1 /* Thread edges through blocks and update the control flow and SSA graphs.
2 Copyright (C) 2004 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2, or (at your option)
11 GCC is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING. If not, write to
18 the Free Software Foundation, 59 Temple Place - Suite 330,
19 Boston, MA 02111-1307, USA. */
23 #include "coretypes.h"
30 #include "basic-block.h"
35 #include "diagnostic.h"
36 #include "tree-flow.h"
37 #include "tree-dump.h"
38 #include "tree-pass.h"
40 /* Given a block B, update the CFG and SSA graph to reflect redirecting
41 one or more in-edges to B to instead reach the destination of an
42 out-edge from B while preserving any side effects in B.
44 i.e., given A->B and B->C, change A->B to be A->C yet still preserve the
45 side effects of executing B.
47 1. Make a copy of B (including its outgoing edges and statements). Call
48 the copy B'. Note B' has no incoming edges or PHIs at this time.
50 2. Remove the control statement at the end of B' and all outgoing edges
53 3. Add a new argument to each PHI in C with the same value as the existing
54 argument associated with edge B->C. Associate the new PHI arguments
57 4. For each PHI in B, find or create a PHI in B' with an identical
58 PHI_RESULT. Add an argument to the PHI in B' which has the same
59 value as the PHI in B associated with the edge A->B. Associate
60 the new argument in the PHI in B' with the edge A->B.
62 5. Change the edge A->B to A->B'.
64 5a. This automatically deletes any PHI arguments associated with the
67 5b. This automatically associates each new argument added in step 4
70 6. Repeat for other incoming edges into B.
72 7. Put the duplicated resources in B and all the B' blocks into SSA form.
74 Note that block duplication can be minimized by first collecting the
75 the set of unique destination blocks that the incoming edges should
76 be threaded to. Block duplication can be further minimized by using
77 B instead of creating B' for one destination if all edges into B are
78 going to be threaded to a successor of B. */
81 /* Main data structure recording information regarding B's duplicate
84 struct redirection_data
86 /* A duplicate of B with the trailing control statement removed and which
87 targets a single successor of B. */
88 basic_block dup_block;
90 /* An outgoing edge from B. DUP_BLOCK will have OUTGOING_EDGE->dest as
91 its single successor. */
95 /* Main data structure to hold information for duplicates of BB. */
96 static varray_type redirection_data;
98 /* Add to the destination of edge E those PHI arguments queued on
102 copy_phis_to_block (edge e)
104 basic_block dest = e->dest;
107 for (var = PENDING_STMT (e); var; var = TREE_CHAIN (var))
109 tree result = TREE_PURPOSE (var);
110 tree arg = TREE_VALUE (var);
113 /* First try to find a PHI node in NEW_BB which has the same
114 PHI_RESULT as the PHI from BB we are currently processing. */
115 for (new_phi = phi_nodes (dest); new_phi;
116 new_phi = PHI_CHAIN (new_phi))
117 if (PHI_RESULT (new_phi) == result)
120 /* If we did not find a suitable PHI in NEW_BB, create one. */
122 new_phi = create_phi_node (result, dest);
124 /* Now add that same argument to the new PHI node in block NEW_BB. */
125 add_phi_arg (&new_phi, arg, e);
128 PENDING_STMT (e) = NULL;
131 /* Remove the last statement in block BB if it is a control statement
132 Also remove all outgoing edges except the edge which reaches DEST_BB.
133 If DEST_BB is NULL, then remove all outgoing edges. */
136 remove_ctrl_stmt_and_useless_edges (basic_block bb, basic_block dest_bb)
138 block_stmt_iterator bsi;
144 /* If the duplicate ends with a control statement, then remove it.
146 Note that if we are duplicating the template block rather than the
147 original basic block, then the duplicate might not have any real
151 && (TREE_CODE (bsi_stmt (bsi)) == COND_EXPR
152 || TREE_CODE (bsi_stmt (bsi)) == SWITCH_EXPR))
155 for (ei = ei_start (bb->succs); (e = ei_safe_edge (ei)); )
157 if (e->dest != dest_bb)
164 /* Create a duplicate of BB which only reaches the destination of the edge
165 stored in RD. Record the duplicate block in RD. */
168 create_block_for_threading (basic_block bb, struct redirection_data *rd)
170 /* We can use the generic block duplication code and simply remove
171 the stuff we do not need. */
172 rd->dup_block = duplicate_block (bb, NULL);
174 /* Zero out the profile, since the block is unreachable for now. */
175 rd->dup_block->frequency = 0;
176 rd->dup_block->count = 0;
178 /* The call to duplicate_block will copy everything, including the
179 useless COND_EXPR or SWITCH_EXPR at the end of BB. We just remove
180 the useless COND_EXPR or SWITCH_EXPR here rather than having a
181 specialized block copier. We also remove all outgoing edges
182 from the duplicate block. The appropriate edge will be created
184 remove_ctrl_stmt_and_useless_edges (rd->dup_block, NULL);
187 /* BB is a block which ends with a COND_EXPR or SWITCH_EXPR and when BB
188 is reached via one or more specific incoming edges, we know which
189 outgoing edge from BB will be traversed.
191 We want to redirect those incoming edges to the target of the
192 appropriate outgoing edge. Doing so avoids a conditional branch
193 and may expose new optimization opportunities. Note that we have
194 to update dominator tree and SSA graph after such changes.
196 The key to keeping the SSA graph update manageable is to duplicate
197 the side effects occurring in BB so that those side effects still
198 occur on the paths which bypass BB after redirecting edges.
200 We accomplish this by creating duplicates of BB and arranging for
201 the duplicates to unconditionally pass control to one specific
202 successor of BB. We then revector the incoming edges into BB to
203 the appropriate duplicate of BB.
205 BB and its duplicates will have assignments to the same set of
206 SSA_NAMEs. Right now, we just call into rewrite_ssa_into_ssa
207 to update the SSA graph for those names.
209 We are also going to experiment with a true incremental update
210 scheme for the duplicated resources. One of the interesting
211 properties we can exploit here is that all the resources set
212 in BB will have the same IDFS, so we have one IDFS computation
213 per block with incoming threaded edges, which can lower the
214 cost of the true incremental update algorithm. */
217 thread_block (basic_block bb)
219 /* E is an incoming edge into BB that we may or may not want to
220 redirect to a duplicate of BB. */
223 basic_block template_block;
225 /* ALL indicates whether or not all incoming edges into BB should
226 be threaded to a duplicate of BB. */
231 VARRAY_GENERIC_PTR_INIT (redirection_data, 2, "redirection data");
233 /* Look at each incoming edge into BB. Record each unique outgoing
234 edge that we want to thread an incoming edge to. Also note if
235 all incoming edges are threaded or not. */
236 FOR_EACH_EDGE (e, ei, bb->preds)
246 /* See if we can find an entry for the destination of this
247 threaded edge that has already been recorded. */
248 for (i = 0; i < VARRAY_ACTIVE_SIZE (redirection_data); i++)
250 struct redirection_data *rd;
253 rd = VARRAY_GENERIC_PTR (redirection_data, i);
256 if (e2->dest == rd->outgoing_edge->dest)
260 /* If the loop did not terminate early, then we have a new
261 destination for the incoming threaded edges. Record it. */
262 if (i == VARRAY_ACTIVE_SIZE (redirection_data))
264 struct redirection_data *rd;
266 rd = ggc_alloc_cleared (sizeof (struct redirection_data));
267 rd->outgoing_edge = e->aux;
268 VARRAY_PUSH_GENERIC_PTR (redirection_data, rd);
273 /* Now create duplicates of BB. Note that if all incoming edges are
274 threaded, then BB is going to become unreachable. In that case
275 we use BB for one of the duplicates rather than wasting memory
276 duplicating BB. Thus the odd starting condition for the loop.
278 Note that for a block with a high outgoing degree we can waste
279 a lot of time and memory creating and destroying useless edges.
281 So we first duplicate BB and remove the control structure at the
282 tail of the duplicate as well as all outgoing edges from the
283 duplicate. We then use that duplicate block as a template for
284 the rest of the duplicates. */
285 template_block = NULL;
286 for (i = (all ? 1 : 0); i < VARRAY_ACTIVE_SIZE (redirection_data); i++)
288 struct redirection_data *rd = VARRAY_GENERIC_PTR (redirection_data, i);
290 if (template_block == NULL)
292 create_block_for_threading (bb, rd);
293 template_block = rd->dup_block;
297 create_block_for_threading (template_block, rd);
301 /* Now created up edges from the duplicate blocks to their new
302 destinations. Doing this as a separate loop after block creation
303 allows us to avoid creating lots of useless edges. */
304 for (i = (all ? 1 : 0); i < VARRAY_ACTIVE_SIZE (redirection_data); i++)
306 struct redirection_data *rd = VARRAY_GENERIC_PTR (redirection_data, i);
310 e = make_edge (rd->dup_block, rd->outgoing_edge->dest, EDGE_FALLTHRU);
312 /* If there are any PHI nodes at the destination of the outgoing edge
313 from the duplicate block, then we will need to add a new argument
314 to them. The argument should have the same value as the argument
315 associated with the outgoing edge stored in RD. */
316 for (phi = phi_nodes (e->dest); phi; phi = PHI_CHAIN (phi))
318 int indx = phi_arg_from_edge (phi, rd->outgoing_edge);
319 add_phi_arg (&phi, PHI_ARG_DEF_TREE (phi, indx), e);
323 /* The loop above created the duplicate blocks (and the statements
324 within the duplicate blocks). This loop creates PHI nodes for the
325 duplicated blocks and redirects the incoming edges into BB to reach
326 the duplicates of BB.
328 Note that redirecting the edge will change e->pred_next, so we have
329 to hold e->pred_next in a temporary.
331 If this turns out to be a performance problem, then we could create
332 a list of incoming edges associated with each entry in
333 REDIRECTION_DATA and walk over that list of edges instead. */
334 for (ei = ei_start (bb->preds); (e = ei_safe_edge (ei)); )
336 edge new_dest = e->aux;
338 /* E was not threaded, then there is nothing to do. */
345 /* Go ahead and clear E->aux. It's not needed anymore and failure
346 to clear it will cause all kinds of unpleasant problems later. */
349 /* We know E is an edge we want to thread. Find the entry associated
350 with E's new destination in the REDIRECTION_DATA array. */
351 for (i = 0; i < VARRAY_ACTIVE_SIZE (redirection_data); i++)
353 struct redirection_data *rd;
355 rd = VARRAY_GENERIC_PTR (redirection_data, i);
357 /* We have found the right entry if the outgoing edge in this
358 entry matches E's new destination. Note that if we have not
359 created a duplicate block (rd->dup_block is NULL), then we
360 are going to re-use BB as a duplicate and we do not need
361 to create PHI nodes or redirect the edge. */
362 if (rd->outgoing_edge == new_dest && rd->dup_block)
366 if (dump_file && (dump_flags & TDF_DETAILS))
367 fprintf (dump_file, " Threaded jump %d --> %d to %d\n",
368 e->src->index, e->dest->index, rd->dup_block->index);
370 e2 = redirect_edge_and_branch (e, rd->dup_block);
371 copy_phis_to_block (e2);
373 if ((dump_file && (dump_flags & TDF_DETAILS))
374 && e->src != e2->src)
375 fprintf (dump_file, " basic block %d created\n",
382 /* If all the incoming edges where threaded, then we used BB as one
383 of the duplicate blocks. We need to fixup BB in that case so that
384 it no longer has a COND_EXPR or SWITCH_EXPR and reaches one destination
388 struct redirection_data *rd;
390 rd = VARRAY_GENERIC_PTR (redirection_data, 0);
392 if (dump_file && (dump_flags & TDF_DETAILS))
393 fprintf (dump_file, " Threaded jump %d --> %d to %d\n",
394 EDGE_PRED (bb, 0)->src->index, bb->index,
395 EDGE_SUCC (bb, 0)->dest->index);
397 remove_ctrl_stmt_and_useless_edges (bb, rd->outgoing_edge->dest);
398 EDGE_SUCC (bb, 0)->flags &= ~(EDGE_TRUE_VALUE | EDGE_FALSE_VALUE);
399 EDGE_SUCC (bb, 0)->flags |= EDGE_FALLTHRU;
402 /* Done with this block. Clear REDIRECTION_DATA. */
403 VARRAY_CLEAR (redirection_data);
406 /* Walk through all blocks and thread incoming edges to the block's
407 destinations as requested. This is the only entry point into this
410 Blocks which have one or more incoming edges have INCOMING_EDGE_THREADED
411 set in the block's annotation.
414 Each edge that should be threaded has the new destination edge stored in
415 the original edge's AUX field.
417 This routine (or one of its callees) will clear INCOMING_EDGE_THREADED
418 in the block annotations and the AUX field in the edges.
420 It is the caller's responsibility to fix the dominance information
421 and rewrite duplicated SSA_NAMEs back into SSA form.
423 Returns true if one or more edges were threaded, false otherwise. */
426 thread_through_all_blocks (void)
433 if (bb_ann (bb)->incoming_edge_threaded)
437 bb_ann (bb)->incoming_edge_threaded = false;