if (src2 == NULL) {
2314 if (do_reserve_copy()) {
2315 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: src2 should not be NULL, exiting SuperWord");})
2316 return; //and reverse to backup IG
2317 }
2318 ShouldNotReachHere();
2319 }
2320 BasicType bt = velt_basic_type(n);
2321 const TypeVect* vt = TypeVect::make(bt, vlen);
2322 vn = new CMoveVDNode(cc, src1, src2, vt);
2323 NOT_PRODUCT(if(is_trace_cmov()) {tty->print("SWPointer::output: created new CMove node %d: ", vn->_idx); vn->dump();})
2324 } else {
2325 if (do_reserve_copy()) {
2326 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: ShouldNotReachHere, exiting SuperWord");})
2327 return; //and reverse to backup IG
2328 }
2329 ShouldNotReachHere();
2330 }
2331
2332 assert(vn != NULL, "sanity");
2333 if (vn == NULL) {
2334 if (do_reserve_copy()){
2335 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("SWPointer::output: got NULL node, cannot proceed, exiting SuperWord");})
2336 return; //and reverse to backup IG
2337 }
2338 ShouldNotReachHere();
2339 }
2340
2341 _block.at_put(i, vn);
2342 _igvn.register_new_node_with_optimizer(vn);
2343 _phase->set_ctrl(vn, _phase->get_ctrl(p->at(0)));
2344 for (uint j = 0; j < p->size(); j++) {
2345 Node* pm = p->at(j);
2346 _igvn.replace_node(pm, vn);
2347 }
2348 _igvn._worklist.push(vn);
2349
2350 if (can_process_post_loop) {
2351 // first check if the vector size if the maximum vector which we can use on the machine,
2352 // other vector size have reduced values for predicated data mapping.
2353 if (vlen_in_bytes != (uint)MaxVectorSize) {
2354 return;
2355 }
2356 }
2357
2358 if (vlen_in_bytes > max_vlen_in_bytes) {
2359 max_vlen = vlen;
2360 max_vlen_in_bytes = vlen_in_bytes;
2361 }
2362 #ifdef ASSERT
2363 if (TraceNewVectors) {
2364 tty->print("new Vector node: ");
2365 vn->dump();
2366 }
2367 #endif
2368 }
2369 }//for (int i = 0; i < _block.length(); i++)
2370
2371 C->set_max_vector_size(max_vlen_in_bytes);
2372
2373 if (SuperWordLoopUnrollAnalysis) {
2374 if (cl->has_passed_slp()) {
2375 uint slp_max_unroll_factor = cl->slp_max_unroll();
2376 if (slp_max_unroll_factor == max_vlen) {
2377 if (TraceSuperWordLoopUnrollAnalysis) {
2378 tty->print_cr("vector loop(unroll=%d, len=%d)\n", max_vlen, max_vlen_in_bytes*BitsPerByte);
2379 }
2380
2381 // For atomic unrolled loops which are vector mapped, instigate more unrolling
2382 cl->set_notpassed_slp();
2383 if (cl->is_main_loop()) {
2384 // if vector resources are limited, do not allow additional unrolling, also
2385 // do not unroll more on pure vector loops which were not reduced so that we can
2386 // program the post loop to single iteration execution.
2387 if (FLOATPRESSURE > 8) {
2388 C->set_major_progress();
2389 cl->mark_do_unroll_only();
2390 }
2391 }
2392
2393 if (do_reserve_copy()) {
2394 cl->mark_loop_vectorized();
2395 if (can_process_post_loop) {
2396 // Now create the difference of trip and limit and use it as our mask index.
2397 // Note: We limited the unroll of the vectorized loop so that
2398 // only vlen-1 size iterations can remain to be mask programmed.
2399 Node *incr = cl->incr();
2400 SubINode *index = new SubINode(cl->limit(), cl->init_trip());
2401 _igvn.register_new_node_with_optimizer(index);
2402 MaskCreateINode *mask = new MaskCreateINode(_phase->get_ctrl(cl->init_trip()), index);
2403 _igvn.register_new_node_with_optimizer(mask);
2404 // make this a single iteration loop
2405 AddINode *new_incr = new AddINode(incr->in(1), mask);
2406 _igvn.register_new_node_with_optimizer(new_incr);
2407 _phase->set_ctrl(new_incr, _phase->get_ctrl(incr));
2408 _igvn.replace_node(incr, new_incr);
2409 cl->mark_is_multiversioned();
2410 }
2411 }
2412 }
2413 }
2414 }
2415
2416 if (do_reserve_copy()) {
2417 make_reversable.use_new();
2418 }
2419 NOT_PRODUCT(if(is_trace_loop_reverse()) {tty->print_cr("\n Final loop after SuperWord"); print_loop(true);})
2420 return;
2421 }
2422
2423 //------------------------------vector_opd---------------------------
2424 // Create a vector operand for the nodes in pack p for operand: in(opd_idx)
2425 Node* SuperWord::vector_opd(Node_List* p, int opd_idx) {
2426 Node* p0 = p->at(0);
2427 uint vlen = p->size();
2428 Node* opd = p0->in(opd_idx);
2429 CountedLoopNode *cl = lpt()->_head->as_CountedLoop();
2430
2431 if (PostLoopMultiversioning && Matcher::has_predicated_vectors() && cl->is_post_loop()) {
2432 // override vlen with the main loops vector length
2433 vlen = cl->slp_max_unroll();
2434 }
2435
2436 if (same_inputs(p, opd_idx)) {
2437 if (opd->is_Vector() || opd->is_LoadVector()) {
2438 assert(((opd_idx != 2) || !VectorNode::is_shift(p0)), "shift's count can't be vector");
2439 if (opd_idx == 2 && VectorNode::is_shift(p0)) {
2440 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("shift's count can't be vector");})
2441 return NULL;
2442 }
2443 return opd; // input is matching vector
2444 }
2445 if ((opd_idx == 2) && VectorNode::is_shift(p0)) {
2446 Compile* C = _phase->C;
2447 Node* cnt = opd;
2448 // Vector instructions do not mask shift count, do it here.
2449 juint mask = (p0->bottom_type() == TypeInt::INT) ? (BitsPerInt - 1) : (BitsPerLong - 1);
2450 const TypeInt* t = opd->find_int_type();
2451 if (t != NULL && t->is_con()) {
2452 juint shift = t->get_con();
2453 if (shift > mask) { // Unsigned cmp
2454 cnt = ConNode::make(TypeInt::make(shift & mask));
2455 }
2456 } else {
2457 if (t == NULL || t->_lo < 0 || t->_hi > (int)mask) {
2458 cnt = ConNode::make(TypeInt::make(mask));
2459 _igvn.register_new_node_with_optimizer(cnt);
2460 cnt = new AndINode(opd, cnt);
2461 _igvn.register_new_node_with_optimizer(cnt);
2462 _phase->set_ctrl(cnt, _phase->get_ctrl(opd));
2463 }
2464 assert(opd->bottom_type()->isa_int(), "int type only");
2465 if (!opd->bottom_type()->isa_int()) {
2466 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("Should be int type only");})
2467 return NULL;
2468 }
2469 // Move non constant shift count into vector register.
2470 cnt = VectorNode::shift_count(p0, cnt, vlen, velt_basic_type(p0));
2471 }
2472 if (cnt != opd) {
2473 _igvn.register_new_node_with_optimizer(cnt);
2474 _phase->set_ctrl(cnt, _phase->get_ctrl(opd));
2475 }
2476 return cnt;
2477 }
2478 assert(!opd->is_StoreVector(), "such vector is not expected here");
2479 if (opd->is_StoreVector()) {
2480 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("StoreVector is not expected here");})
2481 return NULL;
2482 }
2483 // Convert scalar input to vector with the same number of elements as
2484 // p0's vector. Use p0's type because size of operand's container in
2485 // vector should match p0's size regardless operand's size.
2486 const Type* p0_t = velt_type(p0);
2487 VectorNode* vn = VectorNode::scalar2vector(opd, vlen, p0_t);
2488
2489 _igvn.register_new_node_with_optimizer(vn);
2490 _phase->set_ctrl(vn, _phase->get_ctrl(opd));
2491 #ifdef ASSERT
2492 if (TraceNewVectors) {
2493 tty->print("new Vector node: ");
2494 vn->dump();
2495 }
2496 #endif
2497 return vn;
2498 }
2499
2500 // Insert pack operation
2501 BasicType bt = velt_basic_type(p0);
2502 PackNode* pk = PackNode::make(opd, vlen, bt);
2503 DEBUG_ONLY( const BasicType opd_bt = opd->bottom_type()->basic_type(); )
2504
2505 for (uint i = 1; i < vlen; i++) {
2506 Node* pi = p->at(i);
2507 Node* in = pi->in(opd_idx);
2508 assert(my_pack(in) == NULL, "Should already have been unpacked");
2509 if (my_pack(in) != NULL) {
2510 NOT_PRODUCT(if(is_trace_loop_reverse() || TraceLoopOpts) {tty->print_cr("Should already have been unpacked");})
2511 return NULL;
2512 }
2513 assert(opd_bt == in->bottom_type()->basic_type(), "all same type");
2514 pk->add_opd(in);
2515 }
2516 _igvn.register_new_node_with_optimizer(pk);
2517 _phase->set_ctrl(pk, _phase->get_ctrl(opd));
2518 #ifdef ASSERT
2519 if (TraceNewVectors) {
2520 tty->print("new Vector node: ");
2521 pk->dump();
2522 }
2523 #endif
2524 return pk;
2525 }
2526
2527 //------------------------------insert_extracts---------------------------
2528 // If a use of pack p is not a vector use, then replace the
2529 // use with an extract operation.
2530 void SuperWord::insert_extracts(Node_List* p) {
2531 if (p->at(0)->is_Store()) return;
2532 assert(_n_idx_list.is_empty(), "empty (node,index) list");
2533
2534 // Inspect each use of each pack member. For each use that is
2535 // not a vector use, replace the use with an extract operation.
2536
2537 for (uint i = 0; i < p->size(); i++) {
2538 Node* def = p->at(i);
2539 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
2540 Node* use = def->fast_out(j);
2541 for (uint k = 0; k < use->req(); k++) {
2542 Node* n = use->in(k);
2543 if (def == n) {
2544 Node_List* u_pk = my_pack(use);
2545 if ((u_pk == NULL || !is_cmov_pack(u_pk) || use->is_CMove()) && !is_vector_use(use, k)) {
2546 _n_idx_list.push(use, k);
2547 }
2548 }
2549 }
2550 }
2551 }
2552
2553 while (_n_idx_list.is_nonempty()) {
2554 Node* use = _n_idx_list.node();
2555 int idx = _n_idx_list.index();
2556 _n_idx_list.pop();
2557 Node* def = use->in(idx);
2558
2559 if (def->is_reduction()) continue;
2560
2561 // Insert extract operation
2562 _igvn.hash_delete(def);
2563 int def_pos = alignment(def) / data_size(def);
2564
2565 Node* ex = ExtractNode::make(def, def_pos, velt_basic_type(def));
2566 _igvn.register_new_node_with_optimizer(ex);
2567 _phase->set_ctrl(ex, _phase->get_ctrl(def));
2568 _igvn.replace_input_of(use, idx, ex);
2569 _igvn._worklist.push(def);
2570
2571 bb_insert_after(ex, bb_idx(def));
2572 set_velt_type(ex, velt_type(def));
2573 }
2574 }
2575
2576 //------------------------------is_vector_use---------------------------
2577 // Is use->in(u_idx) a vector use?
2578 bool SuperWord::is_vector_use(Node* use, int u_idx) {
2579 Node_List* u_pk = my_pack(use);
2580 if (u_pk == NULL) return false;
2581 if (use->is_reduction()) return true;
2582 Node* def = use->in(u_idx);
2583 Node_List* d_pk = my_pack(def);
2584 if (d_pk == NULL) {
2585 // check for scalar promotion
2586 Node* n = u_pk->at(0)->in(u_idx);
2587 for (uint i = 1; i < u_pk->size(); i++) {
2588 if (u_pk->at(i)->in(u_idx) != n) return false;
2589 }
2590 return true;
2591 }
2592 if (u_pk->size() != d_pk->size())
2593 return false;
2594 for (uint i = 0; i < u_pk->size(); i++) {
2595 Node* ui = u_pk->at(i);
2596 Node* di = d_pk->at(i);
2597 if (ui->in(u_idx) != di || alignment(ui) != alignment(di))
2598 return false;
2599 }
2600 return true;
2601 }
2602
2603 //------------------------------construct_bb---------------------------
2604 // Construct reverse postorder list of block members
2605 bool SuperWord::construct_bb() {
2606 Node* entry = bb();
2607
2608 assert(_stk.length() == 0, "stk is empty");
2609 assert(_block.length() == 0, "block is empty");
2610 assert(_data_entry.length() == 0, "data_entry is empty");
2611 assert(_mem_slice_head.length() == 0, "mem_slice_head is empty");
2612 assert(_mem_slice_tail.length() == 0, "mem_slice_tail is empty");
2613
2614 // Find non-control nodes with no inputs from within block,
2615 // create a temporary map from node _idx to bb_idx for use
2616 // by the visited and post_visited sets,
2617 // and count number of nodes in block.
2618 int bb_ct = 0;
2619 for (uint i = 0; i < lpt()->_body.size(); i++) {
2620 Node *n = lpt()->_body.at(i);
2621 set_bb_idx(n, i); // Create a temporary map
2622 if (in_bb(n)) {
2623 if (n->is_LoadStore() || n->is_MergeMem() ||
2624 (n->is_Proj() && !n->as_Proj()->is_CFG())) {
2625 // Bailout if the loop has LoadStore, MergeMem or data Proj
2626 // nodes. Superword optimization does not work with them.
2627 return false;
2628 }
2629 bb_ct++;
2630 if (!n->is_CFG()) {
2631 bool found = false;
2632 for (uint j = 0; j < n->req(); j++) {
2633 Node* def = n->in(j);
2634 if (def && in_bb(def)) {
2635 found = true;
2636 break;
2637 }
2638 }
2639 if (!found) {
2640 assert(n != entry, "can't be entry");
2641 _data_entry.push(n);
2642 }
2643 }
2644 }
2645 }
2646
2647 // Find memory slices (head and tail)
2648 for (DUIterator_Fast imax, i = lp()->fast_outs(imax); i < imax; i++) {
2649 Node *n = lp()->fast_out(i);
2650 if (in_bb(n) && (n->is_Phi() && n->bottom_type() == Type::MEMORY)) {
2651 Node* n_tail = n->in(LoopNode::LoopBackControl);
2652 if (n_tail != n->in(LoopNode::EntryControl)) {
2653 if (!n_tail->is_Mem()) {
2654 assert(n_tail->is_Mem(), "unexpected node for memory slice: %s", n_tail->Name());
2655 return false; // Bailout
2656 }
2657 _mem_slice_head.push(n);
2658 _mem_slice_tail.push(n_tail);
2659 }
2660 }
2661 }
2662
2663 // Create an RPO list of nodes in block
2664
2665 visited_clear();
2666 post_visited_clear();
2667
2668 // Push all non-control nodes with no inputs from within block, then control entry
2669 for (int j = 0; j < _data_entry.length(); j++) {
2670 Node* n = _data_entry.at(j);
2671 visited_set(n);
2672 _stk.push(n);
2673 }
2674 visited_set(entry);
2675 _stk.push(entry);
2676
2677 // Do a depth first walk over out edges
2678 int rpo_idx = bb_ct - 1;
2679 int size;
2680 int reduction_uses = 0;
2681 while ((size = _stk.length()) > 0) {
2682 Node* n = _stk.top(); // Leave node on stack
2683 if (!visited_test_set(n)) {
2684 // forward arc in graph
2685 } else if (!post_visited_test(n)) {
2686 // cross or back arc
2687 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2688 Node *use = n->fast_out(i);
2689 if (in_bb(use) && !visited_test(use) &&
2690 // Don't go around backedge
2691 (!use->is_Phi() || n == entry)) {
2692 if (use->is_reduction()) {
2693 // First see if we can map the reduction on the given system we are on, then
2694 // make a data entry operation for each reduction we see.
2695 BasicType bt = use->bottom_type()->basic_type();
2696 if (ReductionNode::implemented(use->Opcode(), Matcher::min_vector_size(bt), bt)) {
2697 reduction_uses++;
2698 }
2699 }
2700 _stk.push(use);
2701 }
2702 }
2703 if (_stk.length() == size) {
2704 // There were no additional uses, post visit node now
2705 _stk.pop(); // Remove node from stack
2706 assert(rpo_idx >= 0, "");
2707 _block.at_put_grow(rpo_idx, n);
2708 rpo_idx--;
2709 post_visited_set(n);
2710 assert(rpo_idx >= 0 || _stk.is_empty(), "");
2711 }
2712 } else {
2713 _stk.pop(); // Remove post-visited node from stack
2714 }
2715 }//while
2716
2717 int ii_current = -1;
2718 unsigned int load_idx = (unsigned int)-1;
2719 _ii_order.clear();
2720 // Create real map of block indices for nodes
2721 for (int j = 0; j < _block.length(); j++) {
2722 Node* n = _block.at(j);
2723 set_bb_idx(n, j);
2724 if (_do_vector_loop && n->is_Load()) {
2725 if (ii_current == -1) {
2726 ii_current = _clone_map.gen(n->_idx);
2727 _ii_order.push(ii_current);
2728 load_idx = _clone_map.idx(n->_idx);
2729 } else if (_clone_map.idx(n->_idx) == load_idx && _clone_map.gen(n->_idx) != ii_current) {
2730 ii_current = _clone_map.gen(n->_idx);
2731 _ii_order.push(ii_current);
2732 }
2733 }
2734 }//for
2735
2736 // Ensure extra info is allocated.
2737 initialize_bb();
2738
2739 #ifndef PRODUCT
2740 if (_vector_loop_debug && _ii_order.length() > 0) {
2741 tty->print("SuperWord::construct_bb: List of generations: ");
2742 for (int jj = 0; jj < _ii_order.length(); ++jj) {
2743 tty->print(" %d:%d", jj, _ii_order.at(jj));
2744 }
2745 tty->print_cr(" ");
2746 }
2747 if (TraceSuperWord) {
2748 print_bb();
2749 tty->print_cr("\ndata entry nodes: %s", _data_entry.length() > 0 ? "" : "NONE");
2750 for (int m = 0; m < _data_entry.length(); m++) {
2751 tty->print("%3d ", m);
2752 _data_entry.at(m)->dump();
2753 }
2754 tty->print_cr("\nmemory slices: %s", _mem_slice_head.length() > 0 ? "" : "NONE");
2755 for (int m = 0; m < _mem_slice_head.length(); m++) {
2756 tty->print("%3d ", m); _mem_slice_head.at(m)->dump();
2757 tty->print(" "); _mem_slice_tail.at(m)->dump();
2758 }
2759 }
2760 #endif
2761 assert(rpo_idx == -1 && bb_ct == _block.length(), "all block members found");
2762 return (_mem_slice_head.length() > 0) || (reduction_uses > 0) || (_data_entry.length() > 0);
2763 }
2764
2765 //------------------------------initialize_bb---------------------------
2766 // Initialize per node info
2767 void SuperWord::initialize_bb() {
2768 Node* last = _block.at(_block.length() - 1);
2769 grow_node_info(bb_idx(last));
2770 }
2771
2772 //------------------------------bb_insert_after---------------------------
2773 // Insert n into block after pos
2774 void SuperWord::bb_insert_after(Node* n, int pos) {
2775 int n_pos = pos + 1;
2776 // Make room
2777 for (int i = _block.length() - 1; i >= n_pos; i--) {
2778 _block.at_put_grow(i+1, _block.at(i));
2779 }
2780 for (int j = _node_info.length() - 1; j >= n_pos; j--) {
2781 _node_info.at_put_grow(j+1, _node_info.at(j));
2782 }
2783 // Set value
2784 _block.at_put_grow(n_pos, n);
2785 _node_info.at_put_grow(n_pos, SWNodeInfo::initial);
2786 // Adjust map from node->_idx to _block index
2787 for (int i = n_pos; i < _block.length(); i++) {
2788 set_bb_idx(_block.at(i), i);
2789 }
2790 }
2791
2792 //------------------------------compute_max_depth---------------------------
2793 // Compute max depth for expressions from beginning of block
2794 // Use to prune search paths during test for independence.
2795 void SuperWord::compute_max_depth() {
2796 int ct = 0;
2797 bool again;
2798 do {
2799 again = false;
2800 for (int i = 0; i < _block.length(); i++) {
2801 Node* n = _block.at(i);
2802 if (!n->is_Phi()) {
2803 int d_orig = depth(n);
2804 int d_in = 0;
2805 for (DepPreds preds(n, _dg); !preds.done(); preds.next()) {
2806 Node* pred = preds.current();
2807 if (in_bb(pred)) {
2808 d_in = MAX2(d_in, depth(pred));
2809 }
2810 }
2811 if (d_in + 1 != d_orig) {
2812 set_depth(n, d_in + 1);
2813 again = true;
2814 }
2815 }
2816 }
2817 ct++;
2818 } while (again);
2819
2820 if (TraceSuperWord && Verbose) {
2821 tty->print_cr("compute_max_depth iterated: %d times", ct);
2822 }
2823 }
2824
2825 //-------------------------compute_vector_element_type-----------------------
2826 // Compute necessary vector element type for expressions
2827 // This propagates backwards a narrower integer type when the
2828 // upper bits of the value are not needed.
2829 // Example: char a,b,c; a = b + c;
2830 // Normally the type of the add is integer, but for packed character
2831 // operations the type of the add needs to be char.
2832 void SuperWord::compute_vector_element_type() {
2833 if (TraceSuperWord && Verbose) {
2834 tty->print_cr("\ncompute_velt_type:");
2835 }
2836
2837 // Initial type
2838 for (int i = 0; i < _block.length(); i++) {
2839 Node* n = _block.at(i);
2840 set_velt_type(n, container_type(n));
2841 }
2842
2843 // Propagate integer narrowed type backwards through operations
2844 // that don't depend on higher order bits
2845 for (int i = _block.length() - 1; i >= 0; i--) {
2846 Node* n = _block.at(i);
2847 // Only integer types need be examined
2848 const Type* vtn = velt_type(n);
2849 if (vtn->basic_type() == T_INT) {
2850 uint start, end;
2851 VectorNode::vector_operands(n, &start, &end);
2852
2853 for (uint j = start; j < end; j++) {
2854 Node* in = n->in(j);
2855 // Don't propagate through a memory
2856 if (!in->is_Mem() && in_bb(in) && velt_type(in)->basic_type() == T_INT &&
2857 data_size(n) < data_size(in)) {
2858 bool same_type = true;
2859 for (DUIterator_Fast kmax, k = in->fast_outs(kmax); k < kmax; k++) {
2860 Node *use = in->fast_out(k);
2861 if (!in_bb(use) || !same_velt_type(use, n)) {
2862 same_type = false;
2863 break;
2864 }
2865 }
2866 if (same_type) {
2867 // For right shifts of small integer types (bool, byte, char, short)
2868 // we need precise information about sign-ness. Only Load nodes have
2869 // this information because Store nodes are the same for signed and
2870 // unsigned values. And any arithmetic operation after a load may
2871 // expand a value to signed Int so such right shifts can't be used
2872 // because vector elements do not have upper bits of Int.
2873 const Type* vt = vtn;
2874 if (VectorNode::is_shift(in)) {
2875 Node* load = in->in(1);
2876 if (load->is_Load() && in_bb(load) && (velt_type(load)->basic_type() == T_INT)) {
2877 vt = velt_type(load);
2878 } else if (in->Opcode() != Op_LShiftI) {
2879 // Widen type to Int to avoid creation of right shift vector
2880 // (align + data_size(s1) check in stmts_can_pack() will fail).
2881 // Note, left shifts work regardless type.
2882 vt = TypeInt::INT;
2883 }
2884 }
2885 set_velt_type(in, vt);
2886 }
2887 }
2888 }
2889 }
2890 }
2891 #ifndef PRODUCT
2892 if (TraceSuperWord && Verbose) {
2893 for (int i = 0; i < _block.length(); i++) {
2894 Node* n = _block.at(i);
2895 velt_type(n)->dump();
2896 tty->print("\t");
2897 n->dump();
2898 }
2899 }
2900 #endif
2901 }
2902
2903 //------------------------------memory_alignment---------------------------
2904 // Alignment within a vector memory reference
2905 int SuperWord::memory_alignment(MemNode* s, int iv_adjust) {
2906 #ifndef PRODUCT
2907 if(TraceSuperWord && Verbose) {
2908 tty->print("SuperWord::memory_alignment within a vector memory reference for %d: ", s->_idx); s->dump();
2909 }
2910 #endif
2911 NOT_PRODUCT(SWPointer::Tracer::Depth ddd(0);)
2912 SWPointer p(s, this, NULL, false);
2913 if (!p.valid()) {
2914 NOT_PRODUCT(if(is_trace_alignment()) tty->print("SWPointer::memory_alignment: SWPointer p invalid, return bottom_align");)
2915 return bottom_align;
2916 }
2917 int vw = vector_width_in_bytes(s);
2918 if (vw < 2) {
2919 NOT_PRODUCT(if(is_trace_alignment()) tty->print_cr("SWPointer::memory_alignment: vector_width_in_bytes < 2, return bottom_align");)
2920 return bottom_align; // No vectors for this type
2921 }
2922 int offset = p.offset_in_bytes();
2923 offset += iv_adjust*p.memory_size();
2924 int off_rem = offset % vw;
2925 int off_mod = off_rem >= 0 ? off_rem : off_rem + vw;
2926 if (TraceSuperWord && Verbose) {
2927 tty->print_cr("SWPointer::memory_alignment: off_rem = %d, off_mod = %d", off_rem, off_mod);
2928 }
2929 return off_mod;
2930 }
2931
2932 //---------------------------container_type---------------------------
2933 // Smallest type containing range of values
2934 const Type* SuperWord::container_type(Node* n) {
2935 if (n->is_Mem()) {
2936 BasicType bt = n->as_Mem()->memory_type();
2937 if (n->is_Store() && (bt == T_CHAR)) {
2938 // Use T_SHORT type instead of T_CHAR for stored values because any
2939 // preceding arithmetic operation extends values to signed Int.
2940 bt = T_SHORT;
2941 }
2942 if (n->Opcode() == Op_LoadUB) {
2943 // Adjust type for unsigned byte loads, it is important for right shifts.
2944 // T_BOOLEAN is used because there is no basic type representing type
2945 // TypeInt::UBYTE. Use of T_BOOLEAN for vectors is fine because only
2946 // size (one byte) and sign is important.
2947 bt = T_BOOLEAN;
2948 }
2949 return Type::get_const_basic_type(bt);
2950 }
2951 const Type* t = _igvn.type(n);
2952 if (t->basic_type() == T_INT) {
2953 // A narrow type of arithmetic operations will be determined by
2954 // propagating the type of memory operations.
2955 return TypeInt::INT;
2956 }
2957 return t;
2958 }
2959
2960 bool SuperWord::same_velt_type(Node* n1, Node* n2) {
2961 const Type* vt1 = velt_type(n1);
2962 const Type* vt2 = velt_type(n2);
2963 if (vt1->basic_type() == T_INT && vt2->basic_type() == T_INT) {
2964 // Compare vectors element sizes for integer types.
2965 return data_size(n1) == data_size(n2);
2966 }
2967 return vt1 == vt2;
2968 }
2969
2970 //------------------------------in_packset---------------------------
2971 // Are s1 and s2 in a pack pair and ordered as s1,s2?
2972 bool SuperWord::in_packset(Node* s1, Node* s2) {
2973 for (int i = 0; i < _packset.length(); i++) {
2974 Node_List* p = _packset.at(i);
2975 assert(p->size() == 2, "must be");
2976 if (p->at(0) == s1 && p->at(p->size()-1) == s2) {
2977 return true;
2978 }
2979 }
2980 return false;
2981 }
2982
2983 //------------------------------in_pack---------------------------
2984 // Is s in pack p?
2985 Node_List* SuperWord::in_pack(Node* s, Node_List* p) {
2986 for (uint i = 0; i < p->size(); i++) {
2987 if (p->at(i) == s) {
2988 return p;
2989 }
2990 }
2991 return NULL;
2992 }
2993
2994 //------------------------------remove_pack_at---------------------------
2995 // Remove the pack at position pos in the packset
2996 void SuperWord::remove_pack_at(int pos) {
2997 Node_List* p = _packset.at(pos);
2998 for (uint i = 0; i < p->size(); i++) {
2999 Node* s = p->at(i);
3000 set_my_pack(s, NULL);
3001 }
3002 _packset.remove_at(pos);
3003 }
3004
3005 void SuperWord::packset_sort(int n) {
3006 // simple bubble sort so that we capitalize with O(n) when its already sorted
3007 while (n != 0) {
3008 bool swapped = false;
3009 for (int i = 1; i < n; i++) {
3010 Node_List* q_low = _packset.at(i-1);
3011 Node_List* q_i = _packset.at(i);
3012
3013 // only swap when we find something to swap
3014 if (alignment(q_low->at(0)) > alignment(q_i->at(0))) {
3015 Node_List* t = q_i;
3016 *(_packset.adr_at(i)) = q_low;
3017 *(_packset.adr_at(i-1)) = q_i;
3018 swapped = true;
3019 }
3020 }
3021 if (swapped == false) break;
3022 n--;
3023 }
3024 }
3025
3026 //------------------------------executed_first---------------------------
3027 // Return the node executed first in pack p. Uses the RPO block list
3028 // to determine order.
3029 Node* SuperWord::executed_first(Node_List* p) {
3030 Node* n = p->at(0);
3031 int n_rpo = bb_idx(n);
3032 for (uint i = 1; i < p->size(); i++) {
3033 Node* s = p->at(i);
3034 int s_rpo = bb_idx(s);
3035 if (s_rpo < n_rpo) {
3036 n = s;
3037 n_rpo = s_rpo;
3038 }
3039 }
3040 return n;
3041 }
3042
3043 //------------------------------executed_last---------------------------
3044 // Return the node executed last in pack p.
3045 Node* SuperWord::executed_last(Node_List* p) {
3046 Node* n = p->at(0);
3047 int n_rpo = bb_idx(n);
3048 for (uint i = 1; i < p->size(); i++) {
3049 Node* s = p->at(i);
3050 int s_rpo = bb_idx(s);
3051 if (s_rpo > n_rpo) {
3052 n = s;
3053 n_rpo = s_rpo;
3054 }
3055 }
3056 return n;
3057 }
3058
3059 LoadNode::ControlDependency SuperWord::control_dependency(Node_List* p) {
3060 LoadNode::ControlDependency dep = LoadNode::DependsOnlyOnTest;
3061 for (uint i = 0; i < p->size(); i++) {
3062 Node* n = p->at(i);
3063 assert(n->is_Load(), "only meaningful for loads");
3064 if (!n->depends_only_on_test()) {
3065 dep = LoadNode::Pinned;
3066 }
3067 }
3068 return dep;
3069 }
3070
3071
3072 //----------------------------align_initial_loop_index---------------------------
3073 // Adjust pre-loop limit so that in main loop, a load/store reference
3074 // to align_to_ref will be a position zero in the vector.
3075 // (iv + k) mod vector_align == 0
3076 void SuperWord::align_initial_loop_index(MemNode* align_to_ref) {
3077 CountedLoopNode *main_head = lp()->as_CountedLoop();
3078 assert(main_head->is_main_loop(), "");
3079 CountedLoopEndNode* pre_end = get_pre_loop_end(main_head);
3080 assert(pre_end != NULL, "we must have a correct pre-loop");
3081 Node *pre_opaq1 = pre_end->limit();
3082 assert(pre_opaq1->Opcode() == Op_Opaque1, "");
3083 Opaque1Node *pre_opaq = (Opaque1Node*)pre_opaq1;
3084 Node *lim0 = pre_opaq->in(1);
3085
3086 // Where we put new limit calculations
3087 Node *pre_ctrl = pre_end->loopnode()->in(LoopNode::EntryControl);
3088
3089 // Ensure the original loop limit is available from the
3090 // pre-loop Opaque1 node.
3091 Node *orig_limit = pre_opaq->original_loop_limit();
3092 assert(orig_limit != NULL && _igvn.type(orig_limit) != Type::TOP, "");
3093
3094 SWPointer align_to_ref_p(align_to_ref, this, NULL, false);
3095 assert(align_to_ref_p.valid(), "sanity");
3096
3097 // Given:
3098 // lim0 == original pre loop limit
3099 // V == v_align (power of 2)
3100 // invar == extra invariant piece of the address expression
3101 // e == offset [ +/- invar ]
3102 //
3103 // When reassociating expressions involving '%' the basic rules are:
3104 // (a - b) % k == 0 => a % k == b % k
3105 // and:
3106 // (a + b) % k == 0 => a % k == (k - b) % k
3107 //
3108 // For stride > 0 && scale > 0,
3109 // Derive the new pre-loop limit "lim" such that the two constraints:
3110 // (1) lim = lim0 + N (where N is some positive integer < V)
3111 // (2) (e + lim) % V == 0
3112 // are true.
3113 //
3114 // Substituting (1) into (2),
3115 // (e + lim0 + N) % V == 0
3116 // solve for N:
3117 // N = (V - (e + lim0)) % V
3118 // substitute back into (1), so that new limit
3119 // lim = lim0 + (V - (e + lim0)) % V
3120 //
3121 // For stride > 0 && scale < 0
3122 // Constraints:
3123 // lim = lim0 + N
3124 // (e - lim) % V == 0
3125 // Solving for lim:
3126 // (e - lim0 - N) % V == 0
3127 // N = (e - lim0) % V
3128 // lim = lim0 + (e - lim0) % V
3129 //
3130 // For stride < 0 && scale > 0
3131 // Constraints:
3132 // lim = lim0 - N
3133 // (e + lim) % V == 0
3134 // Solving for lim:
3135 // (e + lim0 - N) % V == 0
3136 // N = (e + lim0) % V
3137 // lim = lim0 - (e + lim0) % V
3138 //
3139 // For stride < 0 && scale < 0
3140 // Constraints:
3141 // lim = lim0 - N
3142 // (e - lim) % V == 0
3143 // Solving for lim:
3144 // (e - lim0 + N) % V == 0
3145 // N = (V - (e - lim0)) % V
3146 // lim = lim0 - (V - (e - lim0)) % V
3147
3148 int vw = vector_width_in_bytes(align_to_ref);
3149 int stride = iv_stride();
3150 int scale = align_to_ref_p.scale_in_bytes();
3151 int elt_size = align_to_ref_p.memory_size();
3152 int v_align = vw / elt_size;
3153 assert(v_align > 1, "sanity");
3154 int offset = align_to_ref_p.offset_in_bytes() / elt_size;
3155 Node *offsn = _igvn.intcon(offset);
3156
3157 Node *e = offsn;
3158 if (align_to_ref_p.invar() != NULL) {
3159 // incorporate any extra invariant piece producing (offset +/- invar) >>> log2(elt)
3160 Node* log2_elt = _igvn.intcon(exact_log2(elt_size));
3161 Node* aref = new URShiftINode(align_to_ref_p.invar(), log2_elt);
3162 _igvn.register_new_node_with_optimizer(aref);
3163 _phase->set_ctrl(aref, pre_ctrl);
3164 if (align_to_ref_p.negate_invar()) {
3165 e = new SubINode(e, aref);
3166 } else {
3167 e = new AddINode(e, aref);
3168 }
3169 _igvn.register_new_node_with_optimizer(e);
3170 _phase->set_ctrl(e, pre_ctrl);
3171 }
3172 if (vw > ObjectAlignmentInBytes) {
3173 // incorporate base e +/- base && Mask >>> log2(elt)
3174 Node* xbase = new CastP2XNode(NULL, align_to_ref_p.base());
3175 _igvn.register_new_node_with_optimizer(xbase);
3176 #ifdef _LP64
3177 xbase = new ConvL2INode(xbase);
3178 _igvn.register_new_node_with_optimizer(xbase);
3179 #endif
3180 Node* mask = _igvn.intcon(vw-1);
3181 Node* masked_xbase = new AndINode(xbase, mask);
3182 _igvn.register_new_node_with_optimizer(masked_xbase);
3183 Node* log2_elt = _igvn.intcon(exact_log2(elt_size));
3184 Node* bref = new URShiftINode(masked_xbase, log2_elt);
3185 _igvn.register_new_node_with_optimizer(bref);
3186 _phase->set_ctrl(bref, pre_ctrl);
3187 e = new AddINode(e, bref);
3188 _igvn.register_new_node_with_optimizer(e);
3189 _phase->set_ctrl(e, pre_ctrl);
3190 }
3191
3192 // compute e +/- lim0
3193 if (scale < 0) {
3194 e = new SubINode(e, lim0);
3195 } else {
3196 e = new AddINode(e, lim0);
3197 }
3198 _igvn.register_new_node_with_optimizer(e);
3199 _phase->set_ctrl(e, pre_ctrl);
3200
3201 if (stride * scale > 0) {
3202 // compute V - (e +/- lim0)
3203 Node* va = _igvn.intcon(v_align);
3204 e = new SubINode(va, e);
3205 _igvn.register_new_node_with_optimizer(e);
3206 _phase->set_ctrl(e, pre_ctrl);
3207 }
3208 // compute N = (exp) % V
3209 Node* va_msk = _igvn.intcon(v_align - 1);
3210 Node* N = new AndINode(e, va_msk);
3211 _igvn.register_new_node_with_optimizer(N);
3212 _phase->set_ctrl(N, pre_ctrl);
3213
3214 // substitute back into (1), so that new limit
3215 // lim = lim0 + N
3216 Node* lim;
3217 if (stride < 0) {
3218 lim = new SubINode(lim0, N);
3219 } else {
3220 lim = new AddINode(lim0, N);
3221 }
3222 _igvn.register_new_node_with_optimizer(lim);
3223 _phase->set_ctrl(lim, pre_ctrl);
3224 Node* constrained =
3225 (stride > 0) ? (Node*) new MinINode(lim, orig_limit)
3226 : (Node*) new MaxINode(lim, orig_limit);
3227 _igvn.register_new_node_with_optimizer(constrained);
3228 _phase->set_ctrl(constrained, pre_ctrl);
3229 _igvn.hash_delete(pre_opaq);
3230 pre_opaq->set_req(1, constrained);
3231 }
3232
3233 //----------------------------get_pre_loop_end---------------------------
3234 // Find pre loop end from main loop. Returns null if none.
3235 CountedLoopEndNode* SuperWord::get_pre_loop_end(CountedLoopNode* cl) {
3236 // The loop cannot be optimized if the graph shape at
3237 // the loop entry is inappropriate.
3238 if (!PhaseIdealLoop::is_canonical_loop_entry(cl)) {
3239 return NULL;
3240 }
3241
3242 Node* p_f = cl->in(LoopNode::EntryControl)->in(0)->in(0);
3243 if (!p_f->is_IfFalse()) return NULL;
3244 if (!p_f->in(0)->is_CountedLoopEnd()) return NULL;
3245 CountedLoopEndNode* pre_end = p_f->in(0)->as_CountedLoopEnd();
3246 CountedLoopNode* loop_node = pre_end->loopnode();
3247 if (loop_node == NULL || !loop_node->is_pre_loop()) return NULL;
3248 return pre_end;
3249 }
3250
3251 //------------------------------init---------------------------
3252 void SuperWord::init() {
3253 _dg.init();
3254 _packset.clear();
3255 _disjoint_ptrs.clear();
3256 _block.clear();
3257 _post_block.clear();
3258 _data_entry.clear();
3259 _mem_slice_head.clear();
3260 _mem_slice_tail.clear();
3261 _iteration_first.clear();
3262 _iteration_last.clear();
3263 _node_info.clear();
3264 _align_to_ref = NULL;
3265 _lpt = NULL;
3266 _lp = NULL;
3267 _bb = NULL;
3268 _iv = NULL;
3269 _race_possible = 0;
3270 _early_return = false;
3271 _num_work_vecs = 0;
3272 _num_reductions = 0;
3273 }
3274
3275 //------------------------------restart---------------------------
3276 void SuperWord::restart() {
3277 _dg.init();
3278 _packset.clear();
3279 _disjoint_ptrs.clear();
3280 _block.clear();
3281 _post_block.clear();
3282 _data_entry.clear();
3283 _mem_slice_head.clear();
3284 _mem_slice_tail.clear();
3285 _node_info.clear();
3286 }
3287
3288 //------------------------------print_packset---------------------------
3289 void SuperWord::print_packset() {
3290 #ifndef PRODUCT
3291 tty->print_cr("packset");
3292 for (int i = 0; i < _packset.length(); i++) {
3293 tty->print_cr("Pack: %d", i);
3294 Node_List* p = _packset.at(i);
3295 print_pack(p);
3296 }
3297 #endif
3298 }
3299
3300 //------------------------------print_pack---------------------------
3301 void SuperWord::print_pack(Node_List* p) {
3302 for (uint i = 0; i < p->size(); i++) {
3303 print_stmt(p->at(i));
3304 }
3305 }
3306
3307 //------------------------------print_bb---------------------------
3308 void SuperWord::print_bb() {
3309 #ifndef PRODUCT
3310 tty->print_cr("\nBlock");
3311 for (int i = 0; i < _block.length(); i++) {
3312 Node* n = _block.at(i);
3313 tty->print("%d ", i);
3314 if (n) {
3315 n->dump();
3316 }
3317 }
3318 #endif
3319 }
3320
3321 //------------------------------print_stmt---------------------------
3322 void SuperWord::print_stmt(Node* s) {
3323 #ifndef PRODUCT
3324 tty->print(" align: %d \t", alignment(s));
3325 s->dump();
3326 #endif
3327 }
3328
3329 //------------------------------blank---------------------------
3330 char* SuperWord::blank(uint depth) {
3331 static char blanks[101];
3332 assert(depth < 101, "too deep");
3333 for (uint i = 0; i < depth; i++) blanks[i] = ' ';
3334 blanks[depth] = '\0';
3335 return blanks;
3336 }
3337
3338
3339 //==============================SWPointer===========================
3340 #ifndef PRODUCT
3341 int SWPointer::Tracer::_depth = 0;
3342 #endif
3343 //----------------------------SWPointer------------------------
3344 SWPointer::SWPointer(MemNode* mem, SuperWord* slp, Node_Stack *nstack, bool analyze_only) :
3345 _mem(mem), _slp(slp), _base(NULL), _adr(NULL),
3346 _scale(0), _offset(0), _invar(NULL), _negate_invar(false),
3347 _nstack(nstack), _analyze_only(analyze_only),
3348 _stack_idx(0)
3349 #ifndef PRODUCT
3350 , _tracer(slp)
3351 #endif
3352 {
3353 NOT_PRODUCT(_tracer.ctor_1(mem);)
3354
3355 Node* adr = mem->in(MemNode::Address);
3356 if (!adr->is_AddP()) {
3357 assert(!valid(), "too complex");
3358 return;
3359 }
3360 // Match AddP(base, AddP(ptr, k*iv [+ invariant]), constant)
3361 Node* base = adr->in(AddPNode::Base);
3362 // The base address should be loop invariant
3363 if (!invariant(base)) {
3364 assert(!valid(), "base address is loop variant");
3365 return;
3366 }
3367 //unsafe reference could not be aligned appropriately without runtime checking
3368 if (base == NULL || base->bottom_type() == Type::TOP) {
3369 assert(!valid(), "unsafe access");
3370 return;
3371 }
3372
3373 NOT_PRODUCT(if(_slp->is_trace_alignment()) _tracer.store_depth();)
3374 NOT_PRODUCT(_tracer.ctor_2(adr);)
3375
3376 int i;
3377 for (i = 0; i < 3; i++) {
3378 NOT_PRODUCT(_tracer.ctor_3(adr, i);)
3379
3380 if (!scaled_iv_plus_offset(adr->in(AddPNode::Offset))) {
3381 assert(!valid(), "too complex");
3382 return;
3383 }
3384 adr = adr->in(AddPNode::Address);
3385 NOT_PRODUCT(_tracer.ctor_4(adr, i);)
3386
3387 if (base == adr || !adr->is_AddP()) {
3388 NOT_PRODUCT(_tracer.ctor_5(adr, base, i);)
3389 break; // stop looking at addp's
3390 }
3391 }
3392 NOT_PRODUCT(if(_slp->is_trace_alignment()) _tracer.restore_depth();)
3393 NOT_PRODUCT(_tracer.ctor_6(mem);)
3394
3395 _base = base;
3396 _adr = adr;
3397 assert(valid(), "Usable");
3398 }
3399
3400 // Following is used to create a temporary object during
3401 // the pattern match of an address expression.
3402 SWPointer::SWPointer(SWPointer* p) :
3403 _mem(p->_mem), _slp(p->_slp), _base(NULL), _adr(NULL),
3404 _scale(0), _offset(0), _invar(NULL), _negate_invar(false),
3405 _nstack(p->_nstack), _analyze_only(p->_analyze_only),
3406 _stack_idx(p->_stack_idx)
3407 #ifndef PRODUCT
3408 , _tracer(p->_slp)
3409 #endif
3410 {}
3411
3412
3413 bool SWPointer::invariant(Node* n) {
3414 NOT_PRODUCT(Tracer::Depth dd;)
3415 Node *n_c = phase()->get_ctrl(n);
3416 NOT_PRODUCT(_tracer.invariant_1(n, n_c);)
3417 return !lpt()->is_member(phase()->get_loop(n_c));
3418 }
3419 //------------------------scaled_iv_plus_offset--------------------
3420 // Match: k*iv + offset
3421 // where: k is a constant that maybe zero, and
3422 // offset is (k2 [+/- invariant]) where k2 maybe zero and invariant is optional
3423 bool SWPointer::scaled_iv_plus_offset(Node* n) {
3424 NOT_PRODUCT(Tracer::Depth ddd;)
3425 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_1(n);)
3426
3427 if (scaled_iv(n)) {
3428 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_2(n);)
3429 return true;
3430 }
3431
3432 if (offset_plus_k(n)) {
3433 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_3(n);)
3434 return true;
3435 }
3436
3437 int opc = n->Opcode();
3438 if (opc == Op_AddI) {
3439 if (scaled_iv(n->in(1)) && offset_plus_k(n->in(2))) {
3440 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_4(n);)
3441 return true;
3442 }
3443 if (scaled_iv(n->in(2)) && offset_plus_k(n->in(1))) {
3444 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_5(n);)
3445 return true;
3446 }
3447 } else if (opc == Op_SubI) {
3448 if (scaled_iv(n->in(1)) && offset_plus_k(n->in(2), true)) {
3449 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_6(n);)
3450 return true;
3451 }
3452 if (scaled_iv(n->in(2)) && offset_plus_k(n->in(1))) {
3453 _scale *= -1;
3454 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_7(n);)
3455 return true;
3456 }
3457 }
3458
3459 NOT_PRODUCT(_tracer.scaled_iv_plus_offset_8(n);)
3460 return false;
3461 }
3462
3463 //----------------------------scaled_iv------------------------
3464 // Match: k*iv where k is a constant that's not zero
3465 bool SWPointer::scaled_iv(Node* n) {
3466 NOT_PRODUCT(Tracer::Depth ddd;)
3467 NOT_PRODUCT(_tracer.scaled_iv_1(n);)
3468
3469 if (_scale != 0) { // already found a scale
3470 NOT_PRODUCT(_tracer.scaled_iv_2(n, _scale);)
3471 return false;
3472 }
3473
3474 if (n == iv()) {
3475 _scale = 1;
3476 NOT_PRODUCT(_tracer.scaled_iv_3(n, _scale);)
3477 return true;
3478 }
3479 if (_analyze_only && (invariant(n) == false)) {
3480 _nstack->push(n, _stack_idx++);
3481 }
3482
3483 int opc = n->Opcode();
3484 if (opc == Op_MulI) {
3485 if (n->in(1) == iv() && n->in(2)->is_Con()) {
3486 _scale = n->in(2)->get_int();
3487 NOT_PRODUCT(_tracer.scaled_iv_4(n, _scale);)
3488 return true;
3489 } else if (n->in(2) == iv() && n->in(1)->is_Con()) {
3490 _scale = n->in(1)->get_int();
3491 NOT_PRODUCT(_tracer.scaled_iv_5(n, _scale);)
3492 return true;
3493 }
3494 } else if (opc == Op_LShiftI) {
3495 if (n->in(1) == iv() && n->in(2)->is_Con()) {
3496 _scale = 1 << n->in(2)->get_int();
3497 NOT_PRODUCT(_tracer.scaled_iv_6(n, _scale);)
3498 return true;
3499 }
3500 } else if (opc == Op_ConvI2L) {
3501 if (n->in(1)->Opcode() == Op_CastII &&
3502 n->in(1)->as_CastII()->has_range_check()) {
3503 // Skip range check dependent CastII nodes
3504 n = n->in(1);
3505 }
3506 if (scaled_iv_plus_offset(n->in(1))) {
3507 NOT_PRODUCT(_tracer.scaled_iv_7(n);)
3508 return true;
3509 }
3510 } else if (opc == Op_LShiftL) {
3511 if (!has_iv() && _invar == NULL) {
3512 // Need to preserve the current _offset value, so
3513 // create a temporary object for this expression subtree.
3514 // Hacky, so should re-engineer the address pattern match.
3515 NOT_PRODUCT(Tracer::Depth dddd;)
3516 SWPointer tmp(this);
3517 NOT_PRODUCT(_tracer.scaled_iv_8(n, &tmp);)
3518
3519 if (tmp.scaled_iv_plus_offset(n->in(1))) {
3520 if (tmp._invar == NULL || _slp->do_vector_loop()) {
3521 int mult = 1 << n->in(2)->get_int();
3522 _scale = tmp._scale * mult;
3523 _offset += tmp._offset * mult;
3524 NOT_PRODUCT(_tracer.scaled_iv_9(n, _scale, _offset, mult);)
3525 return true;
3526 }
3527 }
3528 }
3529 }
3530 NOT_PRODUCT(_tracer.scaled_iv_10(n);)
3531 return false;
3532 }
3533
3534 //----------------------------offset_plus_k------------------------
3535 // Match: offset is (k [+/- invariant])
3536 // where k maybe zero and invariant is optional, but not both.
3537 bool SWPointer::offset_plus_k(Node* n, bool negate) {
3538 NOT_PRODUCT(Tracer::Depth ddd;)
3539 NOT_PRODUCT(_tracer.offset_plus_k_1(n);)
3540
3541 int opc = n->Opcode();
3542 if (opc == Op_ConI) {
3543 _offset += negate ? -(n->get_int()) : n->get_int();
3544 NOT_PRODUCT(_tracer.offset_plus_k_2(n, _offset);)
3545 return true;
3546 } else if (opc == Op_ConL) {
3547 // Okay if value fits into an int
3548 const TypeLong* t = n->find_long_type();
3549 if (t->higher_equal(TypeLong::INT)) {
3550 jlong loff = n->get_long();
3551 jint off = (jint)loff;
3552 _offset += negate ? -off : loff;
3553 NOT_PRODUCT(_tracer.offset_plus_k_3(n, _offset);)
3554 return true;
3555 }
3556 NOT_PRODUCT(_tracer.offset_plus_k_4(n);)
3557 return false;
3558 }
3559 if (_invar != NULL) { // already has an invariant
3560 NOT_PRODUCT(_tracer.offset_plus_k_5(n, _invar);)
3561 return false;
3562 }
3563
3564 if (_analyze_only && (invariant(n) == false)) {
3565 _nstack->push(n, _stack_idx++);
3566 }
3567 if (opc == Op_AddI) {
3568 if (n->in(2)->is_Con() && invariant(n->in(1))) {
3569 _negate_invar = negate;
3570 _invar = n->in(1);
3571 _offset += negate ? -(n->in(2)->get_int()) : n->in(2)->get_int();
3572 NOT_PRODUCT(_tracer.offset_plus_k_6(n, _invar, _negate_invar, _offset);)
3573 return true;
3574 } else if (n->in(1)->is_Con() && invariant(n->in(2))) {
3575 _offset += negate ? -(n->in(1)->get_int()) : n->in(1)->get_int();
3576 _negate_invar = negate;
3577 _invar = n->in(2);
3578 NOT_PRODUCT(_tracer.offset_plus_k_7(n, _invar, _negate_invar, _offset);)
3579 return true;
3580 }
3581 }
3582 if (opc == Op_SubI) {
3583 if (n->in(2)->is_Con() && invariant(n->in(1))) {
3584 _negate_invar = negate;
3585 _invar = n->in(1);
3586 _offset += !negate ? -(n->in(2)->get_int()) : n->in(2)->get_int();
3587 NOT_PRODUCT(_tracer.offset_plus_k_8(n, _invar, _negate_invar, _offset);)
3588 return true;
3589 } else if (n->in(1)->is_Con() && invariant(n->in(2))) {
3590 _offset += negate ? -(n->in(1)->get_int()) : n->in(1)->get_int();
3591 _negate_invar = !negate;
3592 _invar = n->in(2);
3593 NOT_PRODUCT(_tracer.offset_plus_k_9(n, _invar, _negate_invar, _offset);)
3594 return true;
3595 }
3596 }
3597 if (invariant(n)) {
3598 if (opc == Op_ConvI2L) {
3599 n = n->in(1);
3600 if (n->Opcode() == Op_CastII &&
3601 n->as_CastII()->has_range_check()) {
3602 // Skip range check dependent CastII nodes
3603 assert(invariant(n), "sanity");
3604 n = n->in(1);
3605 }
3606 }
3607 if (n->bottom_type()->isa_int()) {
3608 _negate_invar = negate;
3609 _invar = n;
3610 NOT_PRODUCT(_tracer.offset_plus_k_10(n, _invar, _negate_invar, _offset);)
3611 return true;
3612 }
3613 }
3614
3615 NOT_PRODUCT(_tracer.offset_plus_k_11(n);)
3616 return false;
3617 }
3618
3619 //----------------------------print------------------------
3620 void SWPointer::print() {
3621 #ifndef PRODUCT
3622 tty->print("base: %d adr: %d scale: %d offset: %d invar: %c%d\n",
3623 _base != NULL ? _base->_idx : 0,
3624 _adr != NULL ? _adr->_idx : 0,
3625 _scale, _offset,
3626 _negate_invar?'-':'+',
3627 _invar != NULL ? _invar->_idx : 0);
3628 #endif
3629 }
3630
3631 //----------------------------tracing------------------------
3632 #ifndef PRODUCT
3633 void SWPointer::Tracer::print_depth() {
3634 for (int ii = 0; ii<_depth; ++ii) tty->print(" ");
3635 }
3636
3637 void SWPointer::Tracer::ctor_1 (Node* mem) {
3638 if(_slp->is_trace_alignment()) {
3639 print_depth(); tty->print(" %d SWPointer::SWPointer: start alignment analysis", mem->_idx); mem->dump();
3640 }
3641 }
3642
3643 void SWPointer::Tracer::ctor_2(Node* adr) {
3644 if(_slp->is_trace_alignment()) {
3645 //store_depth();
3646 inc_depth();
3647 print_depth(); tty->print(" %d (adr) SWPointer::SWPointer: ", adr->_idx); adr->dump();
3648 inc_depth();
3649 print_depth(); tty->print(" %d (base) SWPointer::SWPointer: ", adr->in(AddPNode::Base)->_idx); adr->in(AddPNode::Base)->dump();
3650 }
3651 }
3652
3653 void SWPointer::Tracer::ctor_3(Node* adr, int i) {
3654 if(_slp->is_trace_alignment()) {
3655 inc_depth();
3656 Node* offset = adr->in(AddPNode::Offset);
3657 print_depth(); tty->print(" %d (offset) SWPointer::SWPointer: i = %d: ", offset->_idx, i); offset->dump();
3658 }
3659 }
3660
3661 void SWPointer::Tracer::ctor_4(Node* adr, int i) {
3662 if(_slp->is_trace_alignment()) {
3663 inc_depth();
3664 print_depth(); tty->print(" %d (adr) SWPointer::SWPointer: i = %d: ", adr->_idx, i); adr->dump();
3665 }
3666 }
3667
3668 void SWPointer::Tracer::ctor_5(Node* adr, Node* base, int i) {
3669 if(_slp->is_trace_alignment()) {
3670 inc_depth();
3671 if (base == adr) {
3672 print_depth(); tty->print_cr(" \\ %d (adr) == %d (base) SWPointer::SWPointer: breaking analysis at i = %d", adr->_idx, base->_idx, i);
3673 } else if (!adr->is_AddP()) {
3674 print_depth(); tty->print_cr(" \\ %d (adr) is NOT Addp SWPointer::SWPointer: breaking analysis at i = %d", adr->_idx, i);
3675 }
3676 }
3677 }
3678
3679 void SWPointer::Tracer::ctor_6(Node* mem) {
3680 if(_slp->is_trace_alignment()) {
3681 //restore_depth();
3682 print_depth(); tty->print_cr(" %d (adr) SWPointer::SWPointer: stop analysis", mem->_idx);
3683 }
3684 }
3685
3686 void SWPointer::Tracer::invariant_1(Node *n, Node *n_c) {
3687 if (_slp->do_vector_loop() && _slp->is_debug() && _slp->_lpt->is_member(_slp->_phase->get_loop(n_c)) != (int)_slp->in_bb(n)) {
3688 int is_member = _slp->_lpt->is_member(_slp->_phase->get_loop(n_c));
3689 int in_bb = _slp->in_bb(n);
3690 print_depth(); tty->print(" \\ "); tty->print_cr(" %d SWPointer::invariant conditions differ: n_c %d", n->_idx, n_c->_idx);
3691 print_depth(); tty->print(" \\ "); tty->print_cr("is_member %d, in_bb %d", is_member, in_bb);
3692 print_depth(); tty->print(" \\ "); n->dump();
3693 print_depth(); tty->print(" \\ "); n_c->dump();
3694 }
3695 }
3696
3697 void SWPointer::Tracer::scaled_iv_plus_offset_1(Node* n) {
3698 if(_slp->is_trace_alignment()) {
3699 print_depth(); tty->print(" %d SWPointer::scaled_iv_plus_offset testing node: ", n->_idx);
3700 n->dump();
3701 }
3702 }
3703
3704 void SWPointer::Tracer::scaled_iv_plus_offset_2(Node* n) {
3705 if(_slp->is_trace_alignment()) {
3706 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv_plus_offset: PASSED", n->_idx);
3707 }
3708 }
3709
3710 void SWPointer::Tracer::scaled_iv_plus_offset_3(Node* n) {
3711 if(_slp->is_trace_alignment()) {
3712 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv_plus_offset: PASSED", n->_idx);
3713 }
3714 }
3715
3716 void SWPointer::Tracer::scaled_iv_plus_offset_4(Node* n) {
3717 if(_slp->is_trace_alignment()) {
3718 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv_plus_offset: Op_AddI PASSED", n->_idx);
3719 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(1) is scaled_iv: ", n->in(1)->_idx); n->in(1)->dump();
3720 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(2) is offset_plus_k: ", n->in(2)->_idx); n->in(2)->dump();
3721 }
3722 }
3723
3724 void SWPointer::Tracer::scaled_iv_plus_offset_5(Node* n) {
3725 if(_slp->is_trace_alignment()) {
3726 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv_plus_offset: Op_AddI PASSED", n->_idx);
3727 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(2) is scaled_iv: ", n->in(2)->_idx); n->in(2)->dump();
3728 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(1) is offset_plus_k: ", n->in(1)->_idx); n->in(1)->dump();
3729 }
3730 }
3731
3732 void SWPointer::Tracer::scaled_iv_plus_offset_6(Node* n) {
3733 if(_slp->is_trace_alignment()) {
3734 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv_plus_offset: Op_SubI PASSED", n->_idx);
3735 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(1) is scaled_iv: ", n->in(1)->_idx); n->in(1)->dump();
3736 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(2) is offset_plus_k: ", n->in(2)->_idx); n->in(2)->dump();
3737 }
3738 }
3739
3740 void SWPointer::Tracer::scaled_iv_plus_offset_7(Node* n) {
3741 if(_slp->is_trace_alignment()) {
3742 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv_plus_offset: Op_SubI PASSED", n->_idx);
3743 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(2) is scaled_iv: ", n->in(2)->_idx); n->in(2)->dump();
3744 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv_plus_offset: in(1) is offset_plus_k: ", n->in(1)->_idx); n->in(1)->dump();
3745 }
3746 }
3747
3748 void SWPointer::Tracer::scaled_iv_plus_offset_8(Node* n) {
3749 if(_slp->is_trace_alignment()) {
3750 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv_plus_offset: FAILED", n->_idx);
3751 }
3752 }
3753
3754 void SWPointer::Tracer::scaled_iv_1(Node* n) {
3755 if(_slp->is_trace_alignment()) {
3756 print_depth(); tty->print(" %d SWPointer::scaled_iv: testing node: ", n->_idx); n->dump();
3757 }
3758 }
3759
3760 void SWPointer::Tracer::scaled_iv_2(Node* n, int scale) {
3761 if(_slp->is_trace_alignment()) {
3762 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: FAILED since another _scale has been detected before", n->_idx);
3763 print_depth(); tty->print_cr(" \\ SWPointer::scaled_iv: _scale (%d) != 0", scale);
3764 }
3765 }
3766
3767 void SWPointer::Tracer::scaled_iv_3(Node* n, int scale) {
3768 if(_slp->is_trace_alignment()) {
3769 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: is iv, setting _scale = %d", n->_idx, scale);
3770 }
3771 }
3772
3773 void SWPointer::Tracer::scaled_iv_4(Node* n, int scale) {
3774 if(_slp->is_trace_alignment()) {
3775 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: Op_MulI PASSED, setting _scale = %d", n->_idx, scale);
3776 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv: in(1) is iv: ", n->in(1)->_idx); n->in(1)->dump();
3777 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv: in(2) is Con: ", n->in(2)->_idx); n->in(2)->dump();
3778 }
3779 }
3780
3781 void SWPointer::Tracer::scaled_iv_5(Node* n, int scale) {
3782 if(_slp->is_trace_alignment()) {
3783 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: Op_MulI PASSED, setting _scale = %d", n->_idx, scale);
3784 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv: in(2) is iv: ", n->in(2)->_idx); n->in(2)->dump();
3785 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv: in(1) is Con: ", n->in(1)->_idx); n->in(1)->dump();
3786 }
3787 }
3788
3789 void SWPointer::Tracer::scaled_iv_6(Node* n, int scale) {
3790 if(_slp->is_trace_alignment()) {
3791 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: Op_LShiftI PASSED, setting _scale = %d", n->_idx, scale);
3792 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv: in(1) is iv: ", n->in(1)->_idx); n->in(1)->dump();
3793 print_depth(); tty->print(" \\ %d SWPointer::scaled_iv: in(2) is Con: ", n->in(2)->_idx); n->in(2)->dump();
3794 }
3795 }
3796
3797 void SWPointer::Tracer::scaled_iv_7(Node* n) {
3798 if(_slp->is_trace_alignment()) {
3799 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: Op_ConvI2L PASSED", n->_idx);
3800 print_depth(); tty->print_cr(" \\ SWPointer::scaled_iv: in(1) %d is scaled_iv_plus_offset: ", n->in(1)->_idx);
3801 inc_depth(); inc_depth();
3802 print_depth(); n->in(1)->dump();
3803 dec_depth(); dec_depth();
3804 }
3805 }
3806
3807 void SWPointer::Tracer::scaled_iv_8(Node* n, SWPointer* tmp) {
3808 if(_slp->is_trace_alignment()) {
3809 print_depth(); tty->print(" %d SWPointer::scaled_iv: Op_LShiftL, creating tmp SWPointer: ", n->_idx); tmp->print();
3810 }
3811 }
3812
3813 void SWPointer::Tracer::scaled_iv_9(Node* n, int scale, int _offset, int mult) {
3814 if(_slp->is_trace_alignment()) {
3815 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: Op_LShiftL PASSED, setting _scale = %d, _offset = %d", n->_idx, scale, _offset);
3816 print_depth(); tty->print_cr(" \\ SWPointer::scaled_iv: in(1) %d is scaled_iv_plus_offset, in(2) %d used to get mult = %d: _scale = %d, _offset = %d",
3817 n->in(1)->_idx, n->in(2)->_idx, mult, scale, _offset);
3818 inc_depth(); inc_depth();
3819 print_depth(); n->in(1)->dump();
3820 print_depth(); n->in(2)->dump();
3821 dec_depth(); dec_depth();
3822 }
3823 }
3824
3825 void SWPointer::Tracer::scaled_iv_10(Node* n) {
3826 if(_slp->is_trace_alignment()) {
3827 print_depth(); tty->print_cr(" %d SWPointer::scaled_iv: FAILED", n->_idx);
3828 }
3829 }
3830
3831 void SWPointer::Tracer::offset_plus_k_1(Node* n) {
3832 if(_slp->is_trace_alignment()) {
3833 print_depth(); tty->print(" %d SWPointer::offset_plus_k: testing node: ", n->_idx); n->dump();
3834 }
3835 }
3836
3837 void SWPointer::Tracer::offset_plus_k_2(Node* n, int _offset) {
3838 if(_slp->is_trace_alignment()) {
3839 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: Op_ConI PASSED, setting _offset = %d", n->_idx, _offset);
3840 }
3841 }
3842
3843 void SWPointer::Tracer::offset_plus_k_3(Node* n, int _offset) {
3844 if(_slp->is_trace_alignment()) {
3845 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: Op_ConL PASSED, setting _offset = %d", n->_idx, _offset);
3846 }
3847 }
3848
3849 void SWPointer::Tracer::offset_plus_k_4(Node* n) {
3850 if(_slp->is_trace_alignment()) {
3851 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: FAILED", n->_idx);
3852 print_depth(); tty->print_cr(" \\ " JLONG_FORMAT " SWPointer::offset_plus_k: Op_ConL FAILED, k is too big", n->get_long());
3853 }
3854 }
3855
3856 void SWPointer::Tracer::offset_plus_k_5(Node* n, Node* _invar) {
3857 if(_slp->is_trace_alignment()) {
3858 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: FAILED since another invariant has been detected before", n->_idx);
3859 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: _invar != NULL: ", _invar->_idx); _invar->dump();
3860 }
3861 }
3862
3863 void SWPointer::Tracer::offset_plus_k_6(Node* n, Node* _invar, bool _negate_invar, int _offset) {
3864 if(_slp->is_trace_alignment()) {
3865 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: Op_AddI PASSED, setting _negate_invar = %d, _invar = %d, _offset = %d",
3866 n->_idx, _negate_invar, _invar->_idx, _offset);
3867 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(2) is Con: ", n->in(2)->_idx); n->in(2)->dump();
3868 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(1) is invariant: ", _invar->_idx); _invar->dump();
3869 }
3870 }
3871
3872 void SWPointer::Tracer::offset_plus_k_7(Node* n, Node* _invar, bool _negate_invar, int _offset) {
3873 if(_slp->is_trace_alignment()) {
3874 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: Op_AddI PASSED, setting _negate_invar = %d, _invar = %d, _offset = %d",
3875 n->_idx, _negate_invar, _invar->_idx, _offset);
3876 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(1) is Con: ", n->in(1)->_idx); n->in(1)->dump();
3877 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(2) is invariant: ", _invar->_idx); _invar->dump();
3878 }
3879 }
3880
3881 void SWPointer::Tracer::offset_plus_k_8(Node* n, Node* _invar, bool _negate_invar, int _offset) {
3882 if(_slp->is_trace_alignment()) {
3883 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: Op_SubI is PASSED, setting _negate_invar = %d, _invar = %d, _offset = %d",
3884 n->_idx, _negate_invar, _invar->_idx, _offset);
3885 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(2) is Con: ", n->in(2)->_idx); n->in(2)->dump();
3886 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(1) is invariant: ", _invar->_idx); _invar->dump();
3887 }
3888 }
3889
3890 void SWPointer::Tracer::offset_plus_k_9(Node* n, Node* _invar, bool _negate_invar, int _offset) {
3891 if(_slp->is_trace_alignment()) {
3892 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: Op_SubI PASSED, setting _negate_invar = %d, _invar = %d, _offset = %d", n->_idx, _negate_invar, _invar->_idx, _offset);
3893 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(1) is Con: ", n->in(1)->_idx); n->in(1)->dump();
3894 print_depth(); tty->print(" \\ %d SWPointer::offset_plus_k: in(2) is invariant: ", _invar->_idx); _invar->dump();
3895 }
3896 }
3897
3898 void SWPointer::Tracer::offset_plus_k_10(Node* n, Node* _invar, bool _negate_invar, int _offset) {
3899 if(_slp->is_trace_alignment()) {
3900 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: PASSED, setting _negate_invar = %d, _invar = %d, _offset = %d", n->_idx, _negate_invar, _invar->_idx, _offset);
3901 print_depth(); tty->print_cr(" \\ %d SWPointer::offset_plus_k: is invariant", n->_idx);
3902 }
3903 }
3904
3905 void SWPointer::Tracer::offset_plus_k_11(Node* n) {
3906 if(_slp->is_trace_alignment()) {
3907 print_depth(); tty->print_cr(" %d SWPointer::offset_plus_k: FAILED", n->_idx);
3908 }
3909 }
3910
3911 #endif
3912 // ========================= OrderedPair =====================
3913
3914 const OrderedPair OrderedPair::initial;
3915
3916 // ========================= SWNodeInfo =====================
3917
3918 const SWNodeInfo SWNodeInfo::initial;
3919
3920
3921 // ============================ DepGraph ===========================
3922
3923 //------------------------------make_node---------------------------
3924 // Make a new dependence graph node for an ideal node.
3925 DepMem* DepGraph::make_node(Node* node) {
3926 DepMem* m = new (_arena) DepMem(node);
3927 if (node != NULL) {
3928 assert(_map.at_grow(node->_idx) == NULL, "one init only");
3929 _map.at_put_grow(node->_idx, m);
3930 }
3931 return m;
3932 }
3933
3934 //------------------------------make_edge---------------------------
3935 // Make a new dependence graph edge from dpred -> dsucc
3936 DepEdge* DepGraph::make_edge(DepMem* dpred, DepMem* dsucc) {
3937 DepEdge* e = new (_arena) DepEdge(dpred, dsucc, dsucc->in_head(), dpred->out_head());
3938 dpred->set_out_head(e);
3939 dsucc->set_in_head(e);
3940 return e;
3941 }
3942
3943 // ========================== DepMem ========================
3944
3945 //------------------------------in_cnt---------------------------
3946 int DepMem::in_cnt() {
3947 int ct = 0;
3948 for (DepEdge* e = _in_head; e != NULL; e = e->next_in()) ct++;
3949 return ct;
3950 }
3951
3952 //------------------------------out_cnt---------------------------
3953 int DepMem::out_cnt() {
3954 int ct = 0;
3955 for (DepEdge* e = _out_head; e != NULL; e = e->next_out()) ct++;
3956 return ct;
3957 }
3958
3959 //------------------------------print-----------------------------
3960 void DepMem::print() {
3961 #ifndef PRODUCT
3962 tty->print(" DepNode %d (", _node->_idx);
3963 for (DepEdge* p = _in_head; p != NULL; p = p->next_in()) {
3964 Node* pred = p->pred()->node();
3965 tty->print(" %d", pred != NULL ? pred->_idx : 0);
3966 }
3967 tty->print(") [");
3968 for (DepEdge* s = _out_head; s != NULL; s = s->next_out()) {
3969 Node* succ = s->succ()->node();
3970 tty->print(" %d", succ != NULL ? succ->_idx : 0);
3971 }
3972 tty->print_cr(" ]");
3973 #endif
3974 }
3975
3976 // =========================== DepEdge =========================
3977
3978 //------------------------------DepPreds---------------------------
3979 void DepEdge::print() {
3980 #ifndef PRODUCT
3981 tty->print_cr("DepEdge: %d [ %d ]", _pred->node()->_idx, _succ->node()->_idx);
3982 #endif
3983 }
3984
3985 // =========================== DepPreds =========================
3986 // Iterator over predecessor edges in the dependence graph.
3987
3988 //------------------------------DepPreds---------------------------
3989 DepPreds::DepPreds(Node* n, DepGraph& dg) {
3990 _n = n;
3991 _done = false;
3992 if (_n->is_Store() || _n->is_Load()) {
3993 _next_idx = MemNode::Address;
3994 _end_idx = n->req();
3995 _dep_next = dg.dep(_n)->in_head();
3996 } else if (_n->is_Mem()) {
3997 _next_idx = 0;
3998 _end_idx = 0;
3999 _dep_next = dg.dep(_n)->in_head();
4000 } else {
4001 _next_idx = 1;
4002 _end_idx = _n->req();
4003 _dep_next = NULL;
4004 }
4005 next();
4006 }
4007
4008 //------------------------------next---------------------------
4009 void DepPreds::next() {
4010 if (_dep_next != NULL) {
4011 _current = _dep_next->pred()->node();
4012 _dep_next = _dep_next->next_in();
4013 } else if (_next_idx < _end_idx) {
4014 _current = _n->in(_next_idx++);
4015 } else {
4016 _done = true;
4017 }
4018 }
4019
4020 // =========================== DepSuccs =========================
4021 // Iterator over successor edges in the dependence graph.
4022
4023 //------------------------------DepSuccs---------------------------
4024 DepSuccs::DepSuccs(Node* n, DepGraph& dg) {
4025 _n = n;
4026 _done = false;
4027 if (_n->is_Load()) {
4028 _next_idx = 0;
4029 _end_idx = _n->outcnt();
4030 _dep_next = dg.dep(_n)->out_head();
4031 } else if (_n->is_Mem() || _n->is_Phi() && _n->bottom_type() == Type::MEMORY) {
4032 _next_idx = 0;
4033 _end_idx = 0;
4034 _dep_next = dg.dep(_n)->out_head();
4035 } else {
4036 _next_idx = 0;
4037 _end_idx = _n->outcnt();
4038 _dep_next = NULL;
4039 }
4040 next();
4041 }
4042
4043 //-------------------------------next---------------------------
4044 void DepSuccs::next() {
4045 if (_dep_next != NULL) {
4046 _current = _dep_next->succ()->node();
4047 _dep_next = _dep_next->next_out();
4048 } else if (_next_idx < _end_idx) {
4049 _current = _n->raw_out(_next_idx++);
4050 } else {
4051 _done = true;
4052 }
4053 }
4054
4055 //
4056 // --------------------------------- vectorization/simd -----------------------------------
4057 //
4058 bool SuperWord::same_origin_idx(Node* a, Node* b) const {
4059 return a != NULL && b != NULL && _clone_map.same_idx(a->_idx, b->_idx);
4060 }
4061 bool SuperWord::same_generation(Node* a, Node* b) const {
4062 return a != NULL && b != NULL && _clone_map.same_gen(a->_idx, b->_idx);
4063 }
4064
4065 Node* SuperWord::find_phi_for_mem_dep(LoadNode* ld) {
4066 assert(in_bb(ld), "must be in block");
4067 if (_clone_map.gen(ld->_idx) == _ii_first) {
4068 #ifndef PRODUCT
4069 if (_vector_loop_debug) {
4070 tty->print_cr("SuperWord::find_phi_for_mem_dep _clone_map.gen(ld->_idx)=%d",
4071 _clone_map.gen(ld->_idx));
4072 }
4073 #endif
4074 return NULL; //we think that any ld in the first gen being vectorizable
4075 }
4076
4077 Node* mem = ld->in(MemNode::Memory);
4078 if (mem->outcnt() <= 1) {
4079 // we don't want to remove the only edge from mem node to load
4080 #ifndef PRODUCT
4081 if (_vector_loop_debug) {
4082 tty->print_cr("SuperWord::find_phi_for_mem_dep input node %d to load %d has no other outputs and edge mem->load cannot be removed",
4083 mem->_idx, ld->_idx);
4084 ld->dump();
4085 mem->dump();
4086 }
4087 #endif
4088 return NULL;
4089 }
4090 if (!in_bb(mem) || same_generation(mem, ld)) {
4091 #ifndef PRODUCT
4092 if (_vector_loop_debug) {
4093 tty->print_cr("SuperWord::find_phi_for_mem_dep _clone_map.gen(mem->_idx)=%d",
4094 _clone_map.gen(mem->_idx));
4095 }
4096 #endif
4097 return NULL; // does not depend on loop volatile node or depends on the same generation
4098 }
4099
4100 //otherwise first node should depend on mem-phi
4101 Node* first = first_node(ld);
4102 assert(first->is_Load(), "must be Load");
4103 Node* phi = first->as_Load()->in(MemNode::Memory);
4104 if (!phi->is_Phi() || phi->bottom_type() != Type::MEMORY) {
4105 #ifndef PRODUCT
4106 if (_vector_loop_debug) {
4107 tty->print_cr("SuperWord::find_phi_for_mem_dep load is not vectorizable node, since it's `first` does not take input from mem phi");
4108 ld->dump();
4109 first->dump();
4110 }
4111 #endif
4112 return NULL;
4113 }
4114
4115 Node* tail = 0;
4116 for (int m = 0; m < _mem_slice_head.length(); m++) {
4117 if (_mem_slice_head.at(m) == phi) {
4118 tail = _mem_slice_tail.at(m);
4119 }
4120 }
4121 if (tail == 0) { //test that found phi is in the list _mem_slice_head
4122 #ifndef PRODUCT
4123 if (_vector_loop_debug) {
4124 tty->print_cr("SuperWord::find_phi_for_mem_dep load %d is not vectorizable node, its phi %d is not _mem_slice_head",
4125 ld->_idx, phi->_idx);
4126 ld->dump();
4127 phi->dump();
4128 }
4129 #endif
4130 return NULL;
4131 }
4132
4133 // now all conditions are met
4134 return phi;
4135 }
4136
4137 Node* SuperWord::first_node(Node* nd) {
4138 for (int ii = 0; ii < _iteration_first.length(); ii++) {
4139 Node* nnn = _iteration_first.at(ii);
4140 if (same_origin_idx(nnn, nd)) {
4141 #ifndef PRODUCT
4142 if (_vector_loop_debug) {
4143 tty->print_cr("SuperWord::first_node: %d is the first iteration node for %d (_clone_map.idx(nnn->_idx) = %d)",
4144 nnn->_idx, nd->_idx, _clone_map.idx(nnn->_idx));
4145 }
4146 #endif
4147 return nnn;
4148 }
4149 }
4150
4151 #ifndef PRODUCT
4152 if (_vector_loop_debug) {
4153 tty->print_cr("SuperWord::first_node: did not find first iteration node for %d (_clone_map.idx(nd->_idx)=%d)",
4154 nd->_idx, _clone_map.idx(nd->_idx));
4155 }
4156 #endif
4157 return 0;
4158 }
4159
4160 Node* SuperWord::last_node(Node* nd) {
4161 for (int ii = 0; ii < _iteration_last.length(); ii++) {
4162 Node* nnn = _iteration_last.at(ii);
4163 if (same_origin_idx(nnn, nd)) {
4164 #ifndef PRODUCT
4165 if (_vector_loop_debug) {
4166 tty->print_cr("SuperWord::last_node _clone_map.idx(nnn->_idx)=%d, _clone_map.idx(nd->_idx)=%d",
4167 _clone_map.idx(nnn->_idx), _clone_map.idx(nd->_idx));
4168 }
4169 #endif
4170 return nnn;
4171 }
4172 }
4173 return 0;
4174 }
4175
4176 int SuperWord::mark_generations() {
4177 Node *ii_err = NULL, *tail_err = NULL;
4178 for (int i = 0; i < _mem_slice_head.length(); i++) {
4179 Node* phi = _mem_slice_head.at(i);
4180 assert(phi->is_Phi(), "must be phi");
4181
4182 Node* tail = _mem_slice_tail.at(i);
4183 if (_ii_last == -1) {
4184 tail_err = tail;
4185 _ii_last = _clone_map.gen(tail->_idx);
4186 }
4187 else if (_ii_last != _clone_map.gen(tail->_idx)) {
4188 #ifndef PRODUCT
4189 if (TraceSuperWord && Verbose) {
4190 tty->print_cr("SuperWord::mark_generations _ii_last error - found different generations in two tail nodes ");
4191 tail->dump();
4192 tail_err->dump();
4193 }
4194 #endif
4195 return -1;
4196 }
4197
4198 // find first iteration in the loop
4199 for (DUIterator_Fast imax, i = phi->fast_outs(imax); i < imax; i++) {
4200 Node* ii = phi->fast_out(i);
4201 if (in_bb(ii) && ii->is_Store()) { // we speculate that normally Stores of one and one only generation have deps from mem phi
4202 if (_ii_first == -1) {
4203 ii_err = ii;
4204 _ii_first = _clone_map.gen(ii->_idx);
4205 } else if (_ii_first != _clone_map.gen(ii->_idx)) {
4206 #ifndef PRODUCT
4207 if (TraceSuperWord && Verbose) {
4208 tty->print_cr("SuperWord::mark_generations: _ii_first was found before and not equal to one in this node (%d)", _ii_first);
4209 ii->dump();
4210 if (ii_err!= 0) {
4211 ii_err->dump();
4212 }
4213 }
4214 #endif
4215 return -1; // this phi has Stores from different generations of unroll and cannot be simd/vectorized
4216 }
4217 }
4218 }//for (DUIterator_Fast imax,
4219 }//for (int i...
4220
4221 if (_ii_first == -1 || _ii_last == -1) {
4222 if (TraceSuperWord && Verbose) {
4223 tty->print_cr("SuperWord::mark_generations unknown error, something vent wrong");
4224 }
4225 return -1; // something vent wrong
4226 }
4227 // collect nodes in the first and last generations
4228 assert(_iteration_first.length() == 0, "_iteration_first must be empty");
4229 assert(_iteration_last.length() == 0, "_iteration_last must be empty");
4230 for (int j = 0; j < _block.length(); j++) {
4231 Node* n = _block.at(j);
4232 node_idx_t gen = _clone_map.gen(n->_idx);
4233 if ((signed)gen == _ii_first) {
4234 _iteration_first.push(n);
4235 } else if ((signed)gen == _ii_last) {
4236 _iteration_last.push(n);
4237 }
4238 }
4239
4240 // building order of iterations
4241 if (_ii_order.length() == 0 && ii_err != 0) {
4242 assert(in_bb(ii_err) && ii_err->is_Store(), "should be Store in bb");
4243 Node* nd = ii_err;
4244 while(_clone_map.gen(nd->_idx) != _ii_last) {
4245 _ii_order.push(_clone_map.gen(nd->_idx));
4246 bool found = false;
4247 for (DUIterator_Fast imax, i = nd->fast_outs(imax); i < imax; i++) {
4248 Node* use = nd->fast_out(i);
4249 if (same_origin_idx(use, nd) && use->as_Store()->in(MemNode::Memory) == nd) {
4250 found = true;
4251 nd = use;
4252 break;
4253 }
4254 }//for
4255
4256 if (found == false) {
4257 if (TraceSuperWord && Verbose) {
4258 tty->print_cr("SuperWord::mark_generations: Cannot build order of iterations - no dependent Store for %d", nd->_idx);
4259 }
4260 _ii_order.clear();
4261 return -1;
4262 }
4263 } //while
4264 _ii_order.push(_clone_map.gen(nd->_idx));
4265 }
4266
4267 #ifndef PRODUCT
4268 if (_vector_loop_debug) {
4269 tty->print_cr("SuperWord::mark_generations");
4270 tty->print_cr("First generation (%d) nodes:", _ii_first);
4271 for (int ii = 0; ii < _iteration_first.length(); ii++) _iteration_first.at(ii)->dump();
4272 tty->print_cr("Last generation (%d) nodes:", _ii_last);
4273 for (int ii = 0; ii < _iteration_last.length(); ii++) _iteration_last.at(ii)->dump();
4274 tty->print_cr(" ");
4275
4276 tty->print("SuperWord::List of generations: ");
4277 for (int jj = 0; jj < _ii_order.length(); ++jj) {
4278 tty->print("%d:%d ", jj, _ii_order.at(jj));
4279 }
4280 tty->print_cr(" ");
4281 }
4282 #endif
4283
4284 return _ii_first;
4285 }
4286
4287 bool SuperWord::fix_commutative_inputs(Node* gold, Node* fix) {
4288 assert(gold->is_Add() && fix->is_Add() || gold->is_Mul() && fix->is_Mul(), "should be only Add or Mul nodes");
4289 assert(same_origin_idx(gold, fix), "should be clones of the same node");
4290 Node* gin1 = gold->in(1);
4291 Node* gin2 = gold->in(2);
4292 Node* fin1 = fix->in(1);
4293 Node* fin2 = fix->in(2);
4294 bool swapped = false;
4295
4296 if (in_bb(gin1) && in_bb(gin2) && in_bb(fin1) && in_bb(fin1)) {
4297 if (same_origin_idx(gin1, fin1) &&
4298 same_origin_idx(gin2, fin2)) {
4299 return true; // nothing to fix
4300 }
4301 if (same_origin_idx(gin1, fin2) &&
4302 same_origin_idx(gin2, fin1)) {
4303 fix->swap_edges(1, 2);
4304 swapped = true;
4305 }
4306 }
4307 // at least one input comes from outside of bb
4308 if (gin1->_idx == fin1->_idx) {
4309 return true; // nothing to fix
4310 }
4311 if (!swapped && (gin1->_idx == fin2->_idx || gin2->_idx == fin1->_idx)) { //swapping is expensive, check condition first
4312 fix->swap_edges(1, 2);
4313 swapped = true;
4314 }
4315
4316 if (swapped) {
4317 #ifndef PRODUCT
4318 if (_vector_loop_debug) {
4319 tty->print_cr("SuperWord::fix_commutative_inputs: fixed node %d", fix->_idx);
4320 }
4321 #endif
4322 return true;
4323 }
4324
4325 if (TraceSuperWord && Verbose) {
4326 tty->print_cr("SuperWord::fix_commutative_inputs: cannot fix node %d", fix->_idx);
4327 }
4328
4329 return false;
4330 }
4331
4332 bool SuperWord::pack_parallel() {
4333 #ifndef PRODUCT
4334 if (_vector_loop_debug) {
4335 tty->print_cr("SuperWord::pack_parallel: START");
4336 }
4337 #endif
4338
4339 _packset.clear();
4340
4341 for (int ii = 0; ii < _iteration_first.length(); ii++) {
4342 Node* nd = _iteration_first.at(ii);
4343 if (in_bb(nd) && (nd->is_Load() || nd->is_Store() || nd->is_Add() || nd->is_Mul())) {
4344 Node_List* pk = new Node_List();
4345 pk->push(nd);
4346 for (int gen = 1; gen < _ii_order.length(); ++gen) {
4347 for (int kk = 0; kk < _block.length(); kk++) {
4348 Node* clone = _block.at(kk);
4349 if (same_origin_idx(clone, nd) &&
4350 _clone_map.gen(clone->_idx) == _ii_order.at(gen)) {
4351 if (nd->is_Add() || nd->is_Mul()) {
4352 fix_commutative_inputs(nd, clone);
4353 }
4354 pk->push(clone);
4355 if (pk->size() == 4) {
4356 _packset.append(pk);
4357 #ifndef PRODUCT
4358 if (_vector_loop_debug) {
4359 tty->print_cr("SuperWord::pack_parallel: added pack ");
4360 pk->dump();
4361 }
4362 #endif
4363 if (_clone_map.gen(clone->_idx) != _ii_last) {
4364 pk = new Node_List();
4365 }
4366 }
4367 break;
4368 }
4369 }
4370 }//for
4371 }//if
4372 }//for
4373
4374 #ifndef PRODUCT
4375 if (_vector_loop_debug) {
4376 tty->print_cr("SuperWord::pack_parallel: END");
4377 }
4378 #endif
4379
4380 return true;
4381 }
4382
4383 bool SuperWord::hoist_loads_in_graph() {
4384 GrowableArray<Node*> loads;
4385
4386 #ifndef PRODUCT
4387 if (_vector_loop_debug) {
4388 tty->print_cr("SuperWord::hoist_loads_in_graph: total number _mem_slice_head.length() = %d", _mem_slice_head.length());
4389 }
4390 #endif
4391
4392 for (int i = 0; i < _mem_slice_head.length(); i++) {
4393 Node* n = _mem_slice_head.at(i);
4394 if ( !in_bb(n) || !n->is_Phi() || n->bottom_type() != Type::MEMORY) {
4395 if (TraceSuperWord && Verbose) {
4396 tty->print_cr("SuperWord::hoist_loads_in_graph: skipping unexpected node n=%d", n->_idx);
4397 }
4398 continue;
4399 }
4400
4401 #ifndef PRODUCT
4402 if (_vector_loop_debug) {
4403 tty->print_cr("SuperWord::hoist_loads_in_graph: processing phi %d = _mem_slice_head.at(%d);", n->_idx, i);
4404 }
4405 #endif
4406
4407 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
4408 Node* ld = n->fast_out(i);
4409 if (ld->is_Load() && ld->as_Load()->in(MemNode::Memory) == n && in_bb(ld)) {
4410 for (int i = 0; i < _block.length(); i++) {
4411 Node* ld2 = _block.at(i);
4412 if (ld2->is_Load() && same_origin_idx(ld, ld2) &&
4413 !same_generation(ld, ld2)) { // <= do not collect the first generation ld
4414 #ifndef PRODUCT
4415 if (_vector_loop_debug) {
4416 tty->print_cr("SuperWord::hoist_loads_in_graph: will try to hoist load ld2->_idx=%d, cloned from %d (ld->_idx=%d)",
4417 ld2->_idx, _clone_map.idx(ld->_idx), ld->_idx);
4418 }
4419 #endif
4420 // could not do on-the-fly, since iterator is immutable
4421 loads.push(ld2);
4422 }
4423 }// for
4424 }//if
4425 }//for (DUIterator_Fast imax,
4426 }//for (int i = 0; i
4427
4428 for (int i = 0; i < loads.length(); i++) {
4429 LoadNode* ld = loads.at(i)->as_Load();
4430 Node* phi = find_phi_for_mem_dep(ld);
4431 if (phi != NULL) {
4432 #ifndef PRODUCT
4433 if (_vector_loop_debug) {
4434 tty->print_cr("SuperWord::hoist_loads_in_graph replacing MemNode::Memory(%d) edge in %d with one from %d",
4435 MemNode::Memory, ld->_idx, phi->_idx);
4436 }
4437 #endif
4438 _igvn.replace_input_of(ld, MemNode::Memory, phi);
4439 }
4440 }//for
4441
4442 restart(); // invalidate all basic structures, since we rebuilt the graph
4443
4444 if (TraceSuperWord && Verbose) {
4445 tty->print_cr("\nSuperWord::hoist_loads_in_graph() the graph was rebuilt, all structures invalidated and need rebuild");
4446 }
4447
4448 return true;
4449 }
4450