Fri, 16 Aug 2013 10:23:55 +0200
8023003: Cleanup the public interface to PhaseCFG
Summary: public methods that don't need to be public should be private.
Reviewed-by: kvn, twisti
1 /*
2 * Copyright (c) 1997, 2012, 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 */
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 #ifdef TARGET_ARCH_MODEL_x86_32
39 # include "adfiles/ad_x86_32.hpp"
40 #endif
41 #ifdef TARGET_ARCH_MODEL_x86_64
42 # include "adfiles/ad_x86_64.hpp"
43 #endif
44 #ifdef TARGET_ARCH_MODEL_sparc
45 # include "adfiles/ad_sparc.hpp"
46 #endif
47 #ifdef TARGET_ARCH_MODEL_zero
48 # include "adfiles/ad_zero.hpp"
49 #endif
50 #ifdef TARGET_ARCH_MODEL_arm
51 # include "adfiles/ad_arm.hpp"
52 #endif
53 #ifdef TARGET_ARCH_MODEL_ppc
54 # include "adfiles/ad_ppc.hpp"
55 #endif
57 // Portions of code courtesy of Clifford Click
59 // Optimization - Graph Style
61 // To avoid float value underflow
62 #define MIN_BLOCK_FREQUENCY 1.e-35f
64 //----------------------------schedule_node_into_block-------------------------
65 // Insert node n into block b. Look for projections of n and make sure they
66 // are in b also.
67 void PhaseCFG::schedule_node_into_block( Node *n, Block *b ) {
68 // Set basic block of n, Add n to b,
69 map_node_to_block(n, b);
70 b->add_inst(n);
72 // After Matching, nearly any old Node may have projections trailing it.
73 // These are usually machine-dependent flags. In any case, they might
74 // float to another block below this one. Move them up.
75 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
76 Node* use = n->fast_out(i);
77 if (use->is_Proj()) {
78 Block* buse = get_block_for_node(use);
79 if (buse != b) { // In wrong block?
80 if (buse != NULL) {
81 buse->find_remove(use); // Remove from wrong block
82 }
83 map_node_to_block(use, b);
84 b->add_inst(use);
85 }
86 }
87 }
88 }
90 //----------------------------replace_block_proj_ctrl-------------------------
91 // Nodes that have is_block_proj() nodes as their control need to use
92 // the appropriate Region for their actual block as their control since
93 // the projection will be in a predecessor block.
94 void PhaseCFG::replace_block_proj_ctrl( Node *n ) {
95 const Node *in0 = n->in(0);
96 assert(in0 != NULL, "Only control-dependent");
97 const Node *p = in0->is_block_proj();
98 if (p != NULL && p != n) { // Control from a block projection?
99 assert(!n->pinned() || n->is_MachConstantBase(), "only pinned MachConstantBase node is expected here");
100 // Find trailing Region
101 Block *pb = get_block_for_node(in0); // Block-projection already has basic block
102 uint j = 0;
103 if (pb->_num_succs != 1) { // More then 1 successor?
104 // Search for successor
105 uint max = pb->_nodes.size();
106 assert( max > 1, "" );
107 uint start = max - pb->_num_succs;
108 // Find which output path belongs to projection
109 for (j = start; j < max; j++) {
110 if( pb->_nodes[j] == in0 )
111 break;
112 }
113 assert( j < max, "must find" );
114 // Change control to match head of successor basic block
115 j -= start;
116 }
117 n->set_req(0, pb->_succs[j]->head());
118 }
119 }
122 //------------------------------schedule_pinned_nodes--------------------------
123 // Set the basic block for Nodes pinned into blocks
124 void PhaseCFG::schedule_pinned_nodes(VectorSet &visited) {
125 // Allocate node stack of size C->unique()+8 to avoid frequent realloc
126 GrowableArray <Node *> spstack(C->unique() + 8);
127 spstack.push(_root);
128 while (spstack.is_nonempty()) {
129 Node* node = spstack.pop();
130 if (!visited.test_set(node->_idx)) { // Test node and flag it as visited
131 if (node->pinned() && !has_block(node)) { // Pinned? Nail it down!
132 assert(node->in(0), "pinned Node must have Control");
133 // Before setting block replace block_proj control edge
134 replace_block_proj_ctrl(node);
135 Node* input = node->in(0);
136 while (!input->is_block_start()) {
137 input = input->in(0);
138 }
139 Block* block = get_block_for_node(input); // Basic block of controlling input
140 schedule_node_into_block(node, block);
141 }
143 // process all inputs that are non NULL
144 for (int i = node->req() - 1; i >= 0; --i) {
145 if (node->in(i) != NULL) {
146 spstack.push(node->in(i));
147 }
148 }
149 }
150 }
151 }
153 #ifdef ASSERT
154 // Assert that new input b2 is dominated by all previous inputs.
155 // Check this by by seeing that it is dominated by b1, the deepest
156 // input observed until b2.
157 static void assert_dom(Block* b1, Block* b2, Node* n, const PhaseCFG* cfg) {
158 if (b1 == NULL) return;
159 assert(b1->_dom_depth < b2->_dom_depth, "sanity");
160 Block* tmp = b2;
161 while (tmp != b1 && tmp != NULL) {
162 tmp = tmp->_idom;
163 }
164 if (tmp != b1) {
165 // Detected an unschedulable graph. Print some nice stuff and die.
166 tty->print_cr("!!! Unschedulable graph !!!");
167 for (uint j=0; j<n->len(); j++) { // For all inputs
168 Node* inn = n->in(j); // Get input
169 if (inn == NULL) continue; // Ignore NULL, missing inputs
170 Block* inb = cfg->get_block_for_node(inn);
171 tty->print("B%d idom=B%d depth=%2d ",inb->_pre_order,
172 inb->_idom ? inb->_idom->_pre_order : 0, inb->_dom_depth);
173 inn->dump();
174 }
175 tty->print("Failing node: ");
176 n->dump();
177 assert(false, "unscheduable graph");
178 }
179 }
180 #endif
182 static Block* find_deepest_input(Node* n, const PhaseCFG* cfg) {
183 // Find the last input dominated by all other inputs.
184 Block* deepb = NULL; // Deepest block so far
185 int deepb_dom_depth = 0;
186 for (uint k = 0; k < n->len(); k++) { // For all inputs
187 Node* inn = n->in(k); // Get input
188 if (inn == NULL) continue; // Ignore NULL, missing inputs
189 Block* inb = cfg->get_block_for_node(inn);
190 assert(inb != NULL, "must already have scheduled this input");
191 if (deepb_dom_depth < (int) inb->_dom_depth) {
192 // The new inb must be dominated by the previous deepb.
193 // The various inputs must be linearly ordered in the dom
194 // tree, or else there will not be a unique deepest block.
195 DEBUG_ONLY(assert_dom(deepb, inb, n, cfg));
196 deepb = inb; // Save deepest block
197 deepb_dom_depth = deepb->_dom_depth;
198 }
199 }
200 assert(deepb != NULL, "must be at least one input to n");
201 return deepb;
202 }
205 //------------------------------schedule_early---------------------------------
206 // Find the earliest Block any instruction can be placed in. Some instructions
207 // are pinned into Blocks. Unpinned instructions can appear in last block in
208 // which all their inputs occur.
209 bool PhaseCFG::schedule_early(VectorSet &visited, Node_List &roots) {
210 // Allocate stack with enough space to avoid frequent realloc
211 Node_Stack nstack(roots.Size() + 8);
212 // _root will be processed among C->top() inputs
213 roots.push(C->top());
214 visited.set(C->top()->_idx);
216 while (roots.size() != 0) {
217 // Use local variables nstack_top_n & nstack_top_i to cache values
218 // on stack's top.
219 Node* parent_node = roots.pop();
220 uint input_index = 0;
222 while (true) {
223 if (input_index == 0) {
224 // Fixup some control. Constants without control get attached
225 // to root and nodes that use is_block_proj() nodes should be attached
226 // to the region that starts their block.
227 const Node* control_input = parent_node->in(0);
228 if (control_input != NULL) {
229 replace_block_proj_ctrl(parent_node);
230 } else {
231 // Is a constant with NO inputs?
232 if (parent_node->req() == 1) {
233 parent_node->set_req(0, _root);
234 }
235 }
236 }
238 // First, visit all inputs and force them to get a block. If an
239 // input is already in a block we quit following inputs (to avoid
240 // cycles). Instead we put that Node on a worklist to be handled
241 // later (since IT'S inputs may not have a block yet).
243 // Assume all n's inputs will be processed
244 bool done = true;
246 while (input_index < parent_node->len()) {
247 Node* in = parent_node->in(input_index++);
248 if (in == NULL) {
249 continue;
250 }
252 int is_visited = visited.test_set(in->_idx);
253 if (!has_block(in)) {
254 if (is_visited) {
255 return false;
256 }
257 // Save parent node and next input's index.
258 nstack.push(parent_node, input_index);
259 // Process current input now.
260 parent_node = in;
261 input_index = 0;
262 // Not all n's inputs processed.
263 done = false;
264 break;
265 } else if (!is_visited) {
266 // Visit this guy later, using worklist
267 roots.push(in);
268 }
269 }
271 if (done) {
272 // All of n's inputs have been processed, complete post-processing.
274 // Some instructions are pinned into a block. These include Region,
275 // Phi, Start, Return, and other control-dependent instructions and
276 // any projections which depend on them.
277 if (!parent_node->pinned()) {
278 // Set earliest legal block.
279 Block* earliest_block = find_deepest_input(parent_node, this);
280 map_node_to_block(parent_node, earliest_block);
281 } else {
282 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");
283 }
285 if (nstack.is_empty()) {
286 // Finished all nodes on stack.
287 // Process next node on the worklist 'roots'.
288 break;
289 }
290 // Get saved parent node and next input's index.
291 parent_node = nstack.node();
292 input_index = nstack.index();
293 nstack.pop();
294 }
295 }
296 }
297 return true;
298 }
300 //------------------------------dom_lca----------------------------------------
301 // Find least common ancestor in dominator tree
302 // LCA is a current notion of LCA, to be raised above 'this'.
303 // As a convenient boundary condition, return 'this' if LCA is NULL.
304 // Find the LCA of those two nodes.
305 Block* Block::dom_lca(Block* LCA) {
306 if (LCA == NULL || LCA == this) return this;
308 Block* anc = this;
309 while (anc->_dom_depth > LCA->_dom_depth)
310 anc = anc->_idom; // Walk up till anc is as high as LCA
312 while (LCA->_dom_depth > anc->_dom_depth)
313 LCA = LCA->_idom; // Walk up till LCA is as high as anc
315 while (LCA != anc) { // Walk both up till they are the same
316 LCA = LCA->_idom;
317 anc = anc->_idom;
318 }
320 return LCA;
321 }
323 //--------------------------raise_LCA_above_use--------------------------------
324 // We are placing a definition, and have been given a def->use edge.
325 // The definition must dominate the use, so move the LCA upward in the
326 // dominator tree to dominate the use. If the use is a phi, adjust
327 // the LCA only with the phi input paths which actually use this def.
328 static Block* raise_LCA_above_use(Block* LCA, Node* use, Node* def, const PhaseCFG* cfg) {
329 Block* buse = cfg->get_block_for_node(use);
330 if (buse == NULL) return LCA; // Unused killing Projs have no use block
331 if (!use->is_Phi()) return buse->dom_lca(LCA);
332 uint pmax = use->req(); // Number of Phi inputs
333 // Why does not this loop just break after finding the matching input to
334 // the Phi? Well...it's like this. I do not have true def-use/use-def
335 // chains. Means I cannot distinguish, from the def-use direction, which
336 // of many use-defs lead from the same use to the same def. That is, this
337 // Phi might have several uses of the same def. Each use appears in a
338 // different predecessor block. But when I enter here, I cannot distinguish
339 // which use-def edge I should find the predecessor block for. So I find
340 // them all. Means I do a little extra work if a Phi uses the same value
341 // more than once.
342 for (uint j=1; j<pmax; j++) { // For all inputs
343 if (use->in(j) == def) { // Found matching input?
344 Block* pred = cfg->get_block_for_node(buse->pred(j));
345 LCA = pred->dom_lca(LCA);
346 }
347 }
348 return LCA;
349 }
351 //----------------------------raise_LCA_above_marks----------------------------
352 // Return a new LCA that dominates LCA and any of its marked predecessors.
353 // Search all my parents up to 'early' (exclusive), looking for predecessors
354 // which are marked with the given index. Return the LCA (in the dom tree)
355 // of all marked blocks. If there are none marked, return the original
356 // LCA.
357 static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark, Block* early, const PhaseCFG* cfg) {
358 Block_List worklist;
359 worklist.push(LCA);
360 while (worklist.size() > 0) {
361 Block* mid = worklist.pop();
362 if (mid == early) continue; // stop searching here
364 // Test and set the visited bit.
365 if (mid->raise_LCA_visited() == mark) continue; // already visited
367 // Don't process the current LCA, otherwise the search may terminate early
368 if (mid != LCA && mid->raise_LCA_mark() == mark) {
369 // Raise the LCA.
370 LCA = mid->dom_lca(LCA);
371 if (LCA == early) break; // stop searching everywhere
372 assert(early->dominates(LCA), "early is high enough");
373 // Resume searching at that point, skipping intermediate levels.
374 worklist.push(LCA);
375 if (LCA == mid)
376 continue; // Don't mark as visited to avoid early termination.
377 } else {
378 // Keep searching through this block's predecessors.
379 for (uint j = 1, jmax = mid->num_preds(); j < jmax; j++) {
380 Block* mid_parent = cfg->get_block_for_node(mid->pred(j));
381 worklist.push(mid_parent);
382 }
383 }
384 mid->set_raise_LCA_visited(mark);
385 }
386 return LCA;
387 }
389 //--------------------------memory_early_block--------------------------------
390 // This is a variation of find_deepest_input, the heart of schedule_early.
391 // Find the "early" block for a load, if we considered only memory and
392 // address inputs, that is, if other data inputs were ignored.
393 //
394 // Because a subset of edges are considered, the resulting block will
395 // be earlier (at a shallower dom_depth) than the true schedule_early
396 // point of the node. We compute this earlier block as a more permissive
397 // site for anti-dependency insertion, but only if subsume_loads is enabled.
398 static Block* memory_early_block(Node* load, Block* early, const PhaseCFG* cfg) {
399 Node* base;
400 Node* index;
401 Node* store = load->in(MemNode::Memory);
402 load->as_Mach()->memory_inputs(base, index);
404 assert(base != NodeSentinel && index != NodeSentinel,
405 "unexpected base/index inputs");
407 Node* mem_inputs[4];
408 int mem_inputs_length = 0;
409 if (base != NULL) mem_inputs[mem_inputs_length++] = base;
410 if (index != NULL) mem_inputs[mem_inputs_length++] = index;
411 if (store != NULL) mem_inputs[mem_inputs_length++] = store;
413 // In the comparision below, add one to account for the control input,
414 // which may be null, but always takes up a spot in the in array.
415 if (mem_inputs_length + 1 < (int) load->req()) {
416 // This "load" has more inputs than just the memory, base and index inputs.
417 // For purposes of checking anti-dependences, we need to start
418 // from the early block of only the address portion of the instruction,
419 // and ignore other blocks that may have factored into the wider
420 // schedule_early calculation.
421 if (load->in(0) != NULL) mem_inputs[mem_inputs_length++] = load->in(0);
423 Block* deepb = NULL; // Deepest block so far
424 int deepb_dom_depth = 0;
425 for (int i = 0; i < mem_inputs_length; i++) {
426 Block* inb = cfg->get_block_for_node(mem_inputs[i]);
427 if (deepb_dom_depth < (int) inb->_dom_depth) {
428 // The new inb must be dominated by the previous deepb.
429 // The various inputs must be linearly ordered in the dom
430 // tree, or else there will not be a unique deepest block.
431 DEBUG_ONLY(assert_dom(deepb, inb, load, cfg));
432 deepb = inb; // Save deepest block
433 deepb_dom_depth = deepb->_dom_depth;
434 }
435 }
436 early = deepb;
437 }
439 return early;
440 }
442 //--------------------------insert_anti_dependences---------------------------
443 // A load may need to witness memory that nearby stores can overwrite.
444 // For each nearby store, either insert an "anti-dependence" edge
445 // from the load to the store, or else move LCA upward to force the
446 // load to (eventually) be scheduled in a block above the store.
447 //
448 // Do not add edges to stores on distinct control-flow paths;
449 // only add edges to stores which might interfere.
450 //
451 // Return the (updated) LCA. There will not be any possibly interfering
452 // store between the load's "early block" and the updated LCA.
453 // Any stores in the updated LCA will have new precedence edges
454 // back to the load. The caller is expected to schedule the load
455 // in the LCA, in which case the precedence edges will make LCM
456 // preserve anti-dependences. The caller may also hoist the load
457 // above the LCA, if it is not the early block.
458 Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
459 assert(load->needs_anti_dependence_check(), "must be a load of some sort");
460 assert(LCA != NULL, "");
461 DEBUG_ONLY(Block* LCA_orig = LCA);
463 // Compute the alias index. Loads and stores with different alias indices
464 // do not need anti-dependence edges.
465 uint load_alias_idx = C->get_alias_index(load->adr_type());
466 #ifdef ASSERT
467 if (load_alias_idx == Compile::AliasIdxBot && C->AliasLevel() > 0 &&
468 (PrintOpto || VerifyAliases ||
469 PrintMiscellaneous && (WizardMode || Verbose))) {
470 // Load nodes should not consume all of memory.
471 // Reporting a bottom type indicates a bug in adlc.
472 // If some particular type of node validly consumes all of memory,
473 // sharpen the preceding "if" to exclude it, so we can catch bugs here.
474 tty->print_cr("*** Possible Anti-Dependence Bug: Load consumes all of memory.");
475 load->dump(2);
476 if (VerifyAliases) assert(load_alias_idx != Compile::AliasIdxBot, "");
477 }
478 #endif
479 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrComp),
480 "String compare is only known 'load' that does not conflict with any stores");
481 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrEquals),
482 "String equals is a 'load' that does not conflict with any stores");
483 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrIndexOf),
484 "String indexOf is a 'load' that does not conflict with any stores");
485 assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_AryEq),
486 "Arrays equals is a 'load' that do not conflict with any stores");
488 if (!C->alias_type(load_alias_idx)->is_rewritable()) {
489 // It is impossible to spoil this load by putting stores before it,
490 // because we know that the stores will never update the value
491 // which 'load' must witness.
492 return LCA;
493 }
495 node_idx_t load_index = load->_idx;
497 // Note the earliest legal placement of 'load', as determined by
498 // by the unique point in the dom tree where all memory effects
499 // and other inputs are first available. (Computed by schedule_early.)
500 // For normal loads, 'early' is the shallowest place (dom graph wise)
501 // to look for anti-deps between this load and any store.
502 Block* early = get_block_for_node(load);
504 // If we are subsuming loads, compute an "early" block that only considers
505 // memory or address inputs. This block may be different than the
506 // schedule_early block in that it could be at an even shallower depth in the
507 // dominator tree, and allow for a broader discovery of anti-dependences.
508 if (C->subsume_loads()) {
509 early = memory_early_block(load, early, this);
510 }
512 ResourceArea *area = Thread::current()->resource_area();
513 Node_List worklist_mem(area); // prior memory state to store
514 Node_List worklist_store(area); // possible-def to explore
515 Node_List worklist_visited(area); // visited mergemem nodes
516 Node_List non_early_stores(area); // all relevant stores outside of early
517 bool must_raise_LCA = false;
519 #ifdef TRACK_PHI_INPUTS
520 // %%% This extra checking fails because MergeMem nodes are not GVNed.
521 // Provide "phi_inputs" to check if every input to a PhiNode is from the
522 // original memory state. This indicates a PhiNode for which should not
523 // prevent the load from sinking. For such a block, set_raise_LCA_mark
524 // may be overly conservative.
525 // Mechanism: count inputs seen for each Phi encountered in worklist_store.
526 DEBUG_ONLY(GrowableArray<uint> phi_inputs(area, C->unique(),0,0));
527 #endif
529 // 'load' uses some memory state; look for users of the same state.
530 // Recurse through MergeMem nodes to the stores that use them.
532 // Each of these stores is a possible definition of memory
533 // that 'load' needs to use. We need to force 'load'
534 // to occur before each such store. When the store is in
535 // the same block as 'load', we insert an anti-dependence
536 // edge load->store.
538 // The relevant stores "nearby" the load consist of a tree rooted
539 // at initial_mem, with internal nodes of type MergeMem.
540 // Therefore, the branches visited by the worklist are of this form:
541 // initial_mem -> (MergeMem ->)* store
542 // The anti-dependence constraints apply only to the fringe of this tree.
544 Node* initial_mem = load->in(MemNode::Memory);
545 worklist_store.push(initial_mem);
546 worklist_visited.push(initial_mem);
547 worklist_mem.push(NULL);
548 while (worklist_store.size() > 0) {
549 // Examine a nearby store to see if it might interfere with our load.
550 Node* mem = worklist_mem.pop();
551 Node* store = worklist_store.pop();
552 uint op = store->Opcode();
554 // MergeMems do not directly have anti-deps.
555 // Treat them as internal nodes in a forward tree of memory states,
556 // the leaves of which are each a 'possible-def'.
557 if (store == initial_mem // root (exclusive) of tree we are searching
558 || op == Op_MergeMem // internal node of tree we are searching
559 ) {
560 mem = store; // It's not a possibly interfering store.
561 if (store == initial_mem)
562 initial_mem = NULL; // only process initial memory once
564 for (DUIterator_Fast imax, i = mem->fast_outs(imax); i < imax; i++) {
565 store = mem->fast_out(i);
566 if (store->is_MergeMem()) {
567 // Be sure we don't get into combinatorial problems.
568 // (Allow phis to be repeated; they can merge two relevant states.)
569 uint j = worklist_visited.size();
570 for (; j > 0; j--) {
571 if (worklist_visited.at(j-1) == store) break;
572 }
573 if (j > 0) continue; // already on work list; do not repeat
574 worklist_visited.push(store);
575 }
576 worklist_mem.push(mem);
577 worklist_store.push(store);
578 }
579 continue;
580 }
582 if (op == Op_MachProj || op == Op_Catch) continue;
583 if (store->needs_anti_dependence_check()) continue; // not really a store
585 // Compute the alias index. Loads and stores with different alias
586 // indices do not need anti-dependence edges. Wide MemBar's are
587 // anti-dependent on everything (except immutable memories).
588 const TypePtr* adr_type = store->adr_type();
589 if (!C->can_alias(adr_type, load_alias_idx)) continue;
591 // Most slow-path runtime calls do NOT modify Java memory, but
592 // they can block and so write Raw memory.
593 if (store->is_Mach()) {
594 MachNode* mstore = store->as_Mach();
595 if (load_alias_idx != Compile::AliasIdxRaw) {
596 // Check for call into the runtime using the Java calling
597 // convention (and from there into a wrapper); it has no
598 // _method. Can't do this optimization for Native calls because
599 // they CAN write to Java memory.
600 if (mstore->ideal_Opcode() == Op_CallStaticJava) {
601 assert(mstore->is_MachSafePoint(), "");
602 MachSafePointNode* ms = (MachSafePointNode*) mstore;
603 assert(ms->is_MachCallJava(), "");
604 MachCallJavaNode* mcj = (MachCallJavaNode*) ms;
605 if (mcj->_method == NULL) {
606 // These runtime calls do not write to Java visible memory
607 // (other than Raw) and so do not require anti-dependence edges.
608 continue;
609 }
610 }
611 // Same for SafePoints: they read/write Raw but only read otherwise.
612 // This is basically a workaround for SafePoints only defining control
613 // instead of control + memory.
614 if (mstore->ideal_Opcode() == Op_SafePoint)
615 continue;
616 } else {
617 // Some raw memory, such as the load of "top" at an allocation,
618 // can be control dependent on the previous safepoint. See
619 // comments in GraphKit::allocate_heap() about control input.
620 // Inserting an anti-dep between such a safepoint and a use
621 // creates a cycle, and will cause a subsequent failure in
622 // local scheduling. (BugId 4919904)
623 // (%%% How can a control input be a safepoint and not a projection??)
624 if (mstore->ideal_Opcode() == Op_SafePoint && load->in(0) == mstore)
625 continue;
626 }
627 }
629 // Identify a block that the current load must be above,
630 // or else observe that 'store' is all the way up in the
631 // earliest legal block for 'load'. In the latter case,
632 // immediately insert an anti-dependence edge.
633 Block* store_block = get_block_for_node(store);
634 assert(store_block != NULL, "unused killing projections skipped above");
636 if (store->is_Phi()) {
637 // 'load' uses memory which is one (or more) of the Phi's inputs.
638 // It must be scheduled not before the Phi, but rather before
639 // each of the relevant Phi inputs.
640 //
641 // Instead of finding the LCA of all inputs to a Phi that match 'mem',
642 // we mark each corresponding predecessor block and do a combined
643 // hoisting operation later (raise_LCA_above_marks).
644 //
645 // Do not assert(store_block != early, "Phi merging memory after access")
646 // PhiNode may be at start of block 'early' with backedge to 'early'
647 DEBUG_ONLY(bool found_match = false);
648 for (uint j = PhiNode::Input, jmax = store->req(); j < jmax; j++) {
649 if (store->in(j) == mem) { // Found matching input?
650 DEBUG_ONLY(found_match = true);
651 Block* pred_block = get_block_for_node(store_block->pred(j));
652 if (pred_block != early) {
653 // If any predecessor of the Phi matches the load's "early block",
654 // we do not need a precedence edge between the Phi and 'load'
655 // since the load will be forced into a block preceding the Phi.
656 pred_block->set_raise_LCA_mark(load_index);
657 assert(!LCA_orig->dominates(pred_block) ||
658 early->dominates(pred_block), "early is high enough");
659 must_raise_LCA = true;
660 } else {
661 // anti-dependent upon PHI pinned below 'early', no edge needed
662 LCA = early; // but can not schedule below 'early'
663 }
664 }
665 }
666 assert(found_match, "no worklist bug");
667 #ifdef TRACK_PHI_INPUTS
668 #ifdef ASSERT
669 // This assert asks about correct handling of PhiNodes, which may not
670 // have all input edges directly from 'mem'. See BugId 4621264
671 int num_mem_inputs = phi_inputs.at_grow(store->_idx,0) + 1;
672 // Increment by exactly one even if there are multiple copies of 'mem'
673 // coming into the phi, because we will run this block several times
674 // if there are several copies of 'mem'. (That's how DU iterators work.)
675 phi_inputs.at_put(store->_idx, num_mem_inputs);
676 assert(PhiNode::Input + num_mem_inputs < store->req(),
677 "Expect at least one phi input will not be from original memory state");
678 #endif //ASSERT
679 #endif //TRACK_PHI_INPUTS
680 } else if (store_block != early) {
681 // 'store' is between the current LCA and earliest possible block.
682 // Label its block, and decide later on how to raise the LCA
683 // to include the effect on LCA of this store.
684 // If this store's block gets chosen as the raised LCA, we
685 // will find him on the non_early_stores list and stick him
686 // with a precedence edge.
687 // (But, don't bother if LCA is already raised all the way.)
688 if (LCA != early) {
689 store_block->set_raise_LCA_mark(load_index);
690 must_raise_LCA = true;
691 non_early_stores.push(store);
692 }
693 } else {
694 // Found a possibly-interfering store in the load's 'early' block.
695 // This means 'load' cannot sink at all in the dominator tree.
696 // Add an anti-dep edge, and squeeze 'load' into the highest block.
697 assert(store != load->in(0), "dependence cycle found");
698 if (verify) {
699 assert(store->find_edge(load) != -1, "missing precedence edge");
700 } else {
701 store->add_prec(load);
702 }
703 LCA = early;
704 // This turns off the process of gathering non_early_stores.
705 }
706 }
707 // (Worklist is now empty; all nearby stores have been visited.)
709 // Finished if 'load' must be scheduled in its 'early' block.
710 // If we found any stores there, they have already been given
711 // precedence edges.
712 if (LCA == early) return LCA;
714 // We get here only if there are no possibly-interfering stores
715 // in the load's 'early' block. Move LCA up above all predecessors
716 // which contain stores we have noted.
717 //
718 // The raised LCA block can be a home to such interfering stores,
719 // but its predecessors must not contain any such stores.
720 //
721 // The raised LCA will be a lower bound for placing the load,
722 // preventing the load from sinking past any block containing
723 // a store that may invalidate the memory state required by 'load'.
724 if (must_raise_LCA)
725 LCA = raise_LCA_above_marks(LCA, load->_idx, early, this);
726 if (LCA == early) return LCA;
728 // Insert anti-dependence edges from 'load' to each store
729 // in the non-early LCA block.
730 // Mine the non_early_stores list for such stores.
731 if (LCA->raise_LCA_mark() == load_index) {
732 while (non_early_stores.size() > 0) {
733 Node* store = non_early_stores.pop();
734 Block* store_block = get_block_for_node(store);
735 if (store_block == LCA) {
736 // add anti_dependence from store to load in its own block
737 assert(store != load->in(0), "dependence cycle found");
738 if (verify) {
739 assert(store->find_edge(load) != -1, "missing precedence edge");
740 } else {
741 store->add_prec(load);
742 }
743 } else {
744 assert(store_block->raise_LCA_mark() == load_index, "block was marked");
745 // Any other stores we found must be either inside the new LCA
746 // or else outside the original LCA. In the latter case, they
747 // did not interfere with any use of 'load'.
748 assert(LCA->dominates(store_block)
749 || !LCA_orig->dominates(store_block), "no stray stores");
750 }
751 }
752 }
754 // Return the highest block containing stores; any stores
755 // within that block have been given anti-dependence edges.
756 return LCA;
757 }
759 // This class is used to iterate backwards over the nodes in the graph.
761 class Node_Backward_Iterator {
763 private:
764 Node_Backward_Iterator();
766 public:
767 // Constructor for the iterator
768 Node_Backward_Iterator(Node *root, VectorSet &visited, Node_List &stack, PhaseCFG &cfg);
770 // Postincrement operator to iterate over the nodes
771 Node *next();
773 private:
774 VectorSet &_visited;
775 Node_List &_stack;
776 PhaseCFG &_cfg;
777 };
779 // Constructor for the Node_Backward_Iterator
780 Node_Backward_Iterator::Node_Backward_Iterator( Node *root, VectorSet &visited, Node_List &stack, PhaseCFG &cfg)
781 : _visited(visited), _stack(stack), _cfg(cfg) {
782 // The stack should contain exactly the root
783 stack.clear();
784 stack.push(root);
786 // Clear the visited bits
787 visited.Clear();
788 }
790 // Iterator for the Node_Backward_Iterator
791 Node *Node_Backward_Iterator::next() {
793 // If the _stack is empty, then just return NULL: finished.
794 if ( !_stack.size() )
795 return NULL;
797 // '_stack' is emulating a real _stack. The 'visit-all-users' loop has been
798 // made stateless, so I do not need to record the index 'i' on my _stack.
799 // Instead I visit all users each time, scanning for unvisited users.
800 // I visit unvisited not-anti-dependence users first, then anti-dependent
801 // children next.
802 Node *self = _stack.pop();
804 // I cycle here when I am entering a deeper level of recursion.
805 // The key variable 'self' was set prior to jumping here.
806 while( 1 ) {
808 _visited.set(self->_idx);
810 // Now schedule all uses as late as possible.
811 const Node* src = self->is_Proj() ? self->in(0) : self;
812 uint src_rpo = _cfg.get_block_for_node(src)->_rpo;
814 // Schedule all nodes in a post-order visit
815 Node *unvisited = NULL; // Unvisited anti-dependent Node, if any
817 // Scan for unvisited nodes
818 for (DUIterator_Fast imax, i = self->fast_outs(imax); i < imax; i++) {
819 // For all uses, schedule late
820 Node* n = self->fast_out(i); // Use
822 // Skip already visited children
823 if ( _visited.test(n->_idx) )
824 continue;
826 // do not traverse backward control edges
827 Node *use = n->is_Proj() ? n->in(0) : n;
828 uint use_rpo = _cfg.get_block_for_node(use)->_rpo;
830 if ( use_rpo < src_rpo )
831 continue;
833 // Phi nodes always precede uses in a basic block
834 if ( use_rpo == src_rpo && use->is_Phi() )
835 continue;
837 unvisited = n; // Found unvisited
839 // Check for possible-anti-dependent
840 if( !n->needs_anti_dependence_check() )
841 break; // Not visited, not anti-dep; schedule it NOW
842 }
844 // Did I find an unvisited not-anti-dependent Node?
845 if ( !unvisited )
846 break; // All done with children; post-visit 'self'
848 // Visit the unvisited Node. Contains the obvious push to
849 // indicate I'm entering a deeper level of recursion. I push the
850 // old state onto the _stack and set a new state and loop (recurse).
851 _stack.push(self);
852 self = unvisited;
853 } // End recursion loop
855 return self;
856 }
858 //------------------------------ComputeLatenciesBackwards----------------------
859 // Compute the latency of all the instructions.
860 void PhaseCFG::compute_latencies_backwards(VectorSet &visited, Node_List &stack) {
861 #ifndef PRODUCT
862 if (trace_opto_pipelining())
863 tty->print("\n#---- ComputeLatenciesBackwards ----\n");
864 #endif
866 Node_Backward_Iterator iter((Node *)_root, visited, stack, *this);
867 Node *n;
869 // Walk over all the nodes from last to first
870 while (n = iter.next()) {
871 // Set the latency for the definitions of this instruction
872 partial_latency_of_defs(n);
873 }
874 } // end ComputeLatenciesBackwards
876 //------------------------------partial_latency_of_defs------------------------
877 // Compute the latency impact of this node on all defs. This computes
878 // a number that increases as we approach the beginning of the routine.
879 void PhaseCFG::partial_latency_of_defs(Node *n) {
880 // Set the latency for this instruction
881 #ifndef PRODUCT
882 if (trace_opto_pipelining()) {
883 tty->print("# latency_to_inputs: node_latency[%d] = %d for node", n->_idx, get_latency_for_node(n));
884 dump();
885 }
886 #endif
888 if (n->is_Proj()) {
889 n = n->in(0);
890 }
892 if (n->is_Root()) {
893 return;
894 }
896 uint nlen = n->len();
897 uint use_latency = get_latency_for_node(n);
898 uint use_pre_order = get_block_for_node(n)->_pre_order;
900 for (uint j = 0; j < nlen; j++) {
901 Node *def = n->in(j);
903 if (!def || def == n) {
904 continue;
905 }
907 // Walk backwards thru projections
908 if (def->is_Proj()) {
909 def = def->in(0);
910 }
912 #ifndef PRODUCT
913 if (trace_opto_pipelining()) {
914 tty->print("# in(%2d): ", j);
915 def->dump();
916 }
917 #endif
919 // If the defining block is not known, assume it is ok
920 Block *def_block = get_block_for_node(def);
921 uint def_pre_order = def_block ? def_block->_pre_order : 0;
923 if ((use_pre_order < def_pre_order) || (use_pre_order == def_pre_order && n->is_Phi())) {
924 continue;
925 }
927 uint delta_latency = n->latency(j);
928 uint current_latency = delta_latency + use_latency;
930 if (get_latency_for_node(def) < current_latency) {
931 set_latency_for_node(def, current_latency);
932 }
934 #ifndef PRODUCT
935 if (trace_opto_pipelining()) {
936 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));
937 }
938 #endif
939 }
940 }
942 //------------------------------latency_from_use-------------------------------
943 // Compute the latency of a specific use
944 int PhaseCFG::latency_from_use(Node *n, const Node *def, Node *use) {
945 // If self-reference, return no latency
946 if (use == n || use->is_Root()) {
947 return 0;
948 }
950 uint def_pre_order = get_block_for_node(def)->_pre_order;
951 uint latency = 0;
953 // If the use is not a projection, then it is simple...
954 if (!use->is_Proj()) {
955 #ifndef PRODUCT
956 if (trace_opto_pipelining()) {
957 tty->print("# out(): ");
958 use->dump();
959 }
960 #endif
962 uint use_pre_order = get_block_for_node(use)->_pre_order;
964 if (use_pre_order < def_pre_order)
965 return 0;
967 if (use_pre_order == def_pre_order && use->is_Phi())
968 return 0;
970 uint nlen = use->len();
971 uint nl = get_latency_for_node(use);
973 for ( uint j=0; j<nlen; j++ ) {
974 if (use->in(j) == n) {
975 // Change this if we want local latencies
976 uint ul = use->latency(j);
977 uint l = ul + nl;
978 if (latency < l) latency = l;
979 #ifndef PRODUCT
980 if (trace_opto_pipelining()) {
981 tty->print_cr("# %d + edge_latency(%d) == %d -> %d, latency = %d",
982 nl, j, ul, l, latency);
983 }
984 #endif
985 }
986 }
987 } else {
988 // This is a projection, just grab the latency of the use(s)
989 for (DUIterator_Fast jmax, j = use->fast_outs(jmax); j < jmax; j++) {
990 uint l = latency_from_use(use, def, use->fast_out(j));
991 if (latency < l) latency = l;
992 }
993 }
995 return latency;
996 }
998 //------------------------------latency_from_uses------------------------------
999 // Compute the latency of this instruction relative to all of it's uses.
1000 // This computes a number that increases as we approach the beginning of the
1001 // routine.
1002 void PhaseCFG::latency_from_uses(Node *n) {
1003 // Set the latency for this instruction
1004 #ifndef PRODUCT
1005 if (trace_opto_pipelining()) {
1006 tty->print("# latency_from_outputs: node_latency[%d] = %d for node", n->_idx, get_latency_for_node(n));
1007 dump();
1008 }
1009 #endif
1010 uint latency=0;
1011 const Node *def = n->is_Proj() ? n->in(0): n;
1013 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1014 uint l = latency_from_use(n, def, n->fast_out(i));
1016 if (latency < l) latency = l;
1017 }
1019 set_latency_for_node(n, latency);
1020 }
1022 //------------------------------hoist_to_cheaper_block-------------------------
1023 // Pick a block for node self, between early and LCA, that is a cheaper
1024 // alternative to LCA.
1025 Block* PhaseCFG::hoist_to_cheaper_block(Block* LCA, Block* early, Node* self) {
1026 const double delta = 1+PROB_UNLIKELY_MAG(4);
1027 Block* least = LCA;
1028 double least_freq = least->_freq;
1029 uint target = get_latency_for_node(self);
1030 uint start_latency = get_latency_for_node(LCA->_nodes[0]);
1031 uint end_latency = get_latency_for_node(LCA->_nodes[LCA->end_idx()]);
1032 bool in_latency = (target <= start_latency);
1033 const Block* root_block = get_block_for_node(_root);
1035 // Turn off latency scheduling if scheduling is just plain off
1036 if (!C->do_scheduling())
1037 in_latency = true;
1039 // Do not hoist (to cover latency) instructions which target a
1040 // single register. Hoisting stretches the live range of the
1041 // single register and may force spilling.
1042 MachNode* mach = self->is_Mach() ? self->as_Mach() : NULL;
1043 if (mach && mach->out_RegMask().is_bound1() && mach->out_RegMask().is_NotEmpty())
1044 in_latency = true;
1046 #ifndef PRODUCT
1047 if (trace_opto_pipelining()) {
1048 tty->print("# Find cheaper block for latency %d: ", get_latency_for_node(self));
1049 self->dump();
1050 tty->print_cr("# B%d: start latency for [%4d]=%d, end latency for [%4d]=%d, freq=%g",
1051 LCA->_pre_order,
1052 LCA->_nodes[0]->_idx,
1053 start_latency,
1054 LCA->_nodes[LCA->end_idx()]->_idx,
1055 end_latency,
1056 least_freq);
1057 }
1058 #endif
1060 int cand_cnt = 0; // number of candidates tried
1062 // Walk up the dominator tree from LCA (Lowest common ancestor) to
1063 // the earliest legal location. Capture the least execution frequency.
1064 while (LCA != early) {
1065 LCA = LCA->_idom; // Follow up the dominator tree
1067 if (LCA == NULL) {
1068 // Bailout without retry
1069 C->record_method_not_compilable("late schedule failed: LCA == NULL");
1070 return least;
1071 }
1073 // Don't hoist machine instructions to the root basic block
1074 if (mach && LCA == root_block)
1075 break;
1077 uint start_lat = get_latency_for_node(LCA->_nodes[0]);
1078 uint end_idx = LCA->end_idx();
1079 uint end_lat = get_latency_for_node(LCA->_nodes[end_idx]);
1080 double LCA_freq = LCA->_freq;
1081 #ifndef PRODUCT
1082 if (trace_opto_pipelining()) {
1083 tty->print_cr("# B%d: start latency for [%4d]=%d, end latency for [%4d]=%d, freq=%g",
1084 LCA->_pre_order, LCA->_nodes[0]->_idx, start_lat, end_idx, end_lat, LCA_freq);
1085 }
1086 #endif
1087 cand_cnt++;
1088 if (LCA_freq < least_freq || // Better Frequency
1089 (StressGCM && Compile::randomized_select(cand_cnt)) || // Should be randomly accepted in stress mode
1090 (!StressGCM && // Otherwise, choose with latency
1091 !in_latency && // No block containing latency
1092 LCA_freq < least_freq * delta && // No worse frequency
1093 target >= end_lat && // within latency range
1094 !self->is_iteratively_computed() ) // But don't hoist IV increments
1095 // because they may end up above other uses of their phi forcing
1096 // their result register to be different from their input.
1097 ) {
1098 least = LCA; // Found cheaper block
1099 least_freq = LCA_freq;
1100 start_latency = start_lat;
1101 end_latency = end_lat;
1102 if (target <= start_lat)
1103 in_latency = true;
1104 }
1105 }
1107 #ifndef PRODUCT
1108 if (trace_opto_pipelining()) {
1109 tty->print_cr("# Choose block B%d with start latency=%d and freq=%g",
1110 least->_pre_order, start_latency, least_freq);
1111 }
1112 #endif
1114 // See if the latency needs to be updated
1115 if (target < end_latency) {
1116 #ifndef PRODUCT
1117 if (trace_opto_pipelining()) {
1118 tty->print_cr("# Change latency for [%4d] from %d to %d", self->_idx, target, end_latency);
1119 }
1120 #endif
1121 set_latency_for_node(self, end_latency);
1122 partial_latency_of_defs(self);
1123 }
1125 return least;
1126 }
1129 //------------------------------schedule_late-----------------------------------
1130 // Now schedule all codes as LATE as possible. This is the LCA in the
1131 // dominator tree of all USES of a value. Pick the block with the least
1132 // loop nesting depth that is lowest in the dominator tree.
1133 extern const char must_clone[];
1134 void PhaseCFG::schedule_late(VectorSet &visited, Node_List &stack) {
1135 #ifndef PRODUCT
1136 if (trace_opto_pipelining())
1137 tty->print("\n#---- schedule_late ----\n");
1138 #endif
1140 Node_Backward_Iterator iter((Node *)_root, visited, stack, *this);
1141 Node *self;
1143 // Walk over all the nodes from last to first
1144 while (self = iter.next()) {
1145 Block* early = get_block_for_node(self); // Earliest legal placement
1147 if (self->is_top()) {
1148 // Top node goes in bb #2 with other constants.
1149 // It must be special-cased, because it has no out edges.
1150 early->add_inst(self);
1151 continue;
1152 }
1154 // No uses, just terminate
1155 if (self->outcnt() == 0) {
1156 assert(self->is_MachProj(), "sanity");
1157 continue; // Must be a dead machine projection
1158 }
1160 // If node is pinned in the block, then no scheduling can be done.
1161 if( self->pinned() ) // Pinned in block?
1162 continue;
1164 MachNode* mach = self->is_Mach() ? self->as_Mach() : NULL;
1165 if (mach) {
1166 switch (mach->ideal_Opcode()) {
1167 case Op_CreateEx:
1168 // Don't move exception creation
1169 early->add_inst(self);
1170 continue;
1171 break;
1172 case Op_CheckCastPP:
1173 // Don't move CheckCastPP nodes away from their input, if the input
1174 // is a rawptr (5071820).
1175 Node *def = self->in(1);
1176 if (def != NULL && def->bottom_type()->base() == Type::RawPtr) {
1177 early->add_inst(self);
1178 #ifdef ASSERT
1179 _raw_oops.push(def);
1180 #endif
1181 continue;
1182 }
1183 break;
1184 }
1185 }
1187 // Gather LCA of all uses
1188 Block *LCA = NULL;
1189 {
1190 for (DUIterator_Fast imax, i = self->fast_outs(imax); i < imax; i++) {
1191 // For all uses, find LCA
1192 Node* use = self->fast_out(i);
1193 LCA = raise_LCA_above_use(LCA, use, self, this);
1194 }
1195 } // (Hide defs of imax, i from rest of block.)
1197 // Place temps in the block of their use. This isn't a
1198 // requirement for correctness but it reduces useless
1199 // interference between temps and other nodes.
1200 if (mach != NULL && mach->is_MachTemp()) {
1201 map_node_to_block(self, LCA);
1202 LCA->add_inst(self);
1203 continue;
1204 }
1206 // Check if 'self' could be anti-dependent on memory
1207 if (self->needs_anti_dependence_check()) {
1208 // Hoist LCA above possible-defs and insert anti-dependences to
1209 // defs in new LCA block.
1210 LCA = insert_anti_dependences(LCA, self);
1211 }
1213 if (early->_dom_depth > LCA->_dom_depth) {
1214 // Somehow the LCA has moved above the earliest legal point.
1215 // (One way this can happen is via memory_early_block.)
1216 if (C->subsume_loads() == true && !C->failing()) {
1217 // Retry with subsume_loads == false
1218 // If this is the first failure, the sentinel string will "stick"
1219 // to the Compile object, and the C2Compiler will see it and retry.
1220 C->record_failure(C2Compiler::retry_no_subsuming_loads());
1221 } else {
1222 // Bailout without retry when (early->_dom_depth > LCA->_dom_depth)
1223 C->record_method_not_compilable("late schedule failed: incorrect graph");
1224 }
1225 return;
1226 }
1228 // If there is no opportunity to hoist, then we're done.
1229 // In stress mode, try to hoist even the single operations.
1230 bool try_to_hoist = StressGCM || (LCA != early);
1232 // Must clone guys stay next to use; no hoisting allowed.
1233 // Also cannot hoist guys that alter memory or are otherwise not
1234 // allocatable (hoisting can make a value live longer, leading to
1235 // anti and output dependency problems which are normally resolved
1236 // by the register allocator giving everyone a different register).
1237 if (mach != NULL && must_clone[mach->ideal_Opcode()])
1238 try_to_hoist = false;
1240 Block* late = NULL;
1241 if (try_to_hoist) {
1242 // Now find the block with the least execution frequency.
1243 // Start at the latest schedule and work up to the earliest schedule
1244 // in the dominator tree. Thus the Node will dominate all its uses.
1245 late = hoist_to_cheaper_block(LCA, early, self);
1246 } else {
1247 // Just use the LCA of the uses.
1248 late = LCA;
1249 }
1251 // Put the node into target block
1252 schedule_node_into_block(self, late);
1254 #ifdef ASSERT
1255 if (self->needs_anti_dependence_check()) {
1256 // since precedence edges are only inserted when we're sure they
1257 // are needed make sure that after placement in a block we don't
1258 // need any new precedence edges.
1259 verify_anti_dependences(late, self);
1260 }
1261 #endif
1262 } // Loop until all nodes have been visited
1264 } // end ScheduleLate
1266 //------------------------------GlobalCodeMotion-------------------------------
1267 void PhaseCFG::global_code_motion() {
1268 ResourceMark rm;
1270 #ifndef PRODUCT
1271 if (trace_opto_pipelining()) {
1272 tty->print("\n---- Start GlobalCodeMotion ----\n");
1273 }
1274 #endif
1276 // Initialize the node to block mapping for things on the proj_list
1277 for (uint i = 0; i < _matcher.number_of_projections(); i++) {
1278 unmap_node_from_block(_matcher.get_projection(i));
1279 }
1281 // Set the basic block for Nodes pinned into blocks
1282 Arena* arena = Thread::current()->resource_area();
1283 VectorSet visited(arena);
1284 schedule_pinned_nodes(visited);
1286 // Find the earliest Block any instruction can be placed in. Some
1287 // instructions are pinned into Blocks. Unpinned instructions can
1288 // appear in last block in which all their inputs occur.
1289 visited.Clear();
1290 Node_List stack(arena);
1291 // Pre-grow the list
1292 stack.map((C->unique() >> 1) + 16, NULL);
1293 if (!schedule_early(visited, stack)) {
1294 // Bailout without retry
1295 C->record_method_not_compilable("early schedule failed");
1296 return;
1297 }
1299 // Build Def-Use edges.
1300 // Compute the latency information (via backwards walk) for all the
1301 // instructions in the graph
1302 _node_latency = new GrowableArray<uint>(); // resource_area allocation
1304 if (C->do_scheduling()) {
1305 compute_latencies_backwards(visited, stack);
1306 }
1308 // Now schedule all codes as LATE as possible. This is the LCA in the
1309 // dominator tree of all USES of a value. Pick the block with the least
1310 // loop nesting depth that is lowest in the dominator tree.
1311 // ( visited.Clear() called in schedule_late()->Node_Backward_Iterator() )
1312 schedule_late(visited, stack);
1313 if (C->failing()) {
1314 // schedule_late fails only when graph is incorrect.
1315 assert(!VerifyGraphEdges, "verification should have failed");
1316 return;
1317 }
1319 #ifndef PRODUCT
1320 if (trace_opto_pipelining()) {
1321 tty->print("\n---- Detect implicit null checks ----\n");
1322 }
1323 #endif
1325 // Detect implicit-null-check opportunities. Basically, find NULL checks
1326 // with suitable memory ops nearby. Use the memory op to do the NULL check.
1327 // I can generate a memory op if there is not one nearby.
1328 if (C->is_method_compilation()) {
1329 // Don't do it for natives, adapters, or runtime stubs
1330 int allowed_reasons = 0;
1331 // ...and don't do it when there have been too many traps, globally.
1332 for (int reason = (int)Deoptimization::Reason_none+1;
1333 reason < Compile::trapHistLength; reason++) {
1334 assert(reason < BitsPerInt, "recode bit map");
1335 if (!C->too_many_traps((Deoptimization::DeoptReason) reason))
1336 allowed_reasons |= nth_bit(reason);
1337 }
1338 // By reversing the loop direction we get a very minor gain on mpegaudio.
1339 // Feel free to revert to a forward loop for clarity.
1340 // for( int i=0; i < (int)matcher._null_check_tests.size(); i+=2 ) {
1341 for (int i = _matcher._null_check_tests.size() - 2; i >= 0; i -= 2) {
1342 Node* proj = _matcher._null_check_tests[i];
1343 Node* val = _matcher._null_check_tests[i + 1];
1344 Block* block = get_block_for_node(proj);
1345 block->implicit_null_check(this, proj, val, allowed_reasons);
1346 // The implicit_null_check will only perform the transformation
1347 // if the null branch is truly uncommon, *and* it leads to an
1348 // uncommon trap. Combined with the too_many_traps guards
1349 // above, this prevents SEGV storms reported in 6366351,
1350 // by recompiling offending methods without this optimization.
1351 }
1352 }
1354 #ifndef PRODUCT
1355 if (trace_opto_pipelining()) {
1356 tty->print("\n---- Start Local Scheduling ----\n");
1357 }
1358 #endif
1360 // Schedule locally. Right now a simple topological sort.
1361 // Later, do a real latency aware scheduler.
1362 GrowableArray<int> ready_cnt(C->unique(), C->unique(), -1);
1363 visited.Clear();
1364 for (uint i = 0; i < number_of_blocks(); i++) {
1365 Block* block = get_block(i);
1366 if (!block->schedule_local(this, _matcher, ready_cnt, visited)) {
1367 if (!C->failure_reason_is(C2Compiler::retry_no_subsuming_loads())) {
1368 C->record_method_not_compilable("local schedule failed");
1369 }
1370 return;
1371 }
1372 }
1374 // If we inserted any instructions between a Call and his CatchNode,
1375 // clone the instructions on all paths below the Catch.
1376 for (uint i = 0; i < number_of_blocks(); i++) {
1377 Block* block = get_block(i);
1378 block->call_catch_cleanup(this, C);
1379 }
1381 #ifndef PRODUCT
1382 if (trace_opto_pipelining()) {
1383 tty->print("\n---- After GlobalCodeMotion ----\n");
1384 for (uint i = 0; i < number_of_blocks(); i++) {
1385 Block* block = get_block(i);
1386 block->dump();
1387 }
1388 }
1389 #endif
1390 // Dead.
1391 _node_latency = (GrowableArray<uint> *)0xdeadbeef;
1392 }
1394 bool PhaseCFG::do_global_code_motion() {
1396 build_dominator_tree();
1397 if (C->failing()) {
1398 return false;
1399 }
1401 NOT_PRODUCT( C->verify_graph_edges(); )
1403 estimate_block_frequency();
1405 global_code_motion();
1407 if (C->failing()) {
1408 return false;
1409 }
1411 return true;
1412 }
1414 //------------------------------Estimate_Block_Frequency-----------------------
1415 // Estimate block frequencies based on IfNode probabilities.
1416 void PhaseCFG::estimate_block_frequency() {
1418 // Force conditional branches leading to uncommon traps to be unlikely,
1419 // not because we get to the uncommon_trap with less relative frequency,
1420 // but because an uncommon_trap typically causes a deopt, so we only get
1421 // there once.
1422 if (C->do_freq_based_layout()) {
1423 Block_List worklist;
1424 Block* root_blk = get_block(0);
1425 for (uint i = 1; i < root_blk->num_preds(); i++) {
1426 Block *pb = get_block_for_node(root_blk->pred(i));
1427 if (pb->has_uncommon_code()) {
1428 worklist.push(pb);
1429 }
1430 }
1431 while (worklist.size() > 0) {
1432 Block* uct = worklist.pop();
1433 if (uct == get_root_block()) {
1434 continue;
1435 }
1436 for (uint i = 1; i < uct->num_preds(); i++) {
1437 Block *pb = get_block_for_node(uct->pred(i));
1438 if (pb->_num_succs == 1) {
1439 worklist.push(pb);
1440 } else if (pb->num_fall_throughs() == 2) {
1441 pb->update_uncommon_branch(uct);
1442 }
1443 }
1444 }
1445 }
1447 // Create the loop tree and calculate loop depth.
1448 _root_loop = create_loop_tree();
1449 _root_loop->compute_loop_depth(0);
1451 // Compute block frequency of each block, relative to a single loop entry.
1452 _root_loop->compute_freq();
1454 // Adjust all frequencies to be relative to a single method entry
1455 _root_loop->_freq = 1.0;
1456 _root_loop->scale_freq();
1458 // Save outmost loop frequency for LRG frequency threshold
1459 _outer_loop_frequency = _root_loop->outer_loop_freq();
1461 // force paths ending at uncommon traps to be infrequent
1462 if (!C->do_freq_based_layout()) {
1463 Block_List worklist;
1464 Block* root_blk = get_block(0);
1465 for (uint i = 1; i < root_blk->num_preds(); i++) {
1466 Block *pb = get_block_for_node(root_blk->pred(i));
1467 if (pb->has_uncommon_code()) {
1468 worklist.push(pb);
1469 }
1470 }
1471 while (worklist.size() > 0) {
1472 Block* uct = worklist.pop();
1473 uct->_freq = PROB_MIN;
1474 for (uint i = 1; i < uct->num_preds(); i++) {
1475 Block *pb = get_block_for_node(uct->pred(i));
1476 if (pb->_num_succs == 1 && pb->_freq > PROB_MIN) {
1477 worklist.push(pb);
1478 }
1479 }
1480 }
1481 }
1483 #ifdef ASSERT
1484 for (uint i = 0; i < number_of_blocks(); i++) {
1485 Block* b = get_block(i);
1486 assert(b->_freq >= MIN_BLOCK_FREQUENCY, "Register Allocator requires meaningful block frequency");
1487 }
1488 #endif
1490 #ifndef PRODUCT
1491 if (PrintCFGBlockFreq) {
1492 tty->print_cr("CFG Block Frequencies");
1493 _root_loop->dump_tree();
1494 if (Verbose) {
1495 tty->print_cr("PhaseCFG dump");
1496 dump();
1497 tty->print_cr("Node dump");
1498 _root->dump(99999);
1499 }
1500 }
1501 #endif
1502 }
1504 //----------------------------create_loop_tree--------------------------------
1505 // Create a loop tree from the CFG
1506 CFGLoop* PhaseCFG::create_loop_tree() {
1508 #ifdef ASSERT
1509 assert(get_block(0) == get_root_block(), "first block should be root block");
1510 for (uint i = 0; i < number_of_blocks(); i++) {
1511 Block* block = get_block(i);
1512 // Check that _loop field are clear...we could clear them if not.
1513 assert(block->_loop == NULL, "clear _loop expected");
1514 // Sanity check that the RPO numbering is reflected in the _blocks array.
1515 // It doesn't have to be for the loop tree to be built, but if it is not,
1516 // then the blocks have been reordered since dom graph building...which
1517 // may question the RPO numbering
1518 assert(block->_rpo == i, "unexpected reverse post order number");
1519 }
1520 #endif
1522 int idct = 0;
1523 CFGLoop* root_loop = new CFGLoop(idct++);
1525 Block_List worklist;
1527 // Assign blocks to loops
1528 for(uint i = number_of_blocks() - 1; i > 0; i-- ) { // skip Root block
1529 Block* block = get_block(i);
1531 if (block->head()->is_Loop()) {
1532 Block* loop_head = block;
1533 assert(loop_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
1534 Node* tail_n = loop_head->pred(LoopNode::LoopBackControl);
1535 Block* tail = get_block_for_node(tail_n);
1537 // Defensively filter out Loop nodes for non-single-entry loops.
1538 // For all reasonable loops, the head occurs before the tail in RPO.
1539 if (i <= tail->_rpo) {
1541 // The tail and (recursive) predecessors of the tail
1542 // are made members of a new loop.
1544 assert(worklist.size() == 0, "nonempty worklist");
1545 CFGLoop* nloop = new CFGLoop(idct++);
1546 assert(loop_head->_loop == NULL, "just checking");
1547 loop_head->_loop = nloop;
1548 // Add to nloop so push_pred() will skip over inner loops
1549 nloop->add_member(loop_head);
1550 nloop->push_pred(loop_head, LoopNode::LoopBackControl, worklist, this);
1552 while (worklist.size() > 0) {
1553 Block* member = worklist.pop();
1554 if (member != loop_head) {
1555 for (uint j = 1; j < member->num_preds(); j++) {
1556 nloop->push_pred(member, j, worklist, this);
1557 }
1558 }
1559 }
1560 }
1561 }
1562 }
1564 // Create a member list for each loop consisting
1565 // of both blocks and (immediate child) loops.
1566 for (uint i = 0; i < number_of_blocks(); i++) {
1567 Block* block = get_block(i);
1568 CFGLoop* lp = block->_loop;
1569 if (lp == NULL) {
1570 // Not assigned to a loop. Add it to the method's pseudo loop.
1571 block->_loop = root_loop;
1572 lp = root_loop;
1573 }
1574 if (lp == root_loop || block != lp->head()) { // loop heads are already members
1575 lp->add_member(block);
1576 }
1577 if (lp != root_loop) {
1578 if (lp->parent() == NULL) {
1579 // Not a nested loop. Make it a child of the method's pseudo loop.
1580 root_loop->add_nested_loop(lp);
1581 }
1582 if (block == lp->head()) {
1583 // Add nested loop to member list of parent loop.
1584 lp->parent()->add_member(lp);
1585 }
1586 }
1587 }
1589 return root_loop;
1590 }
1592 //------------------------------push_pred--------------------------------------
1593 void CFGLoop::push_pred(Block* blk, int i, Block_List& worklist, PhaseCFG* cfg) {
1594 Node* pred_n = blk->pred(i);
1595 Block* pred = cfg->get_block_for_node(pred_n);
1596 CFGLoop *pred_loop = pred->_loop;
1597 if (pred_loop == NULL) {
1598 // Filter out blocks for non-single-entry loops.
1599 // For all reasonable loops, the head occurs before the tail in RPO.
1600 if (pred->_rpo > head()->_rpo) {
1601 pred->_loop = this;
1602 worklist.push(pred);
1603 }
1604 } else if (pred_loop != this) {
1605 // Nested loop.
1606 while (pred_loop->_parent != NULL && pred_loop->_parent != this) {
1607 pred_loop = pred_loop->_parent;
1608 }
1609 // Make pred's loop be a child
1610 if (pred_loop->_parent == NULL) {
1611 add_nested_loop(pred_loop);
1612 // Continue with loop entry predecessor.
1613 Block* pred_head = pred_loop->head();
1614 assert(pred_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
1615 assert(pred_head != head(), "loop head in only one loop");
1616 push_pred(pred_head, LoopNode::EntryControl, worklist, cfg);
1617 } else {
1618 assert(pred_loop->_parent == this && _parent == NULL, "just checking");
1619 }
1620 }
1621 }
1623 //------------------------------add_nested_loop--------------------------------
1624 // Make cl a child of the current loop in the loop tree.
1625 void CFGLoop::add_nested_loop(CFGLoop* cl) {
1626 assert(_parent == NULL, "no parent yet");
1627 assert(cl != this, "not my own parent");
1628 cl->_parent = this;
1629 CFGLoop* ch = _child;
1630 if (ch == NULL) {
1631 _child = cl;
1632 } else {
1633 while (ch->_sibling != NULL) { ch = ch->_sibling; }
1634 ch->_sibling = cl;
1635 }
1636 }
1638 //------------------------------compute_loop_depth-----------------------------
1639 // Store the loop depth in each CFGLoop object.
1640 // Recursively walk the children to do the same for them.
1641 void CFGLoop::compute_loop_depth(int depth) {
1642 _depth = depth;
1643 CFGLoop* ch = _child;
1644 while (ch != NULL) {
1645 ch->compute_loop_depth(depth + 1);
1646 ch = ch->_sibling;
1647 }
1648 }
1650 //------------------------------compute_freq-----------------------------------
1651 // Compute the frequency of each block and loop, relative to a single entry
1652 // into the dominating loop head.
1653 void CFGLoop::compute_freq() {
1654 // Bottom up traversal of loop tree (visit inner loops first.)
1655 // Set loop head frequency to 1.0, then transitively
1656 // compute frequency for all successors in the loop,
1657 // as well as for each exit edge. Inner loops are
1658 // treated as single blocks with loop exit targets
1659 // as the successor blocks.
1661 // Nested loops first
1662 CFGLoop* ch = _child;
1663 while (ch != NULL) {
1664 ch->compute_freq();
1665 ch = ch->_sibling;
1666 }
1667 assert (_members.length() > 0, "no empty loops");
1668 Block* hd = head();
1669 hd->_freq = 1.0f;
1670 for (int i = 0; i < _members.length(); i++) {
1671 CFGElement* s = _members.at(i);
1672 float freq = s->_freq;
1673 if (s->is_block()) {
1674 Block* b = s->as_Block();
1675 for (uint j = 0; j < b->_num_succs; j++) {
1676 Block* sb = b->_succs[j];
1677 update_succ_freq(sb, freq * b->succ_prob(j));
1678 }
1679 } else {
1680 CFGLoop* lp = s->as_CFGLoop();
1681 assert(lp->_parent == this, "immediate child");
1682 for (int k = 0; k < lp->_exits.length(); k++) {
1683 Block* eb = lp->_exits.at(k).get_target();
1684 float prob = lp->_exits.at(k).get_prob();
1685 update_succ_freq(eb, freq * prob);
1686 }
1687 }
1688 }
1690 // For all loops other than the outer, "method" loop,
1691 // sum and normalize the exit probability. The "method" loop
1692 // should keep the initial exit probability of 1, so that
1693 // inner blocks do not get erroneously scaled.
1694 if (_depth != 0) {
1695 // Total the exit probabilities for this loop.
1696 float exits_sum = 0.0f;
1697 for (int i = 0; i < _exits.length(); i++) {
1698 exits_sum += _exits.at(i).get_prob();
1699 }
1701 // Normalize the exit probabilities. Until now, the
1702 // probabilities estimate the possibility of exit per
1703 // a single loop iteration; afterward, they estimate
1704 // the probability of exit per loop entry.
1705 for (int i = 0; i < _exits.length(); i++) {
1706 Block* et = _exits.at(i).get_target();
1707 float new_prob = 0.0f;
1708 if (_exits.at(i).get_prob() > 0.0f) {
1709 new_prob = _exits.at(i).get_prob() / exits_sum;
1710 }
1711 BlockProbPair bpp(et, new_prob);
1712 _exits.at_put(i, bpp);
1713 }
1715 // Save the total, but guard against unreasonable probability,
1716 // as the value is used to estimate the loop trip count.
1717 // An infinite trip count would blur relative block
1718 // frequencies.
1719 if (exits_sum > 1.0f) exits_sum = 1.0;
1720 if (exits_sum < PROB_MIN) exits_sum = PROB_MIN;
1721 _exit_prob = exits_sum;
1722 }
1723 }
1725 //------------------------------succ_prob-------------------------------------
1726 // Determine the probability of reaching successor 'i' from the receiver block.
1727 float Block::succ_prob(uint i) {
1728 int eidx = end_idx();
1729 Node *n = _nodes[eidx]; // Get ending Node
1731 int op = n->Opcode();
1732 if (n->is_Mach()) {
1733 if (n->is_MachNullCheck()) {
1734 // Can only reach here if called after lcm. The original Op_If is gone,
1735 // so we attempt to infer the probability from one or both of the
1736 // successor blocks.
1737 assert(_num_succs == 2, "expecting 2 successors of a null check");
1738 // If either successor has only one predecessor, then the
1739 // probability estimate can be derived using the
1740 // relative frequency of the successor and this block.
1741 if (_succs[i]->num_preds() == 2) {
1742 return _succs[i]->_freq / _freq;
1743 } else if (_succs[1-i]->num_preds() == 2) {
1744 return 1 - (_succs[1-i]->_freq / _freq);
1745 } else {
1746 // Estimate using both successor frequencies
1747 float freq = _succs[i]->_freq;
1748 return freq / (freq + _succs[1-i]->_freq);
1749 }
1750 }
1751 op = n->as_Mach()->ideal_Opcode();
1752 }
1755 // Switch on branch type
1756 switch( op ) {
1757 case Op_CountedLoopEnd:
1758 case Op_If: {
1759 assert (i < 2, "just checking");
1760 // Conditionals pass on only part of their frequency
1761 float prob = n->as_MachIf()->_prob;
1762 assert(prob >= 0.0 && prob <= 1.0, "out of range probability");
1763 // If succ[i] is the FALSE branch, invert path info
1764 if( _nodes[i + eidx + 1]->Opcode() == Op_IfFalse ) {
1765 return 1.0f - prob; // not taken
1766 } else {
1767 return prob; // taken
1768 }
1769 }
1771 case Op_Jump:
1772 // Divide the frequency between all successors evenly
1773 return 1.0f/_num_succs;
1775 case Op_Catch: {
1776 const CatchProjNode *ci = _nodes[i + eidx + 1]->as_CatchProj();
1777 if (ci->_con == CatchProjNode::fall_through_index) {
1778 // Fall-thru path gets the lion's share.
1779 return 1.0f - PROB_UNLIKELY_MAG(5)*_num_succs;
1780 } else {
1781 // Presume exceptional paths are equally unlikely
1782 return PROB_UNLIKELY_MAG(5);
1783 }
1784 }
1786 case Op_Root:
1787 case Op_Goto:
1788 // Pass frequency straight thru to target
1789 return 1.0f;
1791 case Op_NeverBranch:
1792 return 0.0f;
1794 case Op_TailCall:
1795 case Op_TailJump:
1796 case Op_Return:
1797 case Op_Halt:
1798 case Op_Rethrow:
1799 // Do not push out freq to root block
1800 return 0.0f;
1802 default:
1803 ShouldNotReachHere();
1804 }
1806 return 0.0f;
1807 }
1809 //------------------------------num_fall_throughs-----------------------------
1810 // Return the number of fall-through candidates for a block
1811 int Block::num_fall_throughs() {
1812 int eidx = end_idx();
1813 Node *n = _nodes[eidx]; // Get ending Node
1815 int op = n->Opcode();
1816 if (n->is_Mach()) {
1817 if (n->is_MachNullCheck()) {
1818 // In theory, either side can fall-thru, for simplicity sake,
1819 // let's say only the false branch can now.
1820 return 1;
1821 }
1822 op = n->as_Mach()->ideal_Opcode();
1823 }
1825 // Switch on branch type
1826 switch( op ) {
1827 case Op_CountedLoopEnd:
1828 case Op_If:
1829 return 2;
1831 case Op_Root:
1832 case Op_Goto:
1833 return 1;
1835 case Op_Catch: {
1836 for (uint i = 0; i < _num_succs; i++) {
1837 const CatchProjNode *ci = _nodes[i + eidx + 1]->as_CatchProj();
1838 if (ci->_con == CatchProjNode::fall_through_index) {
1839 return 1;
1840 }
1841 }
1842 return 0;
1843 }
1845 case Op_Jump:
1846 case Op_NeverBranch:
1847 case Op_TailCall:
1848 case Op_TailJump:
1849 case Op_Return:
1850 case Op_Halt:
1851 case Op_Rethrow:
1852 return 0;
1854 default:
1855 ShouldNotReachHere();
1856 }
1858 return 0;
1859 }
1861 //------------------------------succ_fall_through-----------------------------
1862 // Return true if a specific successor could be fall-through target.
1863 bool Block::succ_fall_through(uint i) {
1864 int eidx = end_idx();
1865 Node *n = _nodes[eidx]; // Get ending Node
1867 int op = n->Opcode();
1868 if (n->is_Mach()) {
1869 if (n->is_MachNullCheck()) {
1870 // In theory, either side can fall-thru, for simplicity sake,
1871 // let's say only the false branch can now.
1872 return _nodes[i + eidx + 1]->Opcode() == Op_IfFalse;
1873 }
1874 op = n->as_Mach()->ideal_Opcode();
1875 }
1877 // Switch on branch type
1878 switch( op ) {
1879 case Op_CountedLoopEnd:
1880 case Op_If:
1881 case Op_Root:
1882 case Op_Goto:
1883 return true;
1885 case Op_Catch: {
1886 const CatchProjNode *ci = _nodes[i + eidx + 1]->as_CatchProj();
1887 return ci->_con == CatchProjNode::fall_through_index;
1888 }
1890 case Op_Jump:
1891 case Op_NeverBranch:
1892 case Op_TailCall:
1893 case Op_TailJump:
1894 case Op_Return:
1895 case Op_Halt:
1896 case Op_Rethrow:
1897 return false;
1899 default:
1900 ShouldNotReachHere();
1901 }
1903 return false;
1904 }
1906 //------------------------------update_uncommon_branch------------------------
1907 // Update the probability of a two-branch to be uncommon
1908 void Block::update_uncommon_branch(Block* ub) {
1909 int eidx = end_idx();
1910 Node *n = _nodes[eidx]; // Get ending Node
1912 int op = n->as_Mach()->ideal_Opcode();
1914 assert(op == Op_CountedLoopEnd || op == Op_If, "must be a If");
1915 assert(num_fall_throughs() == 2, "must be a two way branch block");
1917 // Which successor is ub?
1918 uint s;
1919 for (s = 0; s <_num_succs; s++) {
1920 if (_succs[s] == ub) break;
1921 }
1922 assert(s < 2, "uncommon successor must be found");
1924 // If ub is the true path, make the proability small, else
1925 // ub is the false path, and make the probability large
1926 bool invert = (_nodes[s + eidx + 1]->Opcode() == Op_IfFalse);
1928 // Get existing probability
1929 float p = n->as_MachIf()->_prob;
1931 if (invert) p = 1.0 - p;
1932 if (p > PROB_MIN) {
1933 p = PROB_MIN;
1934 }
1935 if (invert) p = 1.0 - p;
1937 n->as_MachIf()->_prob = p;
1938 }
1940 //------------------------------update_succ_freq-------------------------------
1941 // Update the appropriate frequency associated with block 'b', a successor of
1942 // a block in this loop.
1943 void CFGLoop::update_succ_freq(Block* b, float freq) {
1944 if (b->_loop == this) {
1945 if (b == head()) {
1946 // back branch within the loop
1947 // Do nothing now, the loop carried frequency will be
1948 // adjust later in scale_freq().
1949 } else {
1950 // simple branch within the loop
1951 b->_freq += freq;
1952 }
1953 } else if (!in_loop_nest(b)) {
1954 // branch is exit from this loop
1955 BlockProbPair bpp(b, freq);
1956 _exits.append(bpp);
1957 } else {
1958 // branch into nested loop
1959 CFGLoop* ch = b->_loop;
1960 ch->_freq += freq;
1961 }
1962 }
1964 //------------------------------in_loop_nest-----------------------------------
1965 // Determine if block b is in the receiver's loop nest.
1966 bool CFGLoop::in_loop_nest(Block* b) {
1967 int depth = _depth;
1968 CFGLoop* b_loop = b->_loop;
1969 int b_depth = b_loop->_depth;
1970 if (depth == b_depth) {
1971 return true;
1972 }
1973 while (b_depth > depth) {
1974 b_loop = b_loop->_parent;
1975 b_depth = b_loop->_depth;
1976 }
1977 return b_loop == this;
1978 }
1980 //------------------------------scale_freq-------------------------------------
1981 // Scale frequency of loops and blocks by trip counts from outer loops
1982 // Do a top down traversal of loop tree (visit outer loops first.)
1983 void CFGLoop::scale_freq() {
1984 float loop_freq = _freq * trip_count();
1985 _freq = loop_freq;
1986 for (int i = 0; i < _members.length(); i++) {
1987 CFGElement* s = _members.at(i);
1988 float block_freq = s->_freq * loop_freq;
1989 if (g_isnan(block_freq) || block_freq < MIN_BLOCK_FREQUENCY)
1990 block_freq = MIN_BLOCK_FREQUENCY;
1991 s->_freq = block_freq;
1992 }
1993 CFGLoop* ch = _child;
1994 while (ch != NULL) {
1995 ch->scale_freq();
1996 ch = ch->_sibling;
1997 }
1998 }
2000 // Frequency of outer loop
2001 float CFGLoop::outer_loop_freq() const {
2002 if (_child != NULL) {
2003 return _child->_freq;
2004 }
2005 return _freq;
2006 }
2008 #ifndef PRODUCT
2009 //------------------------------dump_tree--------------------------------------
2010 void CFGLoop::dump_tree() const {
2011 dump();
2012 if (_child != NULL) _child->dump_tree();
2013 if (_sibling != NULL) _sibling->dump_tree();
2014 }
2016 //------------------------------dump-------------------------------------------
2017 void CFGLoop::dump() const {
2018 for (int i = 0; i < _depth; i++) tty->print(" ");
2019 tty->print("%s: %d trip_count: %6.0f freq: %6.0f\n",
2020 _depth == 0 ? "Method" : "Loop", _id, trip_count(), _freq);
2021 for (int i = 0; i < _depth; i++) tty->print(" ");
2022 tty->print(" members:", _id);
2023 int k = 0;
2024 for (int i = 0; i < _members.length(); i++) {
2025 if (k++ >= 6) {
2026 tty->print("\n ");
2027 for (int j = 0; j < _depth+1; j++) tty->print(" ");
2028 k = 0;
2029 }
2030 CFGElement *s = _members.at(i);
2031 if (s->is_block()) {
2032 Block *b = s->as_Block();
2033 tty->print(" B%d(%6.3f)", b->_pre_order, b->_freq);
2034 } else {
2035 CFGLoop* lp = s->as_CFGLoop();
2036 tty->print(" L%d(%6.3f)", lp->_id, lp->_freq);
2037 }
2038 }
2039 tty->print("\n");
2040 for (int i = 0; i < _depth; i++) tty->print(" ");
2041 tty->print(" exits: ");
2042 k = 0;
2043 for (int i = 0; i < _exits.length(); i++) {
2044 if (k++ >= 7) {
2045 tty->print("\n ");
2046 for (int j = 0; j < _depth+1; j++) tty->print(" ");
2047 k = 0;
2048 }
2049 Block *blk = _exits.at(i).get_target();
2050 float prob = _exits.at(i).get_prob();
2051 tty->print(" ->%d@%d%%", blk->_pre_order, (int)(prob*100));
2052 }
2053 tty->print("\n");
2054 }
2055 #endif