1 /*
2 * Copyright (c) 1997, 2014, 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
25 #include "precompiled.hpp"
26 #include "libadt/vectset.hpp"
27 #include "memory/allocation.inline.hpp"
28 #include "opto/block.hpp"
29 #include "opto/c2compiler.hpp"
30 #include "opto/callnode.hpp"
31 #include "opto/cfgnode.hpp"
32 #include "opto/machnode.hpp"
33 #include "opto/opcodes.hpp"
34 #include "opto/phaseX.hpp"
35 #include "opto/rootnode.hpp"
36 #include "opto/runtime.hpp"
37 #include "runtime/deoptimization.hpp"
38
39 // Portions of code courtesy of Clifford Click
40
41 // Optimization - Graph Style
42
43 // To avoid float value underflow
44 #define MIN_BLOCK_FREQUENCY 1.e-35f
45
46 //----------------------------schedule_node_into_block-------------------------
47 // Insert node n into block b. Look for projections of n and make sure they
48 // are in b also.
49 void PhaseCFG::schedule_node_into_block( Node *n, Block *b ) {
50 // Set basic block of n, Add n to b,
51 map_node_to_block(n, b);
52 b->add_inst(n);
53
54 // After Matching, nearly any old Node may have projections trailing it.
55 // These are usually machine-dependent flags. In any case, they might
56 // float to another block below this one. Move them up.
57 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
58 Node* use = n->fast_out(i);
59 if (use->is_Proj()) {
60 Block* buse = get_block_for_node(use);
61 if (buse != b) { // In wrong block?
62 if (buse != NULL) {
63 buse->find_remove(use); // Remove from wrong block
64 }
65 map_node_to_block(use, b);
66 b->add_inst(use);
67 }
68 }
69 }
70 }
71
72 //----------------------------replace_block_proj_ctrl-------------------------
73 // Nodes that have is_block_proj() nodes as their control need to use
74 // the appropriate Region for their actual block as their control since
75 // the projection will be in a predecessor block.
76 void PhaseCFG::replace_block_proj_ctrl( Node *n ) {
77 const Node *in0 = n->in(0);
78 assert(in0 != NULL, "Only control-dependent");
79 const Node *p = in0->is_block_proj();
80 if (p != NULL && p != n) { // Control from a block projection?
81 assert(!n->pinned() || n->is_MachConstantBase(), "only pinned MachConstantBase node is expected here");
82 // Find trailing Region
83 Block *pb = get_block_for_node(in0); // Block-projection already has basic block
84 uint j = 0;
85 if (pb->_num_succs != 1) { // More then 1 successor?
86 // Search for successor
87 uint max = pb->number_of_nodes();
88 assert( max > 1, "" );
89 uint start = max - pb->_num_succs;
90 // Find which output path belongs to projection
91 for (j = start; j < max; j++) {
92 if( pb->get_node(j) == in0 )
93 break;
94 }
95 assert( j < max, "must find" );
96 // Change control to match head of successor basic block
97 j -= start;
98 }
99 n->set_req(0, pb->_succs[j]->head());
100 }
101 }
102
103
104 //------------------------------schedule_pinned_nodes--------------------------
105 // Set the basic block for Nodes pinned into blocks
106 void PhaseCFG::schedule_pinned_nodes(VectorSet &visited) {
107 // Allocate node stack of size C->unique()+8 to avoid frequent realloc
108 GrowableArray <Node *> spstack(C->unique() + 8);
109 spstack.push(_root);
110 while (spstack.is_nonempty()) {
111 Node* node = spstack.pop();
112 if (!visited.test_set(node->_idx)) { // Test node and flag it as visited
113 if (node->pinned() && !has_block(node)) { // Pinned? Nail it down!
114 assert(node->in(0), "pinned Node must have Control");
115 // Before setting block replace block_proj control edge
116 replace_block_proj_ctrl(node);
117 Node* input = node->in(0);
118 while (!input->is_block_start()) {
119 input = input->in(0);
120 }
121 Block* block = get_block_for_node(input); // Basic block of controlling input
122 schedule_node_into_block(node, block);
123 }
124
125 // process all inputs that are non NULL
126 for (int i = node->req() - 1; i >= 0; --i) {
127 if (node->in(i) != NULL) {
128 spstack.push(node->in(i));
129 }
130 }
131 }
132 }
133 }
134
135 #ifdef ASSERT
136 // Assert that new input b2 is dominated by all previous inputs.
137 // Check this by by seeing that it is dominated by b1, the deepest
138 // input observed until b2.
139 static void assert_dom(Block* b1, Block* b2, Node* n, const PhaseCFG* cfg) {
140 if (b1 == NULL) return;
141 assert(b1->_dom_depth < b2->_dom_depth, "sanity");
142 Block* tmp = b2;
143 while (tmp != b1 && tmp != NULL) {
144 tmp = tmp->_idom;
145 }
146 if (tmp != b1) {
147 // Detected an unschedulable graph. Print some nice stuff and die.
148 tty->print_cr("!!! Unschedulable graph !!!");
149 for (uint j=0; j<n->len(); j++) { // For all inputs
150 Node* inn = n->in(j); // Get input
151 if (inn == NULL) continue; // Ignore NULL, missing inputs
152 Block* inb = cfg->get_block_for_node(inn);
153 tty->print("B%d idom=B%d depth=%2d ",inb->_pre_order,
154 inb->_idom ? inb->_idom->_pre_order : 0, inb->_dom_depth);
155 inn->dump();
156 }
157 tty->print("Failing node: ");
158 n->dump();
159 assert(false, "unscheduable graph");
160 }
161 }
162 #endif
163
164 static Block* find_deepest_input(Node* n, const PhaseCFG* cfg) {
165 // Find the last input dominated by all other inputs.
166 Block* deepb = NULL; // Deepest block so far
167 int deepb_dom_depth = 0;
168 for (uint k = 0; k < n->len(); k++) { // For all inputs
169 Node* inn = n->in(k); // Get input
170 if (inn == NULL) continue; // Ignore NULL, missing inputs
171 Block* inb = cfg->get_block_for_node(inn);
172 assert(inb != NULL, "must already have scheduled this input");
173 if (deepb_dom_depth < (int) inb->_dom_depth) {
174 // The new inb must be dominated by the previous deepb.
175 // The various inputs must be linearly ordered in the dom
176 // tree, or else there will not be a unique deepest block.
177 DEBUG_ONLY(assert_dom(deepb, inb, n, cfg));
178 deepb = inb; // Save deepest block
179 deepb_dom_depth = deepb->_dom_depth;
180 }
181 }
182 assert(deepb != NULL, "must be at least one input to n");
183 return deepb;
184 }
185
186
187 //------------------------------schedule_early---------------------------------
188 // Find the earliest Block any instruction can be placed in. Some instructions
189 // are pinned into Blocks. Unpinned instructions can appear in last block in
190 // which all their inputs occur.
191 bool PhaseCFG::schedule_early(VectorSet &visited, Node_List &roots) {
192 // Allocate stack with enough space to avoid frequent realloc
193 Node_Stack nstack(roots.Size() + 8);
194 // _root will be processed among C->top() inputs
195 roots.push(C->top());
196 visited.set(C->top()->_idx);
197
198 while (roots.size() != 0) {
199 // Use local variables nstack_top_n & nstack_top_i to cache values
200 // on stack's top.
201 Node* parent_node = roots.pop();
202 uint input_index = 0;
203
204 while (true) {
205 if (input_index == 0) {
206 // Fixup some control. Constants without control get attached
207 // to root and nodes that use is_block_proj() nodes should be attached
208 // to the region that starts their block.
209 const Node* control_input = parent_node->in(0);
210 if (control_input != NULL) {
211 replace_block_proj_ctrl(parent_node);
212 } else {
213 // Is a constant with NO inputs?
214 if (parent_node->req() == 1) {
215 parent_node->set_req(0, _root);
216 }
217 }
218 }
219
220 // First, visit all inputs and force them to get a block. If an
221 // input is already in a block we quit following inputs (to avoid
222 // cycles). Instead we put that Node on a worklist to be handled
223 // later (since IT'S inputs may not have a block yet).
224
225 // Assume all n's inputs will be processed
226 bool done = true;
227
228 while (input_index < parent_node->len()) {
229 Node* in = parent_node->in(input_index++);
230 if (in == NULL) {
231 continue;
232 }
233
234 int is_visited = visited.test_set(in->_idx);
235 if (!has_block(in)) {
236 if (is_visited) {
237 return false;
238 }
239 // Save parent node and next input's index.
240 nstack.push(parent_node, input_index);
241 // Process current input now.
242 parent_node = in;
243 input_index = 0;
244 // Not all n's inputs processed.
245 done = false;
246 break;
247 } else if (!is_visited) {
248 // Visit this guy later, using worklist
249 roots.push(in);
250 }
251 }
252
253 if (done) {
254 // All of n's inputs have been processed, complete post-processing.
255
256 // Some instructions are pinned into a block. These include Region,
257 // Phi, Start, Return, and other control-dependent instructions and
258 // any projections which depend on them.
259 if (!parent_node->pinned()) {
260 // Set earliest legal block.
261 Block* earliest_block = find_deepest_input(parent_node, this);
262 map_node_to_block(parent_node, earliest_block);
263 } else {
264 assert(get_block_for_node(parent_node) == get_block_for_node(parent_node->in(0)), "Pinned Node should be at the same block as its control edge");
265 }
266
267 if (nstack.is_empty()) {
268 // Finished all nodes on stack.
269 // Process next node on the worklist 'roots'.
270 break;
271 }
272 // Get saved parent node and next input's index.
273 parent_node = nstack.node();
274 input_index = nstack.index();
275 nstack.pop();
276 }
277 }
278 }
279 return true;
280 }
281
282 //------------------------------dom_lca----------------------------------------
283 // Find least common ancestor in dominator tree
284 // LCA is a current notion of LCA, to be raised above 'this'.
285 // As a convenient boundary condition, return 'this' if LCA is NULL.
286 // Find the LCA of those two nodes.
287 Block* Block::dom_lca(Block* LCA) {
288 if (LCA == NULL || LCA == this) return this;
289
290 Block* anc = this;
291 while (anc->_dom_depth > LCA->_dom_depth)
292 anc = anc->_idom; // Walk up till anc is as high as LCA
293
294 while (LCA->_dom_depth > anc->_dom_depth)
295 LCA = LCA->_idom; // Walk up till LCA is as high as anc
296
297 while (LCA != anc) { // Walk both up till they are the same
298 LCA = LCA->_idom;
299 anc = anc->_idom;
300 }
301
302 return LCA;
303 }
304
305 //--------------------------raise_LCA_above_use--------------------------------
306 // We are placing a definition, and have been given a def->use edge.
307 // The definition must dominate the use, so move the LCA upward in the
308 // dominator tree to dominate the use. If the use is a phi, adjust
309 // the LCA only with the phi input paths which actually use this def.
310 static Block* raise_LCA_above_use(Block* LCA, Node* use, Node* def, const PhaseCFG* cfg) {
311 Block* buse = cfg->get_block_for_node(use);
312 if (buse == NULL) return LCA; // Unused killing Projs have no use block
313 if (!use->is_Phi()) return buse->dom_lca(LCA);
314 uint pmax = use->req(); // Number of Phi inputs
315 // Why does not this loop just break after finding the matching input to
316 // the Phi? Well...it's like this. I do not have true def-use/use-def
317 // chains. Means I cannot distinguish, from the def-use direction, which
318 // of many use-defs lead from the same use to the same def. That is, this
319 // Phi might have several uses of the same def. Each use appears in a
320 // different predecessor block. But when I enter here, I cannot distinguish
321 // which use-def edge I should find the predecessor block for. So I find
322 // them all. Means I do a little extra work if a Phi uses the same value
323 // more than once.
324 for (uint j=1; j<pmax; j++) { // For all inputs
325 if (use->in(j) == def) { // Found matching input?
326 Block* pred = cfg->get_block_for_node(buse->pred(j));
327 LCA = pred->dom_lca(LCA);
328 }
329 }
330 return LCA;
331 }
332
333 //----------------------------raise_LCA_above_marks----------------------------
334 // Return a new LCA that dominates LCA and any of its marked predecessors.
335 // Search all my parents up to 'early' (exclusive), looking for predecessors
336 // which are marked with the given index. Return the LCA (in the dom tree)
337 // of all marked blocks. If there are none marked, return the original
338 // LCA.
339 static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark, Block* early, const PhaseCFG* cfg) {
340 Block_List worklist;
341 worklist.push(LCA);
342 while (worklist.size() > 0) {
343 Block* mid = worklist.pop();
344 if (mid == early) continue; // stop searching here
345
346 // Test and set the visited bit.
347 if (mid->raise_LCA_visited() == mark) continue; // already visited
348
349 // Don't process the current LCA, otherwise the search may terminate early
350 if (mid != LCA && mid->raise_LCA_mark() == mark) {
351 // Raise the LCA.
352 LCA = mid->dom_lca(LCA);
353 if (LCA == early) break; // stop searching everywhere
354 assert(early->dominates(LCA), "early is high enough");
355 // Resume searching at that point, skipping intermediate levels.
356 worklist.push(LCA);
357 if (LCA == mid)
358 continue; // Don't mark as visited to avoid early termination.
359 } else {
360 // Keep searching through this block's predecessors.
361 for (uint j = 1, jmax = mid->num_preds(); j < jmax; j++) {
362 Block* mid_parent = cfg->get_block_for_node(mid->pred(j));
363 worklist.push(mid_parent);
364 }
365 }
366 mid->set_raise_LCA_visited(mark);
367 }
368 return LCA;
369 }
370
371 //--------------------------memory_early_block--------------------------------
372 // This is a variation of find_deepest_input, the heart of schedule_early.
373 // Find the "early" block for a load, if we considered only memory and
374 // address inputs, that is, if other data inputs were ignored.
375 //
376 // Because a subset of edges are considered, the resulting block will
377 // be earlier (at a shallower dom_depth) than the true schedule_early
378 // point of the node. We compute this earlier block as a more permissive
379 // site for anti-dependency insertion, but only if subsume_loads is enabled.
380 static Block* memory_early_block(Node* load, Block* early, const PhaseCFG* cfg) {
381 Node* base;
382 Node* index;
383 Node* store = load->in(MemNode::Memory);
384 load->as_Mach()->memory_inputs(base, index);
385
386 assert(base != NodeSentinel && index != NodeSentinel,
387 "unexpected base/index inputs");
388
389 Node* mem_inputs[4];
390 int mem_inputs_length = 0;
391 if (base != NULL) mem_inputs[mem_inputs_length++] = base;
392 if (index != NULL) mem_inputs[mem_inputs_length++] = index;
393 if (store != NULL) mem_inputs[mem_inputs_length++] = store;
394
395 // In the comparision below, add one to account for the control input,
396 // which may be null, but always takes up a spot in the in array.
397 if (mem_inputs_length + 1 < (int) load->req()) {
398 // This "load" has more inputs than just the memory, base and index inputs.
399 // For purposes of checking anti-dependences, we need to start
400 // from the early block of only the address portion of the instruction,
401 // and ignore other blocks that may have factored into the wider
402 // schedule_early calculation.
403 if (load->in(0) != NULL) mem_inputs[mem_inputs_length++] = load->in(0);
404
405 Block* deepb = NULL; // Deepest block so far
406 int deepb_dom_depth = 0;
407 for (int i = 0; i < mem_inputs_length; i++) {
408 Block* inb = cfg->get_block_for_node(mem_inputs[i]);
409 if (deepb_dom_depth < (int) inb->_dom_depth) {
410 // The new inb must be dominated by the previous deepb.
411 // The various inputs must be linearly ordered in the dom
412 // tree, or else there will not be a unique deepest block.
413 DEBUG_ONLY(assert_dom(deepb, inb, load, cfg));
414 deepb = inb; // Save deepest block
415 deepb_dom_depth = deepb->_dom_depth;
416 }
417 }
418 early = deepb;
419 }
420
421 return early;
422 }
423
424 //--------------------------insert_anti_dependences---------------------------
425 // A load may need to witness memory that nearby stores can overwrite.
426 // For each nearby store, either insert an "anti-dependence" edge
427 // from the load to the store, or else move LCA upward to force the
428 // load to (eventually) be scheduled in a block above the store.
429 //
430 // Do not add edges to stores on distinct control-flow paths;
431 // only add edges to stores which might interfere.
432 //
433 // Return the (updated) LCA. There will not be any possibly interfering
434 // store between the load's "early block" and the updated LCA.
435 // Any stores in the updated LCA will have new precedence edges
436 // back to the load. The caller is expected to schedule the load
437 // in the LCA, in which case the precedence edges will make LCM
438 // preserve anti-dependences. The caller may also hoist the load
439 // above the LCA, if it is not the early block.
440 Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
441 assert(load->needs_anti_dependence_check(), "must be a load of some sort");
442 assert(LCA != NULL, "");
443 DEBUG_ONLY(Block* LCA_orig = LCA);
444
445 // Compute the alias index. Loads and stores with different alias indices
446 // do not need anti-dependence edges.
447 uint load_alias_idx = C->get_alias_index(load->adr_type());
448 #ifdef ASSERT
449 if (load_alias_idx == Compile::AliasIdxBot && C->AliasLevel() > 0 &&
450 (PrintOpto || VerifyAliases ||
451 PrintMiscellaneous && (WizardMode || Verbose))) {
452 // Load nodes should not consume all of memory.
453 // Reporting a bottom type indicates a bug in adlc.
454 // If some particular type of node validly consumes all of memory,
455 // sharpen the preceding "if" to exclude it, so we can catch bugs here.
456 tty->print_cr("*** Possible Anti-Dependence Bug: Load consumes all of memory.");
457 load->dump(2);
458 if (VerifyAliases) assert(load_alias_idx != Compile::AliasIdxBot, "");
459 }
460 #endif
461 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrComp),
462 "String compare is only known 'load' that does not conflict with any stores");
463 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrEquals),
464 "String equals is a 'load' that does not conflict with any stores");
465 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrIndexOf),
466 "String indexOf is a 'load' that does not conflict with any stores");
467 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_AryEq),
468 "Arrays equals is a 'load' that do not conflict with any stores");
469
470 if (!C->alias_type(load_alias_idx)->is_rewritable()) {
471 // It is impossible to spoil this load by putting stores before it,
472 // because we know that the stores will never update the value
473 // which 'load' must witness.
474 return LCA;
475 }
476
477 node_idx_t load_index = load->_idx;
478
479 // Note the earliest legal placement of 'load', as determined by
480 // by the unique point in the dom tree where all memory effects
481 // and other inputs are first available. (Computed by schedule_early.)
482 // For normal loads, 'early' is the shallowest place (dom graph wise)
483 // to look for anti-deps between this load and any store.
484 Block* early = get_block_for_node(load);
485
486 // If we are subsuming loads, compute an "early" block that only considers
487 // memory or address inputs. This block may be different than the
488 // schedule_early block in that it could be at an even shallower depth in the
489 // dominator tree, and allow for a broader discovery of anti-dependences.
490 if (C->subsume_loads()) {
491 early = memory_early_block(load, early, this);
492 }
493
494 ResourceArea *area = Thread::current()->resource_area();
495 Node_List worklist_mem(area); // prior memory state to store
496 Node_List worklist_store(area); // possible-def to explore
497 Node_List worklist_visited(area); // visited mergemem nodes
498 Node_List non_early_stores(area); // all relevant stores outside of early
499 bool must_raise_LCA = false;
500
501 #ifdef TRACK_PHI_INPUTS
502 // %%% This extra checking fails because MergeMem nodes are not GVNed.
503 // Provide "phi_inputs" to check if every input to a PhiNode is from the
504 // original memory state. This indicates a PhiNode for which should not
505 // prevent the load from sinking. For such a block, set_raise_LCA_mark
506 // may be overly conservative.
507 // Mechanism: count inputs seen for each Phi encountered in worklist_store.
508 DEBUG_ONLY(GrowableArray<uint> phi_inputs(area, C->unique(),0,0));
509 #endif
510
511 // 'load' uses some memory state; look for users of the same state.
512 // Recurse through MergeMem nodes to the stores that use them.
513
514 // Each of these stores is a possible definition of memory
515 // that 'load' needs to use. We need to force 'load'
516 // to occur before each such store. When the store is in
517 // the same block as 'load', we insert an anti-dependence
518 // edge load->store.
519
520 // The relevant stores "nearby" the load consist of a tree rooted
521 // at initial_mem, with internal nodes of type MergeMem.
522 // Therefore, the branches visited by the worklist are of this form:
523 // initial_mem -> (MergeMem ->)* store
524 // The anti-dependence constraints apply only to the fringe of this tree.
525
526 Node* initial_mem = load->in(MemNode::Memory);
527 worklist_store.push(initial_mem);
528 worklist_visited.push(initial_mem);
529 worklist_mem.push(NULL);
530 while (worklist_store.size() > 0) {
531 // Examine a nearby store to see if it might interfere with our load.
532 Node* mem = worklist_mem.pop();
533 Node* store = worklist_store.pop();
534 uint op = store->Opcode();
535
536 // MergeMems do not directly have anti-deps.
537 // Treat them as internal nodes in a forward tree of memory states,
538 // the leaves of which are each a 'possible-def'.
539 if (store == initial_mem // root (exclusive) of tree we are searching
540 || op == Op_MergeMem // internal node of tree we are searching
541 ) {
542 mem = store; // It's not a possibly interfering store.
543 if (store == initial_mem)
544 initial_mem = NULL; // only process initial memory once
545
546 for (DUIterator_Fast imax, i = mem->fast_outs(imax); i < imax; i++) {
547 store = mem->fast_out(i);
548 if (store->is_MergeMem()) {
549 // Be sure we don't get into combinatorial problems.
550 // (Allow phis to be repeated; they can merge two relevant states.)
551 uint j = worklist_visited.size();
552 for (; j > 0; j--) {
553 if (worklist_visited.at(j-1) == store) break;
554 }
555 if (j > 0) continue; // already on work list; do not repeat
556 worklist_visited.push(store);
557 }
558 worklist_mem.push(mem);
559 worklist_store.push(store);
560 }
561 continue;
562 }
563
564 if (op == Op_MachProj || op == Op_Catch) continue;
565 if (store->needs_anti_dependence_check()) continue; // not really a store
566
567 // Compute the alias index. Loads and stores with different alias
568 // indices do not need anti-dependence edges. Wide MemBar's are
569 // anti-dependent on everything (except immutable memories).
570 const TypePtr* adr_type = store->adr_type();
571 if (!C->can_alias(adr_type, load_alias_idx)) continue;
572
573 // Most slow-path runtime calls do NOT modify Java memory, but
574 // they can block and so write Raw memory.
575 if (store->is_Mach()) {
576 MachNode* mstore = store->as_Mach();
577 if (load_alias_idx != Compile::AliasIdxRaw) {
578 // Check for call into the runtime using the Java calling
579 // convention (and from there into a wrapper); it has no
580 // _method. Can't do this optimization for Native calls because
581 // they CAN write to Java memory.
582 if (mstore->ideal_Opcode() == Op_CallStaticJava) {
583 assert(mstore->is_MachSafePoint(), "");
584 MachSafePointNode* ms = (MachSafePointNode*) mstore;
585 assert(ms->is_MachCallJava(), "");
586 MachCallJavaNode* mcj = (MachCallJavaNode*) ms;
587 if (mcj->_method == NULL) {
588 // These runtime calls do not write to Java visible memory
589 // (other than Raw) and so do not require anti-dependence edges.
590 continue;
591 }
592 }
593 // Same for SafePoints: they read/write Raw but only read otherwise.
594 // This is basically a workaround for SafePoints only defining control
595 // instead of control + memory.
596 if (mstore->ideal_Opcode() == Op_SafePoint)
597 continue;
598 } else {
599 // Some raw memory, such as the load of "top" at an allocation,
600 // can be control dependent on the previous safepoint. See
601 // comments in GraphKit::allocate_heap() about control input.
602 // Inserting an anti-dep between such a safepoint and a use
603 // creates a cycle, and will cause a subsequent failure in
604 // local scheduling. (BugId 4919904)
605 // (%%% How can a control input be a safepoint and not a projection??)
606 if (mstore->ideal_Opcode() == Op_SafePoint && load->in(0) == mstore)
607 continue;
608 }
609 }
610
611 // Identify a block that the current load must be above,
612 // or else observe that 'store' is all the way up in the
613 // earliest legal block for 'load'. In the latter case,
614 // immediately insert an anti-dependence edge.
615 Block* store_block = get_block_for_node(store);
616 assert(store_block != NULL, "unused killing projections skipped above");
617
618 if (store->is_Phi()) {
619 // 'load' uses memory which is one (or more) of the Phi's inputs.
620 // It must be scheduled not before the Phi, but rather before
621 // each of the relevant Phi inputs.
622 //
623 // Instead of finding the LCA of all inputs to a Phi that match 'mem',
624 // we mark each corresponding predecessor block and do a combined
625 // hoisting operation later (raise_LCA_above_marks).
626 //
627 // Do not assert(store_block != early, "Phi merging memory after access")
628 // PhiNode may be at start of block 'early' with backedge to 'early'
629 DEBUG_ONLY(bool found_match = false);
630 for (uint j = PhiNode::Input, jmax = store->req(); j < jmax; j++) {
631 if (store->in(j) == mem) { // Found matching input?
632 DEBUG_ONLY(found_match = true);
633 Block* pred_block = get_block_for_node(store_block->pred(j));
634 if (pred_block != early) {
635 // If any predecessor of the Phi matches the load's "early block",
636 // we do not need a precedence edge between the Phi and 'load'
637 // since the load will be forced into a block preceding the Phi.
638 pred_block->set_raise_LCA_mark(load_index);
639 assert(!LCA_orig->dominates(pred_block) ||
640 early->dominates(pred_block), "early is high enough");
641 must_raise_LCA = true;
642 } else {
643 // anti-dependent upon PHI pinned below 'early', no edge needed
644 LCA = early; // but can not schedule below 'early'
645 }
646 }
647 }
648 assert(found_match, "no worklist bug");
649 #ifdef TRACK_PHI_INPUTS
650 #ifdef ASSERT
651 // This assert asks about correct handling of PhiNodes, which may not
652 // have all input edges directly from 'mem'. See BugId 4621264
653 int num_mem_inputs = phi_inputs.at_grow(store->_idx,0) + 1;
654 // Increment by exactly one even if there are multiple copies of 'mem'
655 // coming into the phi, because we will run this block several times
656 // if there are several copies of 'mem'. (That's how DU iterators work.)
657 phi_inputs.at_put(store->_idx, num_mem_inputs);
658 assert(PhiNode::Input + num_mem_inputs < store->req(),
659 "Expect at least one phi input will not be from original memory state");
660 #endif //ASSERT
661 #endif //TRACK_PHI_INPUTS
662 } else if (store_block != early) {
663 // 'store' is between the current LCA and earliest possible block.
664 // Label its block, and decide later on how to raise the LCA
665 // to include the effect on LCA of this store.
666 // If this store's block gets chosen as the raised LCA, we
667 // will find him on the non_early_stores list and stick him
668 // with a precedence edge.
669 // (But, don't bother if LCA is already raised all the way.)
670 if (LCA != early) {
671 store_block->set_raise_LCA_mark(load_index);
672 must_raise_LCA = true;
673 non_early_stores.push(store);
674 }
675 } else {
676 // Found a possibly-interfering store in the load's 'early' block.
677 // This means 'load' cannot sink at all in the dominator tree.
678 // Add an anti-dep edge, and squeeze 'load' into the highest block.
679 assert(store != load->in(0), "dependence cycle found");
680 if (verify) {
681 assert(store->find_edge(load) != -1, "missing precedence edge");
682 } else {
683 store->add_prec(load);
684 }
685 LCA = early;
686 // This turns off the process of gathering non_early_stores.
687 }
688 }
689 // (Worklist is now empty; all nearby stores have been visited.)
690
691 // Finished if 'load' must be scheduled in its 'early' block.
692 // If we found any stores there, they have already been given
693 // precedence edges.
694 if (LCA == early) return LCA;
695
696 // We get here only if there are no possibly-interfering stores
697 // in the load's 'early' block. Move LCA up above all predecessors
698 // which contain stores we have noted.
699 //
700 // The raised LCA block can be a home to such interfering stores,
701 // but its predecessors must not contain any such stores.
702 //
703 // The raised LCA will be a lower bound for placing the load,
704 // preventing the load from sinking past any block containing
705 // a store that may invalidate the memory state required by 'load'.
706 if (must_raise_LCA)
707 LCA = raise_LCA_above_marks(LCA, load->_idx, early, this);
708 if (LCA == early) return LCA;
709
710 // Insert anti-dependence edges from 'load' to each store
711 // in the non-early LCA block.
712 // Mine the non_early_stores list for such stores.
713 if (LCA->raise_LCA_mark() == load_index) {
714 while (non_early_stores.size() > 0) {
715 Node* store = non_early_stores.pop();
716 Block* store_block = get_block_for_node(store);
717 if (store_block == LCA) {
718 // add anti_dependence from store to load in its own block
719 assert(store != load->in(0), "dependence cycle found");
720 if (verify) {
721 assert(store->find_edge(load) != -1, "missing precedence edge");
722 } else {
723 store->add_prec(load);
724 }
725 } else {
726 assert(store_block->raise_LCA_mark() == load_index, "block was marked");
727 // Any other stores we found must be either inside the new LCA
728 // or else outside the original LCA. In the latter case, they
729 // did not interfere with any use of 'load'.
730 assert(LCA->dominates(store_block)
731 || !LCA_orig->dominates(store_block), "no stray stores");
732 }
733 }
734 }
735
736 // Return the highest block containing stores; any stores
737 // within that block have been given anti-dependence edges.
738 return LCA;
739 }
740
741 // This class is used to iterate backwards over the nodes in the graph.
742
743 class Node_Backward_Iterator {
744
745 private:
746 Node_Backward_Iterator();
747
748 public:
749 // Constructor for the iterator
750 Node_Backward_Iterator(Node *root, VectorSet &visited, Node_List &stack, PhaseCFG &cfg);
751
752 // Postincrement operator to iterate over the nodes
753 Node *next();
754
755 private:
756 VectorSet &_visited;
757 Node_List &_stack;
758 PhaseCFG &_cfg;
759 };
760
761 // Constructor for the Node_Backward_Iterator
762 Node_Backward_Iterator::Node_Backward_Iterator( Node *root, VectorSet &visited, Node_List &stack, PhaseCFG &cfg)
763 : _visited(visited), _stack(stack), _cfg(cfg) {
764 // The stack should contain exactly the root
765 stack.clear();
766 stack.push(root);
767
768 // Clear the visited bits
769 visited.Clear();
770 }
771
772 // Iterator for the Node_Backward_Iterator
773 Node *Node_Backward_Iterator::next() {
774
775 // If the _stack is empty, then just return NULL: finished.
776 if ( !_stack.size() )
777 return NULL;
778
779 // '_stack' is emulating a real _stack. The 'visit-all-users' loop has been
780 // made stateless, so I do not need to record the index 'i' on my _stack.
781 // Instead I visit all users each time, scanning for unvisited users.
782 // I visit unvisited not-anti-dependence users first, then anti-dependent
783 // children next.
784 Node *self = _stack.pop();
785
786 // I cycle here when I am entering a deeper level of recursion.
787 // The key variable 'self' was set prior to jumping here.
788 while( 1 ) {
789
790 _visited.set(self->_idx);
791
792 // Now schedule all uses as late as possible.
793 const Node* src = self->is_Proj() ? self->in(0) : self;
794 uint src_rpo = _cfg.get_block_for_node(src)->_rpo;
795
796 // Schedule all nodes in a post-order visit
797 Node *unvisited = NULL; // Unvisited anti-dependent Node, if any
798
799 // Scan for unvisited nodes
800 for (DUIterator_Fast imax, i = self->fast_outs(imax); i < imax; i++) {
801 // For all uses, schedule late
802 Node* n = self->fast_out(i); // Use
803
804 // Skip already visited children
805 if ( _visited.test(n->_idx) )
806 continue;
807
808 // do not traverse backward control edges
809 Node *use = n->is_Proj() ? n->in(0) : n;
810 uint use_rpo = _cfg.get_block_for_node(use)->_rpo;
811
812 if ( use_rpo < src_rpo )
813 continue;
814
815 // Phi nodes always precede uses in a basic block
816 if ( use_rpo == src_rpo && use->is_Phi() )
817 continue;
818
819 unvisited = n; // Found unvisited
820
821 // Check for possible-anti-dependent
822 if( !n->needs_anti_dependence_check() )
823 break; // Not visited, not anti-dep; schedule it NOW
824 }
825
826 // Did I find an unvisited not-anti-dependent Node?
827 if ( !unvisited )
828 break; // All done with children; post-visit 'self'
829
830 // Visit the unvisited Node. Contains the obvious push to
831 // indicate I'm entering a deeper level of recursion. I push the
832 // old state onto the _stack and set a new state and loop (recurse).
833 _stack.push(self);
834 self = unvisited;
835 } // End recursion loop
836
837 return self;
838 }
839
840 //------------------------------ComputeLatenciesBackwards----------------------
841 // Compute the latency of all the instructions.
842 void PhaseCFG::compute_latencies_backwards(VectorSet &visited, Node_List &stack) {
843 #ifndef PRODUCT
844 if (trace_opto_pipelining())
845 tty->print("\n#---- ComputeLatenciesBackwards ----\n");
846 #endif
847
848 Node_Backward_Iterator iter((Node *)_root, visited, stack, *this);
849 Node *n;
850
851 // Walk over all the nodes from last to first
852 while (n = iter.next()) {
853 // Set the latency for the definitions of this instruction
854 partial_latency_of_defs(n);
855 }
856 } // end ComputeLatenciesBackwards
857
858 //------------------------------partial_latency_of_defs------------------------
859 // Compute the latency impact of this node on all defs. This computes
860 // a number that increases as we approach the beginning of the routine.
861 void PhaseCFG::partial_latency_of_defs(Node *n) {
862 // Set the latency for this instruction
863 #ifndef PRODUCT
864 if (trace_opto_pipelining()) {
865 tty->print("# latency_to_inputs: node_latency[%d] = %d for node", n->_idx, get_latency_for_node(n));
866 dump();
867 }
868 #endif
869
870 if (n->is_Proj()) {
871 n = n->in(0);
872 }
873
874 if (n->is_Root()) {
875 return;
876 }
877
878 uint nlen = n->len();
879 uint use_latency = get_latency_for_node(n);
880 uint use_pre_order = get_block_for_node(n)->_pre_order;
881
882 for (uint j = 0; j < nlen; j++) {
883 Node *def = n->in(j);
884
885 if (!def || def == n) {
886 continue;
887 }
888
889 // Walk backwards thru projections
890 if (def->is_Proj()) {
891 def = def->in(0);
892 }
893
894 #ifndef PRODUCT
895 if (trace_opto_pipelining()) {
896 tty->print("# in(%2d): ", j);
897 def->dump();
898 }
899 #endif
900
901 // If the defining block is not known, assume it is ok
902 Block *def_block = get_block_for_node(def);
903 uint def_pre_order = def_block ? def_block->_pre_order : 0;
904
905 if ((use_pre_order < def_pre_order) || (use_pre_order == def_pre_order && n->is_Phi())) {
906 continue;
907 }
908
909 uint delta_latency = n->latency(j);
910 uint current_latency = delta_latency + use_latency;
911
912 if (get_latency_for_node(def) < current_latency) {
913 set_latency_for_node(def, current_latency);
914 }
915
916 #ifndef PRODUCT
917 if (trace_opto_pipelining()) {
918 tty->print_cr("# %d + edge_latency(%d) == %d -> %d, node_latency[%d] = %d", use_latency, j, delta_latency, current_latency, def->_idx, get_latency_for_node(def));
919 }
920 #endif
921 }
922 }
923
924 //------------------------------latency_from_use-------------------------------
925 // Compute the latency of a specific use
926 int PhaseCFG::latency_from_use(Node *n, const Node *def, Node *use) {
927 // If self-reference, return no latency
928 if (use == n || use->is_Root()) {
929 return 0;
930 }
931
932 uint def_pre_order = get_block_for_node(def)->_pre_order;
933 uint latency = 0;
934
935 // If the use is not a projection, then it is simple...
936 if (!use->is_Proj()) {
937 #ifndef PRODUCT
938 if (trace_opto_pipelining()) {
939 tty->print("# out(): ");
940 use->dump();
941 }
942 #endif
943
944 uint use_pre_order = get_block_for_node(use)->_pre_order;
945
946 if (use_pre_order < def_pre_order)
947 return 0;
948
949 if (use_pre_order == def_pre_order && use->is_Phi())
950 return 0;
951
952 uint nlen = use->len();
953 uint nl = get_latency_for_node(use);
954
955 for ( uint j=0; j<nlen; j++ ) {
956 if (use->in(j) == n) {
957 // Change this if we want local latencies
958 uint ul = use->latency(j);
959 uint l = ul + nl;
960 if (latency < l) latency = l;
961 #ifndef PRODUCT
962 if (trace_opto_pipelining()) {
963 tty->print_cr("# %d + edge_latency(%d) == %d -> %d, latency = %d",
964 nl, j, ul, l, latency);
965 }
966 #endif
967 }
968 }
969 } else {
970 // This is a projection, just grab the latency of the use(s)
971 for (DUIterator_Fast jmax, j = use->fast_outs(jmax); j < jmax; j++) {
972 uint l = latency_from_use(use, def, use->fast_out(j));
973 if (latency < l) latency = l;
974 }
975 }
976
977 return latency;
978 }
979
980 //------------------------------latency_from_uses------------------------------
981 // Compute the latency of this instruction relative to all of it's uses.
982 // This computes a number that increases as we approach the beginning of the
983 // routine.
984 void PhaseCFG::latency_from_uses(Node *n) {
985 // Set the latency for this instruction
986 #ifndef PRODUCT
987 if (trace_opto_pipelining()) {
988 tty->print("# latency_from_outputs: node_latency[%d] = %d for node", n->_idx, get_latency_for_node(n));
989 dump();
990 }
991 #endif
992 uint latency=0;
993 const Node *def = n->is_Proj() ? n->in(0): n;
994
995 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
996 uint l = latency_from_use(n, def, n->fast_out(i));
997
998 if (latency < l) latency = l;
999 }
1000
1001 set_latency_for_node(n, latency);
1002 }
1003
1004 //------------------------------hoist_to_cheaper_block-------------------------
1005 // Pick a block for node self, between early and LCA, that is a cheaper
1006 // alternative to LCA.
1007 Block* PhaseCFG::hoist_to_cheaper_block(Block* LCA, Block* early, Node* self) {
1008 const double delta = 1+PROB_UNLIKELY_MAG(4);
1009 Block* least = LCA;
1010 double least_freq = least->_freq;
1011 uint target = get_latency_for_node(self);
1012 uint start_latency = get_latency_for_node(LCA->head());
1013 uint end_latency = get_latency_for_node(LCA->get_node(LCA->end_idx()));
1014 bool in_latency = (target <= start_latency);
1015 const Block* root_block = get_block_for_node(_root);
1016
1017 // Turn off latency scheduling if scheduling is just plain off
1018 if (!C->do_scheduling())
1019 in_latency = true;
1020
1021 // Do not hoist (to cover latency) instructions which target a
1022 // single register. Hoisting stretches the live range of the
1023 // single register and may force spilling.
1024 MachNode* mach = self->is_Mach() ? self->as_Mach() : NULL;
1025 if (mach && mach->out_RegMask().is_bound1() && mach->out_RegMask().is_NotEmpty())
1026 in_latency = true;
1027
1028 #ifndef PRODUCT
1029 if (trace_opto_pipelining()) {
1030 tty->print("# Find cheaper block for latency %d: ", get_latency_for_node(self));
1031 self->dump();
1032 tty->print_cr("# B%d: start latency for [%4d]=%d, end latency for [%4d]=%d, freq=%g",
1033 LCA->_pre_order,
1034 LCA->head()->_idx,
1035 start_latency,
1036 LCA->get_node(LCA->end_idx())->_idx,
1037 end_latency,
1038 least_freq);
1039 }
1040 #endif
1041
1042 int cand_cnt = 0; // number of candidates tried
1043
1044 // Walk up the dominator tree from LCA (Lowest common ancestor) to
1045 // the earliest legal location. Capture the least execution frequency.
1046 while (LCA != early) {
1047 LCA = LCA->_idom; // Follow up the dominator tree
1048
1049 if (LCA == NULL) {
1050 // Bailout without retry
1051 C->record_method_not_compilable("late schedule failed: LCA == NULL");
1052 return least;
1053 }
1054
1055 // Don't hoist machine instructions to the root basic block
1056 if (mach && LCA == root_block)
1057 break;
1058
1059 uint start_lat = get_latency_for_node(LCA->head());
1060 uint end_idx = LCA->end_idx();
1061 uint end_lat = get_latency_for_node(LCA->get_node(end_idx));
1062 double LCA_freq = LCA->_freq;
1063 #ifndef PRODUCT
1064 if (trace_opto_pipelining()) {
1065 tty->print_cr("# B%d: start latency for [%4d]=%d, end latency for [%4d]=%d, freq=%g",
1066 LCA->_pre_order, LCA->head()->_idx, start_lat, end_idx, end_lat, LCA_freq);
1067 }
1068 #endif
1069 cand_cnt++;
1070 if (LCA_freq < least_freq || // Better Frequency
1071 (StressGCM && Compile::randomized_select(cand_cnt)) || // Should be randomly accepted in stress mode
1072 (!StressGCM && // Otherwise, choose with latency
1073 !in_latency && // No block containing latency
1074 LCA_freq < least_freq * delta && // No worse frequency
1075 target >= end_lat && // within latency range
1076 !self->is_iteratively_computed() ) // But don't hoist IV increments
1077 // because they may end up above other uses of their phi forcing
1078 // their result register to be different from their input.
1079 ) {
1080 least = LCA; // Found cheaper block
1081 least_freq = LCA_freq;
1082 start_latency = start_lat;
1083 end_latency = end_lat;
1084 if (target <= start_lat)
1085 in_latency = true;
1086 }
1087 }
1088
1089 #ifndef PRODUCT
1090 if (trace_opto_pipelining()) {
1091 tty->print_cr("# Choose block B%d with start latency=%d and freq=%g",
1092 least->_pre_order, start_latency, least_freq);
1093 }
1094 #endif
1095
1096 // See if the latency needs to be updated
1097 if (target < end_latency) {
1098 #ifndef PRODUCT
1099 if (trace_opto_pipelining()) {
1100 tty->print_cr("# Change latency for [%4d] from %d to %d", self->_idx, target, end_latency);
1101 }
1102 #endif
1103 set_latency_for_node(self, end_latency);
1104 partial_latency_of_defs(self);
1105 }
1106
1107 return least;
1108 }
1109
1110
1111 //------------------------------schedule_late-----------------------------------
1112 // Now schedule all codes as LATE as possible. This is the LCA in the
1113 // dominator tree of all USES of a value. Pick the block with the least
1114 // loop nesting depth that is lowest in the dominator tree.
1115 extern const char must_clone[];
1116 void PhaseCFG::schedule_late(VectorSet &visited, Node_List &stack) {
1117 #ifndef PRODUCT
1118 if (trace_opto_pipelining())
1119 tty->print("\n#---- schedule_late ----\n");
1120 #endif
1121
1122 Node_Backward_Iterator iter((Node *)_root, visited, stack, *this);
1123 Node *self;
1124
1125 // Walk over all the nodes from last to first
1126 while (self = iter.next()) {
1127 Block* early = get_block_for_node(self); // Earliest legal placement
1128
1129 if (self->is_top()) {
1130 // Top node goes in bb #2 with other constants.
1131 // It must be special-cased, because it has no out edges.
1132 early->add_inst(self);
1133 continue;
1134 }
1135
1136 // No uses, just terminate
1137 if (self->outcnt() == 0) {
1138 assert(self->is_MachProj(), "sanity");
1139 continue; // Must be a dead machine projection
1140 }
1141
1142 // If node is pinned in the block, then no scheduling can be done.
1143 if( self->pinned() ) // Pinned in block?
1144 continue;
1145
1146 MachNode* mach = self->is_Mach() ? self->as_Mach() : NULL;
1147 if (mach) {
1148 switch (mach->ideal_Opcode()) {
1149 case Op_CreateEx:
1150 // Don't move exception creation
1151 early->add_inst(self);
1152 continue;
1153 break;
1154 case Op_CheckCastPP:
1155 // Don't move CheckCastPP nodes away from their input, if the input
1156 // is a rawptr (5071820).
1157 Node *def = self->in(1);
1158 if (def != NULL && def->bottom_type()->base() == Type::RawPtr) {
1159 early->add_inst(self);
1160 #ifdef ASSERT
1161 _raw_oops.push(def);
1162 #endif
1163 continue;
1164 }
1165 break;
1166 }
1167 }
1168
1169 // Gather LCA of all uses
1170 Block *LCA = NULL;
1171 {
1172 for (DUIterator_Fast imax, i = self->fast_outs(imax); i < imax; i++) {
1173 // For all uses, find LCA
1174 Node* use = self->fast_out(i);
1175 LCA = raise_LCA_above_use(LCA, use, self, this);
1176 }
1177 } // (Hide defs of imax, i from rest of block.)
1178
1179 // Place temps in the block of their use. This isn't a
1180 // requirement for correctness but it reduces useless
1181 // interference between temps and other nodes.
1182 if (mach != NULL && mach->is_MachTemp()) {
1183 map_node_to_block(self, LCA);
1184 LCA->add_inst(self);
1185 continue;
1186 }
1187
1188 // Check if 'self' could be anti-dependent on memory
1189 if (self->needs_anti_dependence_check()) {
1190 // Hoist LCA above possible-defs and insert anti-dependences to
1191 // defs in new LCA block.
1192 LCA = insert_anti_dependences(LCA, self);
1193 }
1194
1195 if (early->_dom_depth > LCA->_dom_depth) {
1196 // Somehow the LCA has moved above the earliest legal point.
1197 // (One way this can happen is via memory_early_block.)
1198 if (C->subsume_loads() == true && !C->failing()) {
1199 // Retry with subsume_loads == false
1200 // If this is the first failure, the sentinel string will "stick"
1201 // to the Compile object, and the C2Compiler will see it and retry.
1202 C->record_failure(C2Compiler::retry_no_subsuming_loads());
1203 } else {
1204 // Bailout without retry when (early->_dom_depth > LCA->_dom_depth)
1205 C->record_method_not_compilable("late schedule failed: incorrect graph");
1206 }
1207 return;
1208 }
1209
1210 // If there is no opportunity to hoist, then we're done.
1211 // In stress mode, try to hoist even the single operations.
1212 bool try_to_hoist = StressGCM || (LCA != early);
1213
1214 // Must clone guys stay next to use; no hoisting allowed.
1215 // Also cannot hoist guys that alter memory or are otherwise not
1216 // allocatable (hoisting can make a value live longer, leading to
1217 // anti and output dependency problems which are normally resolved
1218 // by the register allocator giving everyone a different register).
1219 if (mach != NULL && must_clone[mach->ideal_Opcode()])
1220 try_to_hoist = false;
1221
1222 Block* late = NULL;
1223 if (try_to_hoist) {
1224 // Now find the block with the least execution frequency.
1225 // Start at the latest schedule and work up to the earliest schedule
1226 // in the dominator tree. Thus the Node will dominate all its uses.
1227 late = hoist_to_cheaper_block(LCA, early, self);
1228 } else {
1229 // Just use the LCA of the uses.
1230 late = LCA;
1231 }
1232
1233 // Put the node into target block
1234 schedule_node_into_block(self, late);
1235
1236 #ifdef ASSERT
1237 if (self->needs_anti_dependence_check()) {
1238 // since precedence edges are only inserted when we're sure they
1239 // are needed make sure that after placement in a block we don't
1240 // need any new precedence edges.
1241 verify_anti_dependences(late, self);
1242 }
1243 #endif
1244 } // Loop until all nodes have been visited
1245
1246 } // end ScheduleLate
1247
1248 //------------------------------GlobalCodeMotion-------------------------------
1249 void PhaseCFG::global_code_motion() {
1250 ResourceMark rm;
1251
1252 #ifndef PRODUCT
1253 if (trace_opto_pipelining()) {
1254 tty->print("\n---- Start GlobalCodeMotion ----\n");
1255 }
1256 #endif
1257
1258 // Initialize the node to block mapping for things on the proj_list
1259 for (uint i = 0; i < _matcher.number_of_projections(); i++) {
1260 unmap_node_from_block(_matcher.get_projection(i));
1261 }
1262
1263 // Set the basic block for Nodes pinned into blocks
1264 Arena* arena = Thread::current()->resource_area();
1265 VectorSet visited(arena);
1266 schedule_pinned_nodes(visited);
1267
1268 // Find the earliest Block any instruction can be placed in. Some
1269 // instructions are pinned into Blocks. Unpinned instructions can
1270 // appear in last block in which all their inputs occur.
1271 visited.Clear();
1272 Node_List stack(arena);
1273 // Pre-grow the list
1274 stack.map((C->unique() >> 1) + 16, NULL);
1275 if (!schedule_early(visited, stack)) {
1276 // Bailout without retry
1277 C->record_method_not_compilable("early schedule failed");
1278 return;
1279 }
1280
1281 // Build Def-Use edges.
1282 // Compute the latency information (via backwards walk) for all the
1283 // instructions in the graph
1284 _node_latency = new GrowableArray<uint>(); // resource_area allocation
1285
1286 if (C->do_scheduling()) {
1287 compute_latencies_backwards(visited, stack);
1288 }
1289
1290 // Now schedule all codes as LATE as possible. This is the LCA in the
1291 // dominator tree of all USES of a value. Pick the block with the least
1292 // loop nesting depth that is lowest in the dominator tree.
1293 // ( visited.Clear() called in schedule_late()->Node_Backward_Iterator() )
1294 schedule_late(visited, stack);
1295 if (C->failing()) {
1296 // schedule_late fails only when graph is incorrect.
1297 assert(!VerifyGraphEdges, "verification should have failed");
1298 return;
1299 }
1300
1301 #ifndef PRODUCT
1302 if (trace_opto_pipelining()) {
1303 tty->print("\n---- Detect implicit null checks ----\n");
1304 }
1305 #endif
1306
1307 // Detect implicit-null-check opportunities. Basically, find NULL checks
1308 // with suitable memory ops nearby. Use the memory op to do the NULL check.
1309 // I can generate a memory op if there is not one nearby.
1310 if (C->is_method_compilation()) {
1311 // By reversing the loop direction we get a very minor gain on mpegaudio.
1312 // Feel free to revert to a forward loop for clarity.
1313 // for( int i=0; i < (int)matcher._null_check_tests.size(); i+=2 ) {
1314 for (int i = _matcher._null_check_tests.size() - 2; i >= 0; i -= 2) {
1315 Node* proj = _matcher._null_check_tests[i];
1316 Node* val = _matcher._null_check_tests[i + 1];
1317 Block* block = get_block_for_node(proj);
1318 implicit_null_check(block, proj, val, C->allowed_deopt_reasons());
1319 // The implicit_null_check will only perform the transformation
1320 // if the null branch is truly uncommon, *and* it leads to an
1321 // uncommon trap. Combined with the too_many_traps guards
1322 // above, this prevents SEGV storms reported in 6366351,
1323 // by recompiling offending methods without this optimization.
1324 }
1325 }
1326
1327 #ifndef PRODUCT
1328 if (trace_opto_pipelining()) {
1329 tty->print("\n---- Start Local Scheduling ----\n");
1330 }
1331 #endif
1332
1333 // Schedule locally. Right now a simple topological sort.
1334 // Later, do a real latency aware scheduler.
1335 GrowableArray<int> ready_cnt(C->unique(), C->unique(), -1);
1336 visited.Clear();
1337 for (uint i = 0; i < number_of_blocks(); i++) {
1338 Block* block = get_block(i);
1339 if (!schedule_local(block, ready_cnt, visited)) {
1340 if (!C->failure_reason_is(C2Compiler::retry_no_subsuming_loads())) {
1341 C->record_method_not_compilable("local schedule failed");
1342 }
1343 return;
1344 }
1345 }
1346
1347 // If we inserted any instructions between a Call and his CatchNode,
1348 // clone the instructions on all paths below the Catch.
1349 for (uint i = 0; i < number_of_blocks(); i++) {
1350 Block* block = get_block(i);
1351 call_catch_cleanup(block);
1352 }
1353
1354 #ifndef PRODUCT
1355 if (trace_opto_pipelining()) {
1356 tty->print("\n---- After GlobalCodeMotion ----\n");
1357 for (uint i = 0; i < number_of_blocks(); i++) {
1358 Block* block = get_block(i);
1359 block->dump();
1360 }
1361 }
1362 #endif
1363 // Dead.
1364 _node_latency = (GrowableArray<uint> *)0xdeadbeef;
1365 }
1366
1367 bool PhaseCFG::do_global_code_motion() {
1368
1369 build_dominator_tree();
1370 if (C->failing()) {
1371 return false;
1372 }
1373
1374 NOT_PRODUCT( C->verify_graph_edges(); )
1375
1376 estimate_block_frequency();
1377
1378 global_code_motion();
1379
1380 if (C->failing()) {
1381 return false;
1382 }
1383
1384 return true;
1385 }
1386
1387 //------------------------------Estimate_Block_Frequency-----------------------
1388 // Estimate block frequencies based on IfNode probabilities.
1389 void PhaseCFG::estimate_block_frequency() {
1390
1391 // Force conditional branches leading to uncommon traps to be unlikely,
1392 // not because we get to the uncommon_trap with less relative frequency,
1393 // but because an uncommon_trap typically causes a deopt, so we only get
1394 // there once.
1395 if (C->do_freq_based_layout()) {
1396 Block_List worklist;
1397 Block* root_blk = get_block(0);
1398 for (uint i = 1; i < root_blk->num_preds(); i++) {
1399 Block *pb = get_block_for_node(root_blk->pred(i));
1400 if (pb->has_uncommon_code()) {
1401 worklist.push(pb);
1402 }
1403 }
1404 while (worklist.size() > 0) {
1405 Block* uct = worklist.pop();
1406 if (uct == get_root_block()) {
1407 continue;
1408 }
1409 for (uint i = 1; i < uct->num_preds(); i++) {
1410 Block *pb = get_block_for_node(uct->pred(i));
1411 if (pb->_num_succs == 1) {
1412 worklist.push(pb);
1413 } else if (pb->num_fall_throughs() == 2) {
1414 pb->update_uncommon_branch(uct);
1415 }
1416 }
1417 }
1418 }
1419
1420 // Create the loop tree and calculate loop depth.
1421 _root_loop = create_loop_tree();
1422 _root_loop->compute_loop_depth(0);
1423
1424 // Compute block frequency of each block, relative to a single loop entry.
1425 _root_loop->compute_freq();
1426
1427 // Adjust all frequencies to be relative to a single method entry
1428 _root_loop->_freq = 1.0;
1429 _root_loop->scale_freq();
1430
1431 // Save outmost loop frequency for LRG frequency threshold
1432 _outer_loop_frequency = _root_loop->outer_loop_freq();
1433
1434 // force paths ending at uncommon traps to be infrequent
1435 if (!C->do_freq_based_layout()) {
1436 Block_List worklist;
1437 Block* root_blk = get_block(0);
1438 for (uint i = 1; i < root_blk->num_preds(); i++) {
1439 Block *pb = get_block_for_node(root_blk->pred(i));
1440 if (pb->has_uncommon_code()) {
1441 worklist.push(pb);
1442 }
1443 }
1444 while (worklist.size() > 0) {
1445 Block* uct = worklist.pop();
1446 uct->_freq = PROB_MIN;
1447 for (uint i = 1; i < uct->num_preds(); i++) {
1448 Block *pb = get_block_for_node(uct->pred(i));
1449 if (pb->_num_succs == 1 && pb->_freq > PROB_MIN) {
1450 worklist.push(pb);
1451 }
1452 }
1453 }
1454 }
1455
1456 #ifdef ASSERT
1457 for (uint i = 0; i < number_of_blocks(); i++) {
1458 Block* b = get_block(i);
1459 assert(b->_freq >= MIN_BLOCK_FREQUENCY, "Register Allocator requires meaningful block frequency");
1460 }
1461 #endif
1462
1463 #ifndef PRODUCT
1464 if (PrintCFGBlockFreq) {
1465 tty->print_cr("CFG Block Frequencies");
1466 _root_loop->dump_tree();
1467 if (Verbose) {
1468 tty->print_cr("PhaseCFG dump");
1469 dump();
1470 tty->print_cr("Node dump");
1471 _root->dump(99999);
1472 }
1473 }
1474 #endif
1475 }
1476
1477 //----------------------------create_loop_tree--------------------------------
1478 // Create a loop tree from the CFG
1479 CFGLoop* PhaseCFG::create_loop_tree() {
1480
1481 #ifdef ASSERT
1482 assert(get_block(0) == get_root_block(), "first block should be root block");
1483 for (uint i = 0; i < number_of_blocks(); i++) {
1484 Block* block = get_block(i);
1485 // Check that _loop field are clear...we could clear them if not.
1486 assert(block->_loop == NULL, "clear _loop expected");
1487 // Sanity check that the RPO numbering is reflected in the _blocks array.
1488 // It doesn't have to be for the loop tree to be built, but if it is not,
1489 // then the blocks have been reordered since dom graph building...which
1490 // may question the RPO numbering
1491 assert(block->_rpo == i, "unexpected reverse post order number");
1492 }
1493 #endif
1494
1495 int idct = 0;
1496 CFGLoop* root_loop = new CFGLoop(idct++);
1497
1498 Block_List worklist;
1499
1500 // Assign blocks to loops
1501 for(uint i = number_of_blocks() - 1; i > 0; i-- ) { // skip Root block
1502 Block* block = get_block(i);
1503
1504 if (block->head()->is_Loop()) {
1505 Block* loop_head = block;
1506 assert(loop_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
1507 Node* tail_n = loop_head->pred(LoopNode::LoopBackControl);
1508 Block* tail = get_block_for_node(tail_n);
1509
1510 // Defensively filter out Loop nodes for non-single-entry loops.
1511 // For all reasonable loops, the head occurs before the tail in RPO.
1512 if (i <= tail->_rpo) {
1513
1514 // The tail and (recursive) predecessors of the tail
1515 // are made members of a new loop.
1516
1517 assert(worklist.size() == 0, "nonempty worklist");
1518 CFGLoop* nloop = new CFGLoop(idct++);
1519 assert(loop_head->_loop == NULL, "just checking");
1520 loop_head->_loop = nloop;
1521 // Add to nloop so push_pred() will skip over inner loops
1522 nloop->add_member(loop_head);
1523 nloop->push_pred(loop_head, LoopNode::LoopBackControl, worklist, this);
1524
1525 while (worklist.size() > 0) {
1526 Block* member = worklist.pop();
1527 if (member != loop_head) {
1528 for (uint j = 1; j < member->num_preds(); j++) {
1529 nloop->push_pred(member, j, worklist, this);
1530 }
1531 }
1532 }
1533 }
1534 }
1535 }
1536
1537 // Create a member list for each loop consisting
1538 // of both blocks and (immediate child) loops.
1539 for (uint i = 0; i < number_of_blocks(); i++) {
1540 Block* block = get_block(i);
1541 CFGLoop* lp = block->_loop;
1542 if (lp == NULL) {
1543 // Not assigned to a loop. Add it to the method's pseudo loop.
1544 block->_loop = root_loop;
1545 lp = root_loop;
1546 }
1547 if (lp == root_loop || block != lp->head()) { // loop heads are already members
1548 lp->add_member(block);
1549 }
1550 if (lp != root_loop) {
1551 if (lp->parent() == NULL) {
1552 // Not a nested loop. Make it a child of the method's pseudo loop.
1553 root_loop->add_nested_loop(lp);
1554 }
1555 if (block == lp->head()) {
1556 // Add nested loop to member list of parent loop.
1557 lp->parent()->add_member(lp);
1558 }
1559 }
1560 }
1561
1562 return root_loop;
1563 }
1564
1565 //------------------------------push_pred--------------------------------------
1566 void CFGLoop::push_pred(Block* blk, int i, Block_List& worklist, PhaseCFG* cfg) {
1567 Node* pred_n = blk->pred(i);
1568 Block* pred = cfg->get_block_for_node(pred_n);
1569 CFGLoop *pred_loop = pred->_loop;
1570 if (pred_loop == NULL) {
1571 // Filter out blocks for non-single-entry loops.
1572 // For all reasonable loops, the head occurs before the tail in RPO.
1573 if (pred->_rpo > head()->_rpo) {
1574 pred->_loop = this;
1575 worklist.push(pred);
1576 }
1577 } else if (pred_loop != this) {
1578 // Nested loop.
1579 while (pred_loop->_parent != NULL && pred_loop->_parent != this) {
1580 pred_loop = pred_loop->_parent;
1581 }
1582 // Make pred's loop be a child
1583 if (pred_loop->_parent == NULL) {
1584 add_nested_loop(pred_loop);
1585 // Continue with loop entry predecessor.
1586 Block* pred_head = pred_loop->head();
1587 assert(pred_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
1588 assert(pred_head != head(), "loop head in only one loop");
1589 push_pred(pred_head, LoopNode::EntryControl, worklist, cfg);
1590 } else {
1591 assert(pred_loop->_parent == this && _parent == NULL, "just checking");
1592 }
1593 }
1594 }
1595
1596 //------------------------------add_nested_loop--------------------------------
1597 // Make cl a child of the current loop in the loop tree.
1598 void CFGLoop::add_nested_loop(CFGLoop* cl) {
1599 assert(_parent == NULL, "no parent yet");
1600 assert(cl != this, "not my own parent");
1601 cl->_parent = this;
1602 CFGLoop* ch = _child;
1603 if (ch == NULL) {
1604 _child = cl;
1605 } else {
1606 while (ch->_sibling != NULL) { ch = ch->_sibling; }
1607 ch->_sibling = cl;
1608 }
1609 }
1610
1611 //------------------------------compute_loop_depth-----------------------------
1612 // Store the loop depth in each CFGLoop object.
1613 // Recursively walk the children to do the same for them.
1614 void CFGLoop::compute_loop_depth(int depth) {
1615 _depth = depth;
1616 CFGLoop* ch = _child;
1617 while (ch != NULL) {
1618 ch->compute_loop_depth(depth + 1);
1619 ch = ch->_sibling;
1620 }
1621 }
1622
1623 //------------------------------compute_freq-----------------------------------
1624 // Compute the frequency of each block and loop, relative to a single entry
1625 // into the dominating loop head.
1626 void CFGLoop::compute_freq() {
1627 // Bottom up traversal of loop tree (visit inner loops first.)
1628 // Set loop head frequency to 1.0, then transitively
1629 // compute frequency for all successors in the loop,
1630 // as well as for each exit edge. Inner loops are
1631 // treated as single blocks with loop exit targets
1632 // as the successor blocks.
1633
1634 // Nested loops first
1635 CFGLoop* ch = _child;
1636 while (ch != NULL) {
1637 ch->compute_freq();
1638 ch = ch->_sibling;
1639 }
1640 assert (_members.length() > 0, "no empty loops");
1641 Block* hd = head();
1642 hd->_freq = 1.0;
1643 for (int i = 0; i < _members.length(); i++) {
1644 CFGElement* s = _members.at(i);
1645 double freq = s->_freq;
1646 if (s->is_block()) {
1647 Block* b = s->as_Block();
1648 for (uint j = 0; j < b->_num_succs; j++) {
1649 Block* sb = b->_succs[j];
1650 update_succ_freq(sb, freq * b->succ_prob(j));
1651 }
1652 } else {
1653 CFGLoop* lp = s->as_CFGLoop();
1654 assert(lp->_parent == this, "immediate child");
1655 for (int k = 0; k < lp->_exits.length(); k++) {
1656 Block* eb = lp->_exits.at(k).get_target();
1657 double prob = lp->_exits.at(k).get_prob();
1658 update_succ_freq(eb, freq * prob);
1659 }
1660 }
1661 }
1662
1663 // For all loops other than the outer, "method" loop,
1664 // sum and normalize the exit probability. The "method" loop
1665 // should keep the initial exit probability of 1, so that
1666 // inner blocks do not get erroneously scaled.
1667 if (_depth != 0) {
1668 // Total the exit probabilities for this loop.
1669 double exits_sum = 0.0f;
1670 for (int i = 0; i < _exits.length(); i++) {
1671 exits_sum += _exits.at(i).get_prob();
1672 }
1673
1674 // Normalize the exit probabilities. Until now, the
1675 // probabilities estimate the possibility of exit per
1676 // a single loop iteration; afterward, they estimate
1677 // the probability of exit per loop entry.
1678 for (int i = 0; i < _exits.length(); i++) {
1679 Block* et = _exits.at(i).get_target();
1680 float new_prob = 0.0f;
1681 if (_exits.at(i).get_prob() > 0.0f) {
1682 new_prob = _exits.at(i).get_prob() / exits_sum;
1683 }
1684 BlockProbPair bpp(et, new_prob);
1685 _exits.at_put(i, bpp);
1686 }
1687
1688 // Save the total, but guard against unreasonable probability,
1689 // as the value is used to estimate the loop trip count.
1690 // An infinite trip count would blur relative block
1691 // frequencies.
1692 if (exits_sum > 1.0f) exits_sum = 1.0;
1693 if (exits_sum < PROB_MIN) exits_sum = PROB_MIN;
1694 _exit_prob = exits_sum;
1695 }
1696 }
1697
1698 //------------------------------succ_prob-------------------------------------
1699 // Determine the probability of reaching successor 'i' from the receiver block.
1700 float Block::succ_prob(uint i) {
1701 int eidx = end_idx();
1702 Node *n = get_node(eidx); // Get ending Node
1703
1704 int op = n->Opcode();
1705 if (n->is_Mach()) {
1706 if (n->is_MachNullCheck()) {
1707 // Can only reach here if called after lcm. The original Op_If is gone,
1708 // so we attempt to infer the probability from one or both of the
1709 // successor blocks.
1710 assert(_num_succs == 2, "expecting 2 successors of a null check");
1711 // If either successor has only one predecessor, then the
1712 // probability estimate can be derived using the
1713 // relative frequency of the successor and this block.
1714 if (_succs[i]->num_preds() == 2) {
1715 return _succs[i]->_freq / _freq;
1716 } else if (_succs[1-i]->num_preds() == 2) {
1717 return 1 - (_succs[1-i]->_freq / _freq);
1718 } else {
1719 // Estimate using both successor frequencies
1720 float freq = _succs[i]->_freq;
1721 return freq / (freq + _succs[1-i]->_freq);
1722 }
1723 }
1724 op = n->as_Mach()->ideal_Opcode();
1725 }
1726
1727
1728 // Switch on branch type
1729 switch( op ) {
1730 case Op_CountedLoopEnd:
1731 case Op_If: {
1732 assert (i < 2, "just checking");
1733 // Conditionals pass on only part of their frequency
1734 float prob = n->as_MachIf()->_prob;
1735 assert(prob >= 0.0 && prob <= 1.0, "out of range probability");
1736 // If succ[i] is the FALSE branch, invert path info
1737 if( get_node(i + eidx + 1)->Opcode() == Op_IfFalse ) {
1738 return 1.0f - prob; // not taken
1739 } else {
1740 return prob; // taken
1741 }
1742 }
1743
1744 case Op_Jump:
1745 // Divide the frequency between all successors evenly
1746 return 1.0f/_num_succs;
1747
1748 case Op_Catch: {
1749 const CatchProjNode *ci = get_node(i + eidx + 1)->as_CatchProj();
1750 if (ci->_con == CatchProjNode::fall_through_index) {
1751 // Fall-thru path gets the lion's share.
1752 return 1.0f - PROB_UNLIKELY_MAG(5)*_num_succs;
1753 } else {
1754 // Presume exceptional paths are equally unlikely
1755 return PROB_UNLIKELY_MAG(5);
1756 }
1757 }
1758
1759 case Op_Root:
1760 case Op_Goto:
1761 // Pass frequency straight thru to target
1762 return 1.0f;
1763
1764 case Op_NeverBranch:
1765 return 0.0f;
1766
1767 case Op_TailCall:
1768 case Op_TailJump:
1769 case Op_Return:
1770 case Op_Halt:
1771 case Op_Rethrow:
1772 // Do not push out freq to root block
1773 return 0.0f;
1774
1775 default:
1776 ShouldNotReachHere();
1777 }
1778
1779 return 0.0f;
1780 }
1781
1782 //------------------------------num_fall_throughs-----------------------------
1783 // Return the number of fall-through candidates for a block
1784 int Block::num_fall_throughs() {
1785 int eidx = end_idx();
1786 Node *n = get_node(eidx); // Get ending Node
1787
1788 int op = n->Opcode();
1789 if (n->is_Mach()) {
1790 if (n->is_MachNullCheck()) {
1791 // In theory, either side can fall-thru, for simplicity sake,
1792 // let's say only the false branch can now.
1793 return 1;
1794 }
1795 op = n->as_Mach()->ideal_Opcode();
1796 }
1797
1798 // Switch on branch type
1799 switch( op ) {
1800 case Op_CountedLoopEnd:
1801 case Op_If:
1802 return 2;
1803
1804 case Op_Root:
1805 case Op_Goto:
1806 return 1;
1807
1808 case Op_Catch: {
1809 for (uint i = 0; i < _num_succs; i++) {
1810 const CatchProjNode *ci = get_node(i + eidx + 1)->as_CatchProj();
1811 if (ci->_con == CatchProjNode::fall_through_index) {
1812 return 1;
1813 }
1814 }
1815 return 0;
1816 }
1817
1818 case Op_Jump:
1819 case Op_NeverBranch:
1820 case Op_TailCall:
1821 case Op_TailJump:
1822 case Op_Return:
1823 case Op_Halt:
1824 case Op_Rethrow:
1825 return 0;
1826
1827 default:
1828 ShouldNotReachHere();
1829 }
1830
1831 return 0;
1832 }
1833
1834 //------------------------------succ_fall_through-----------------------------
1835 // Return true if a specific successor could be fall-through target.
1836 bool Block::succ_fall_through(uint i) {
1837 int eidx = end_idx();
1838 Node *n = get_node(eidx); // Get ending Node
1839
1840 int op = n->Opcode();
1841 if (n->is_Mach()) {
1842 if (n->is_MachNullCheck()) {
1843 // In theory, either side can fall-thru, for simplicity sake,
1844 // let's say only the false branch can now.
1845 return get_node(i + eidx + 1)->Opcode() == Op_IfFalse;
1846 }
1847 op = n->as_Mach()->ideal_Opcode();
1848 }
1849
1850 // Switch on branch type
1851 switch( op ) {
1852 case Op_CountedLoopEnd:
1853 case Op_If:
1854 case Op_Root:
1855 case Op_Goto:
1856 return true;
1857
1858 case Op_Catch: {
1859 const CatchProjNode *ci = get_node(i + eidx + 1)->as_CatchProj();
1860 return ci->_con == CatchProjNode::fall_through_index;
1861 }
1862
1863 case Op_Jump:
1864 case Op_NeverBranch:
1865 case Op_TailCall:
1866 case Op_TailJump:
1867 case Op_Return:
1868 case Op_Halt:
1869 case Op_Rethrow:
1870 return false;
1871
1872 default:
1873 ShouldNotReachHere();
1874 }
1875
1876 return false;
1877 }
1878
1879 //------------------------------update_uncommon_branch------------------------
1880 // Update the probability of a two-branch to be uncommon
1881 void Block::update_uncommon_branch(Block* ub) {
1882 int eidx = end_idx();
1883 Node *n = get_node(eidx); // Get ending Node
1884
1885 int op = n->as_Mach()->ideal_Opcode();
1886
1887 assert(op == Op_CountedLoopEnd || op == Op_If, "must be a If");
1888 assert(num_fall_throughs() == 2, "must be a two way branch block");
1889
1890 // Which successor is ub?
1891 uint s;
1892 for (s = 0; s <_num_succs; s++) {
1893 if (_succs[s] == ub) break;
1894 }
1895 assert(s < 2, "uncommon successor must be found");
1896
1897 // If ub is the true path, make the proability small, else
1898 // ub is the false path, and make the probability large
1899 bool invert = (get_node(s + eidx + 1)->Opcode() == Op_IfFalse);
1900
1901 // Get existing probability
1902 float p = n->as_MachIf()->_prob;
1903
1904 if (invert) p = 1.0 - p;
1905 if (p > PROB_MIN) {
1906 p = PROB_MIN;
1907 }
1908 if (invert) p = 1.0 - p;
1909
1910 n->as_MachIf()->_prob = p;
1911 }
1912
1913 //------------------------------update_succ_freq-------------------------------
1914 // Update the appropriate frequency associated with block 'b', a successor of
1915 // a block in this loop.
1916 void CFGLoop::update_succ_freq(Block* b, double freq) {
1917 if (b->_loop == this) {
1918 if (b == head()) {
1919 // back branch within the loop
1920 // Do nothing now, the loop carried frequency will be
1921 // adjust later in scale_freq().
1922 } else {
1923 // simple branch within the loop
1924 b->_freq += freq;
1925 }
1926 } else if (!in_loop_nest(b)) {
1927 // branch is exit from this loop
1928 BlockProbPair bpp(b, freq);
1929 _exits.append(bpp);
1930 } else {
1931 // branch into nested loop
1932 CFGLoop* ch = b->_loop;
1933 ch->_freq += freq;
1934 }
1935 }
1936
1937 //------------------------------in_loop_nest-----------------------------------
1938 // Determine if block b is in the receiver's loop nest.
1939 bool CFGLoop::in_loop_nest(Block* b) {
1940 int depth = _depth;
1941 CFGLoop* b_loop = b->_loop;
1942 int b_depth = b_loop->_depth;
1943 if (depth == b_depth) {
1944 return true;
1945 }
1946 while (b_depth > depth) {
1947 b_loop = b_loop->_parent;
1948 b_depth = b_loop->_depth;
1949 }
1950 return b_loop == this;
1951 }
1952
1953 //------------------------------scale_freq-------------------------------------
1954 // Scale frequency of loops and blocks by trip counts from outer loops
1955 // Do a top down traversal of loop tree (visit outer loops first.)
1956 void CFGLoop::scale_freq() {
1957 double loop_freq = _freq * trip_count();
1958 _freq = loop_freq;
1959 for (int i = 0; i < _members.length(); i++) {
1960 CFGElement* s = _members.at(i);
1961 double block_freq = s->_freq * loop_freq;
1962 if (g_isnan(block_freq) || block_freq < MIN_BLOCK_FREQUENCY)
1963 block_freq = MIN_BLOCK_FREQUENCY;
1964 s->_freq = block_freq;
1965 }
1966 CFGLoop* ch = _child;
1967 while (ch != NULL) {
1968 ch->scale_freq();
1969 ch = ch->_sibling;
1970 }
1971 }
1972
1973 // Frequency of outer loop
1974 double CFGLoop::outer_loop_freq() const {
1975 if (_child != NULL) {
1976 return _child->_freq;
1977 }
1978 return _freq;
1979 }
1980
1981 #ifndef PRODUCT
1982 //------------------------------dump_tree--------------------------------------
1983 void CFGLoop::dump_tree() const {
1984 dump();
1985 if (_child != NULL) _child->dump_tree();
1986 if (_sibling != NULL) _sibling->dump_tree();
1987 }
1988
1989 //------------------------------dump-------------------------------------------
1990 void CFGLoop::dump() const {
1991 for (int i = 0; i < _depth; i++) tty->print(" ");
1992 tty->print("%s: %d trip_count: %6.0f freq: %6.0f\n",
1993 _depth == 0 ? "Method" : "Loop", _id, trip_count(), _freq);
1994 for (int i = 0; i < _depth; i++) tty->print(" ");
1995 tty->print(" members:");
1996 int k = 0;
1997 for (int i = 0; i < _members.length(); i++) {
1998 if (k++ >= 6) {
1999 tty->print("\n ");
2000 for (int j = 0; j < _depth+1; j++) tty->print(" ");
2001 k = 0;
2002 }
2003 CFGElement *s = _members.at(i);
2004 if (s->is_block()) {
2005 Block *b = s->as_Block();
2006 tty->print(" B%d(%6.3f)", b->_pre_order, b->_freq);
2007 } else {
2008 CFGLoop* lp = s->as_CFGLoop();
2009 tty->print(" L%d(%6.3f)", lp->_id, lp->_freq);
2010 }
2011 }
2012 tty->print("\n");
2013 for (int i = 0; i < _depth; i++) tty->print(" ");
2014 tty->print(" exits: ");
2015 k = 0;
2016 for (int i = 0; i < _exits.length(); i++) {
2017 if (k++ >= 7) {
2018 tty->print("\n ");
2019 for (int j = 0; j < _depth+1; j++) tty->print(" ");
2020 k = 0;
2021 }
2022 Block *blk = _exits.at(i).get_target();
2023 double prob = _exits.at(i).get_prob();
2024 tty->print(" ->%d@%d%%", blk->_pre_order, (int)(prob*100));
2025 }
2026 tty->print("\n");
2027 }
2028 #endif