2 * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Code Placement. Pins all floating nodes to a block where they
23 * will be executed only if needed.
24 * @author Christian Schaefer, Goetz Lindenmaier, Sebastian Felis,
30 #include "iroptimize.h"
38 * Returns non-zero, is a block is not reachable from Start.
40 * @param block the block to test
42 static int is_Block_unreachable(ir_node *block)
44 return is_Block_dead(block) || get_Block_dom_depth(block) < 0;
48 * Find the earliest correct block for node n. --- Place n into the
49 * same Block as its dominance-deepest Input.
51 * We have to avoid calls to get_nodes_block() here
52 * because the graph is floating.
54 * move_out_of_loops() expects that place_floats_early() have placed
55 * all "living" nodes into a living block. That's why we must
56 * move nodes in dead block with "live" successors into a valid
58 * We move them just into the same block as it's successor (or
59 * in case of a Phi into the effective use block). For Phi successors,
60 * this may still be a dead block, but then there is no real use, as
61 * the control flow will be dead later.
63 * @param n the node to be placed
64 * @param worklist a worklist, predecessors of non-floating nodes are placed here
66 static void place_floats_early(ir_node *n, waitq *worklist)
70 /* we must not run into an infinite loop */
71 assert(!irn_visited(n));
74 /* Place floating nodes. */
75 if (get_irn_pinned(n) == op_pin_state_floats) {
76 ir_node *curr_block = get_nodes_block(n);
77 int in_dead_block = is_Block_unreachable(curr_block);
79 ir_node *b = NULL; /* The block to place this node in */
83 if (is_irn_start_block_placed(n)) {
84 /* These nodes will not be placed by the loop below. */
85 b = get_irg_start_block(current_ir_graph);
89 /* find the block for this node. */
90 irn_arity = get_irn_arity(n);
91 for (i = 0; i < irn_arity; i++) {
92 ir_node *pred = get_irn_n(n, i);
95 if (!irn_visited(pred)
96 && (get_irn_pinned(pred) == op_pin_state_floats)) {
99 * If the current node is NOT in a dead block, but one of its
100 * predecessors is, we must move the predecessor to a live
102 * Such thing can happen, if global CSE chose a node from a
103 * dead block. We move it simply to our block.
104 * Note that neither Phi nor End nodes are floating, so we don't
105 * need to handle them here.
107 if (! in_dead_block) {
108 if (get_irn_pinned(pred) == op_pin_state_floats &&
109 is_Block_unreachable(get_nodes_block(pred)))
110 set_nodes_block(pred, curr_block);
112 place_floats_early(pred, worklist);
116 * A node in the Bad block must stay in the bad block,
117 * so don't compute a new block for it.
122 /* Because all loops contain at least one op_pin_state_pinned node,
123 now all our inputs are either op_pin_state_pinned or
124 place_early() has already been finished on them.
125 We do not have any unfinished inputs! */
126 pred_block = get_nodes_block(pred);
127 if ((!is_Block_dead(pred_block)) &&
128 (get_Block_dom_depth(pred_block) > depth)) {
130 depth = get_Block_dom_depth(pred_block);
132 /* Avoid that the node is placed in the Start block if we are not
133 in the backend phase. */
135 get_Block_dom_depth(get_nodes_block(n)) > 1 &&
136 get_irg_phase_state(current_ir_graph) != phase_backend) {
137 b = get_Block_cfg_out(get_irg_start_block(current_ir_graph), 0);
138 assert(b != get_irg_start_block(current_ir_graph));
143 set_nodes_block(n, b);
147 * Add predecessors of non floating nodes and non-floating predecessors
148 * of floating nodes to worklist and fix their blocks if the are in dead
151 irn_arity = get_irn_arity(n);
155 * Simplest case: End node. Predecessors are keep-alives,
156 * no need to move out of dead block.
158 for (i = -1; i < irn_arity; ++i) {
159 ir_node *pred = get_irn_n(n, i);
160 if (!irn_visited(pred))
161 waitq_put(worklist, pred);
163 } else if (is_Block(n)) {
165 * Blocks: Predecessors are control flow, no need to move
166 * them out of dead block.
168 for (i = irn_arity - 1; i >= 0; --i) {
169 ir_node *pred = get_irn_n(n, i);
170 if (!irn_visited(pred))
171 waitq_put(worklist, pred);
173 } else if (is_Phi(n)) {
175 ir_node *curr_block = get_nodes_block(n);
176 int in_dead_block = is_Block_unreachable(curr_block);
179 * Phi nodes: move nodes from dead blocks into the effective use
180 * of the Phi-input if the Phi is not in a bad block.
182 pred = get_nodes_block(n);
183 if (!irn_visited(pred))
184 waitq_put(worklist, pred);
186 for (i = irn_arity - 1; i >= 0; --i) {
187 ir_node *pred = get_irn_n(n, i);
189 if (!irn_visited(pred)) {
190 if (! in_dead_block &&
191 get_irn_pinned(pred) == op_pin_state_floats &&
192 is_Block_unreachable(get_nodes_block(pred))) {
193 set_nodes_block(pred, get_Block_cfgpred_block(curr_block, i));
195 waitq_put(worklist, pred);
200 ir_node *curr_block = get_nodes_block(n);
201 int in_dead_block = is_Block_unreachable(curr_block);
204 * All other nodes: move nodes from dead blocks into the same block.
206 pred = get_nodes_block(n);
207 if (!irn_visited(pred))
208 waitq_put(worklist, pred);
210 for (i = irn_arity - 1; i >= 0; --i) {
211 ir_node *pred = get_irn_n(n, i);
213 if (!irn_visited(pred)) {
214 if (! in_dead_block &&
215 get_irn_pinned(pred) == op_pin_state_floats &&
216 is_Block_unreachable(get_nodes_block(pred))) {
217 set_nodes_block(pred, curr_block);
219 waitq_put(worklist, pred);
226 * Floating nodes form subgraphs that begin at nodes as Const, Load,
227 * Start, Call and that end at op_pin_state_pinned nodes as Store, Call.
228 * Place_early places all floating nodes reachable from its argument through
229 * floating nodes and adds all beginnings at op_pin_state_pinned nodes to the
232 * @param worklist a worklist, used for the algorithm, empty on in/output
234 static void place_early(waitq *worklist)
237 inc_irg_visited(current_ir_graph);
239 /* this inits the worklist */
240 place_floats_early(get_irg_end(current_ir_graph), worklist);
242 /* Work the content of the worklist. */
243 while (!waitq_empty(worklist)) {
244 ir_node *n = waitq_get(worklist);
246 place_floats_early(n, worklist);
248 set_irg_pinned(current_ir_graph, op_pin_state_pinned);
252 * Compute the deepest common dominator tree ancestor of block and dca.
254 * @param dca the deepest common dominator tree ancestor so far,
256 * @param block a block
258 * @return the deepest common dominator tree ancestor of block and dca
260 static ir_node *calc_dom_dca(ir_node *dca, ir_node *block)
262 assert(block != NULL);
264 /* we do not want to place nodes in dead blocks */
265 if (is_Block_dead(block))
268 /* We found a first legal placement. */
269 if (!dca) return block;
271 /* Find a placement that is dominates both, dca and block. */
272 while (get_Block_dom_depth(block) > get_Block_dom_depth(dca))
273 block = get_Block_idom(block);
275 while (get_Block_dom_depth(dca) > get_Block_dom_depth(block)) {
276 dca = get_Block_idom(dca);
279 while (block != dca) {
280 block = get_Block_idom(block); dca = get_Block_idom(dca);
286 * Deepest common dominance ancestor of DCA and CONSUMER of PRODUCER.
287 * I.e., DCA is the block where we might place PRODUCER.
288 * A data flow edge points from producer to consumer.
290 static ir_node *consumer_dom_dca(ir_node *dca, ir_node *consumer, ir_node *producer)
292 /* Compute the last block into which we can place a node so that it is
294 if (is_Phi(consumer)) {
295 /* our consumer is a Phi-node, the effective use is in all those
296 blocks through which the Phi-node reaches producer */
297 ir_node *phi_block = get_nodes_block(consumer);
298 int arity = get_irn_arity(consumer);
301 for (i = 0; i < arity; i++) {
302 if (get_Phi_pred(consumer, i) == producer) {
303 ir_node *new_block = get_Block_cfgpred_block(phi_block, i);
305 if (!is_Block_unreachable(new_block))
306 dca = calc_dom_dca(dca, new_block);
310 dca = calc_dom_dca(dca, get_nodes_block(consumer));
315 static inline int get_block_loop_depth(ir_node *block)
317 return get_loop_depth(get_irn_loop(block));
321 * Move n to a block with less loop depth than it's current block. The
322 * new block must be dominated by early.
324 * @param n the node that should be moved
325 * @param early the earliest block we can n move to
327 static void move_out_of_loops(ir_node *n, ir_node *early)
333 /* Find the region deepest in the dominator tree dominating
334 dca with the least loop nesting depth, but still dominated
335 by our early placement. */
336 dca = get_nodes_block(n);
339 while (dca != early) {
340 dca = get_Block_idom(dca);
341 if (!dca || is_Bad(dca)) break; /* may be Bad if not reachable from Start */
342 if (get_block_loop_depth(dca) < get_block_loop_depth(best)) {
346 if (best != get_nodes_block(n))
347 set_nodes_block(n, best);
351 * Calculate the deepest common ancestor in the dominator tree of all nodes'
352 * blocks depending on node; our final placement has to dominate DCA.
354 * @param node the definition node
355 * @param dca the deepest common ancestor block so far, initially
358 * @return the deepest common dominator ancestor of all blocks of node's users
360 static ir_node *get_deepest_common_dom_ancestor(ir_node *node, ir_node *dca)
364 for (i = get_irn_n_outs(node) - 1; i >= 0; --i) {
365 ir_node *succ = get_irn_out(node, i);
369 * This consumer is the End node, a keep alive edge.
370 * This is not a real consumer, so we ignore it
376 /* Proj nodes are in the same block as node, so
377 * the users of Proj are our users. */
378 dca = get_deepest_common_dom_ancestor(succ, dca);
380 /* ignore if succ is in dead code */
381 ir_node *succ_blk = get_nodes_block(succ);
382 if (is_Block_unreachable(succ_blk))
384 dca = consumer_dom_dca(dca, succ, node);
391 * Put all the Proj nodes of a node into a given block.
393 * @param node the mode_T node
394 * @param block the block to put the Proj nodes to
396 static void set_projs_block(ir_node *node, ir_node *block)
400 for (i = get_irn_n_outs(node) - 1; i >= 0; --i) {
401 ir_node *succ = get_irn_out(node, i);
403 assert(is_Proj(succ));
405 if (get_irn_mode(succ) == mode_T) {
406 set_projs_block(succ, block);
408 set_nodes_block(succ, block);
413 * Find the latest legal block for N and place N into the
414 * `optimal' Block between the latest and earliest legal block.
415 * The `optimal' block is the dominance-deepest block of those
416 * with the least loop-nesting-depth. This places N out of as many
417 * loops as possible and then makes it as control dependent as
420 * @param n the node to be placed
421 * @param worklist a worklist, all successors of non-floating nodes are
424 static void place_floats_late(ir_node *n, pdeq *worklist)
429 assert(!irn_visited(n)); /* no multiple placement */
433 /* no need to place block nodes, control nodes are already placed. */
436 (get_irn_mode(n) != mode_X)) {
437 /* Remember the early_blk placement of this block to move it
438 out of loop no further than the early_blk placement. */
439 early_blk = get_nodes_block(n);
442 * BEWARE: Here we also get code, that is live, but
443 * was in a dead block. If the node is life, but because
444 * of CSE in a dead block, we still might need it.
447 /* Assure that our users are all placed, except the Phi-nodes.
448 --- Each data flow cycle contains at least one Phi-node. We
449 have to break the `user has to be placed before the
450 producer' dependence cycle and the Phi-nodes are the
451 place to do so, because we need to base our placement on the
452 final region of our users, which is OK with Phi-nodes, as they
453 are op_pin_state_pinned, and they never have to be placed after a
454 producer of one of their inputs in the same block anyway. */
455 for (i = get_irn_n_outs(n) - 1; i >= 0; --i) {
456 ir_node *succ = get_irn_out(n, i);
457 if (!irn_visited(succ) && !is_Phi(succ))
458 place_floats_late(succ, worklist);
461 if (! is_Block_dead(early_blk)) {
462 /* do only move things that where not dead */
463 ir_op *op = get_irn_op(n);
465 /* We have to determine the final block of this node... except for
466 constants and Projs */
467 if ((get_irn_pinned(n) == op_pin_state_floats) &&
469 (op != op_SymConst) &&
472 /* deepest common ancestor in the dominator tree of all nodes'
473 blocks depending on us; our final placement has to dominate
475 ir_node *dca = get_deepest_common_dom_ancestor(n, NULL);
477 set_nodes_block(n, dca);
478 move_out_of_loops(n, early_blk);
479 if (get_irn_mode(n) == mode_T) {
480 set_projs_block(n, get_nodes_block(n));
487 /* Add successors of all non-floating nodes on list. (Those of floating
488 nodes are placed already and therefore are marked.) */
489 n_outs = get_irn_n_outs(n);
490 for (i = 0; i < n_outs; i++) {
491 ir_node *succ = get_irn_out(n, i);
492 if (!irn_visited(succ)) {
493 pdeq_putr(worklist, succ);
499 * Place floating nodes on the given worklist as late as possible using
500 * the dominance tree.
502 * @param worklist the worklist containing the nodes to place
504 static void place_late(waitq *worklist)
507 inc_irg_visited(current_ir_graph);
509 /* This fills the worklist initially. */
510 place_floats_late(get_irg_start_block(current_ir_graph), worklist);
512 /* And now empty the worklist again... */
513 while (!waitq_empty(worklist)) {
514 ir_node *n = waitq_get(worklist);
516 place_floats_late(n, worklist);
520 /* Code Placement. */
521 void place_code(ir_graph *irg)
524 ir_graph *rem = current_ir_graph;
526 current_ir_graph = irg;
527 remove_critical_cf_edges(irg);
529 /* Handle graph state */
530 assert(get_irg_phase_state(irg) != phase_building);
531 assure_irg_outs(irg);
534 if (1 || get_irg_loopinfo_state(irg) != loopinfo_consistent) {
535 free_loop_information(irg);
536 construct_cf_backedges(irg);
539 /* Place all floating nodes as early as possible. This guarantees
540 a legal code placement. */
541 worklist = new_waitq();
542 place_early(worklist);
544 /* Note: place_early changes only blocks, no data edges. So, the
545 * data out edges are still valid, no need to recalculate them here. */
547 /* Now move the nodes down in the dominator tree. This reduces the
548 unnecessary executions of the node. */
549 place_late(worklist);
551 set_irg_outs_inconsistent(irg);
552 set_irg_loopinfo_inconsistent(irg);
554 current_ir_graph = rem;
558 * Wrapper for place_code() inside the place_code pass.
560 static void place_code_wrapper(ir_graph *irg)
562 set_opt_global_cse(1);
563 optimize_graph_df(irg);
565 set_opt_global_cse(0);
568 ir_graph_pass_t *place_code_pass(const char *name)
570 return def_graph_pass(name ? name : "place", place_code_wrapper);