GCC Middle and Back End API Reference
tree-cfgcleanup.c File Reference


static bool remove_fallthru_edge ()
static bool cleanup_control_expr_graph ()
static bool cleanup_control_flow_bb ()
static bool tree_forwarder_block_p ()
static bool phi_alternatives_equal ()
static bool remove_forwarder_block ()
bool fixup_noreturn_call ()
static bool split_bbs_on_noreturn_calls ()
static bool cleanup_tree_cfg_bb ()
static bool cleanup_tree_cfg_1 ()
static bool cleanup_tree_cfg_noloop ()
static void repair_loop_structures ()
bool cleanup_tree_cfg ()
static bool remove_forwarder_block_with_phi ()
static unsigned int merge_phi_nodes ()
static bool gate_merge_phi ()
gimple_opt_passmake_pass_merge_phi ()
static unsigned int execute_cleanup_cfg_post_optimizing ()
gimple_opt_passmake_pass_cleanup_cfg_post_optimizing ()


bitmap cfgcleanup_altered_bbs

Function Documentation

static bool cleanup_control_expr_graph ( )
   Disconnect an unreachable block in the control expression starting
   at block BB.  
         Remove all the edges except the one that is always executed.  
static bool cleanup_control_flow_bb ( )
   Try to remove superfluous control structures in basic block BB.  Returns
   true if anything changes.  
     If the last statement of the block could throw and now cannot,
     we need to prune cfg.  
         If we had a computed goto which has a compile-time determinable
         destination, then we can eliminate the goto.  
         First look at all the outgoing edges.  Delete any outgoing
         edges which do not go to the right block.  For the one
         edge which goes to the right block, fix up its flags.  
                 Turn off the EDGE_ABNORMAL flag.  
                 And set EDGE_FALLTHRU.  
         Remove the GOTO_EXPR as it is not needed.  The CFG has all the
         relevant information we need.  
     Check for indirect calls that have been turned into
     noreturn calls.  
bool cleanup_tree_cfg ( void  )
   Cleanup cfg and repair loop structures.  
static bool cleanup_tree_cfg_1 ( )
   Iterate the cfg cleanups, while anything changes.  
     Prepare the worklists of altered blocks.  
     During forwarder block cleanup, we may redirect edges out of
     SWITCH_EXPRs, which can get expensive.  So we want to enable
     recording of edge to CASE_LABEL_EXPR.  
     Start by iterating over all basic blocks.  We cannot use FOR_EACH_BB,
     since the basic blocks may get removed.  
     Now process the altered blocks, as long as any are available.  
         Rerun split_bbs_on_noreturn_calls, in case we have altered any noreturn

References calculate_dominance_info(), CDI_DOMINATORS, changed, compact_blocks(), delete_unreachable_blocks(), dom_info_available_p(), LOOPS_NEED_FIXUP, loops_state_set(), timevar_pop(), timevar_push(), verify_dominators(), and verify_flow_info().

static bool cleanup_tree_cfg_bb ( )
   Tries to cleanup cfg in basic block BB.  Returns true if anything
     Merging the blocks may create new opportunities for folding
     conditional branches (due to the elimination of single-valued PHI
static bool cleanup_tree_cfg_noloop ( )
   Remove unreachable blocks and other miscellaneous clean up work.
   Return true if the flowgraph was modified, false otherwise.  
     Iterate until there are no more cleanups left to do.  If any
     iteration changed the flowgraph, set CHANGED to true.

     If dominance information is available, there cannot be any unreachable
static unsigned int execute_cleanup_cfg_post_optimizing ( )
   Pass: cleanup the CFG just before expanding trees to RTL.
   This is just a round of label cleanups and case node grouping
   because after the tree optimizers have run such cleanups may
   be necessary.  
bool fixup_noreturn_call ( )
   STMT is a call that has been discovered noreturn.  Fixup the CFG
   and remove LHS.  Return true if something changed.  
     First split basic block if stmt is not last.  
     If there is LHS, remove it.  
         We need to remove SSA name to avoid checking errors.
         All uses are dominated by the noreturn and thus will
         be removed afterwards.
         We proactively remove affected non-PHI statements to avoid
         fixup_cfg from trying to update them and crashing.  
     Similarly remove VDEF if there is any.  

References cfun, changed, gimple_bb(), gimple_call_noreturn_p(), function::gimple_df, basic_block_def::index, and vec_safe_length().

static bool gate_merge_phi ( )
gimple_opt_pass* make_pass_cleanup_cfg_post_optimizing ( )
gimple_opt_pass* make_pass_merge_phi ( )
static unsigned int merge_phi_nodes ( )

This pass merges PHI nodes if one feeds into another. For example, suppose we have the following:

goto <bb 9> (<L9>);

<L8>:; tem_17 = foo ();

tem_6 = PHI <tem_17(8), tem_23(7)>;


tem_3 = PHI <tem_6(9), tem_2(5)>;


Then we merge the first PHI node into the second one like so:

goto <bb 9> (<L10>);

<L8>:; tem_17 = foo ();

tem_3 = PHI <tem_23(7), tem_2(5), tem_17(8)>;


     Find all PHI nodes that we may be able to merge.  
         Look for a forwarder block with PHI nodes.  
         We have to feed into another basic block with PHI
             We don't want to deal with a basic block with
             abnormal edges.  
             If BB does not dominate DEST, then the PHI nodes at
             DEST must be the only users of the results of the PHI
             nodes at BB.  
             BB dominates DEST.  There may be many users of the PHI
             nodes in BB.  However, there is still a trivial case we
             can handle.  If the result of every PHI in BB is used
             only by a PHI in DEST, then we can trivially merge the
             PHI nodes from BB into DEST.  
                 If the PHI's result is never used, then we can just
                 ignore it.  
                 Get the single use of the result of this PHI node.  
             If the loop above iterated through all the PHI nodes
             in BB, then we can merge the PHIs from BB into DEST.  
     Now let's drain WORKLIST.  
     Removing forwarder blocks can cause formerly irreducible loops
     to become reducible if we merged two entry blocks.  

References remove_forwarder_block_with_phi().

static bool phi_alternatives_equal ( )
   If all the PHI nodes in DEST have alternatives for E1 and E2 and
   those alternatives are equal in each of the PHI nodes, then return
   true, else return false.  

References edge_def::dest, first_stmt(), gimple_label_label(), and single_succ_edge().

Referenced by remove_forwarder_block().

static bool remove_fallthru_edge ( )
   Remove any fallthru edge from EV.  Return true if an edge was removed.  

References edge_def::flags, and remove_edge_and_dominated_blocks().

static bool remove_forwarder_block ( )
   Removes forwarder block BB.  Returns false if this failed.  
     We check for infinite loops already in tree_forwarder_block_p.
     However it may happen that the infinite loop is created
     afterwards due to removal of forwarders.  
     If the destination block consists of a nonlocal label or is a
     EH landing pad, do not merge it.  
     If there is an abnormal edge to basic block BB, but not into
     dest, problems might occur during removal of the phi node at out
     of ssa due to overlapping live ranges of registers.

     If there is an abnormal edge in DEST, the problems would occur
     anyway since cleanup_dead_labels would then merge the labels for
     two different eh regions, and rest of exception handling code
     does not like it.

     So if there is an abnormal edge to BB, proceed only if there is
     no abnormal edge to DEST and there are no phi nodes in DEST.  
     If there are phi nodes in DEST, and some of the blocks that are
     predecessors of BB are also predecessors of DEST, check that the
     phi node arguments match.  
     Redirect the edges.  
             If there is an abnormal edge, redirect it anyway, and
             move the labels to the new block to make it legal.  
             Create arguments for the phi nodes, since the edge was not
             here before.  
     Move nonlocal labels and computed goto targets as well as user
     defined labels and labels with an EH landing pad number to the
     new block, so that the redirection of the abnormal edges works,
     jump targets end up in a sane place and debug information for
     labels is retained.  
     Move debug statements if the destination has a single predecessor.  
     Update the dominators.  
             Shortcut to avoid calling (relatively expensive)
             nearest_common_dominator unless necessary.  
     And kill the forwarder block.  

References find_edge(), phi_alternatives_equal(), basic_block_def::preds, and edge_def::src.

static bool remove_forwarder_block_with_phi ( )
   Tries to merge the PHI nodes at BB into those at BB's sole successor.
   Returns true if successful.  
     We check for infinite loops already in tree_forwarder_block_p.
     However it may happen that the infinite loop is created
     afterwards due to removal of forwarders.  
     If the destination block consists of a nonlocal label, do not
     merge it.  
     Redirect each incoming edge to BB to DEST.  
             We already have an edge S from E->src to DEST.  If S and
             E->dest's sole successor edge have the same PHI arguments
             at DEST, redirect S to DEST.  
             PHI arguments are different.  Create a forwarder block by
             splitting E so that we can merge PHI arguments on E to
         redirect_edge_and_branch must not create a new edge.  
         Add to the PHI nodes at DEST each PHI argument removed at the
         destination of E.  
                 If DEF is one of the results of PHI nodes removed during
                 redirection, replace it with the PHI argument that used
                 to be on E.  
     Update the dominators.  
         Shortcut to avoid calling (relatively expensive)
         nearest_common_dominator unless necessary.  
     Remove BB since all of BB's incoming edges have been redirected
     to DEST.  

References redirect_edge_var_map_def(), redirect_edge_var_map_location(), redirect_edge_var_map_result(), and redirect_edge_var_map_vector().

Referenced by merge_phi_nodes().

static void repair_loop_structures ( )
   Repairs loop structures.  
     This usually does nothing.  But sometimes parts of cfg that originally
     were inside a loop get out of it due to edge removal (since they
     become unreachable by back edges from latch).  Also a former
     irreducible loop can become reducible - in this case force a full
     rewrite into loop-closed SSA form.  
static bool split_bbs_on_noreturn_calls ( )
   Split basic blocks on calls in the middle of a basic block that are now
   known not to return, and remove the unreachable code.  
     Detect cases where a mid-block call is now known not to return.  
           BB might be deleted at this point, so verify first
           BB is present in the cfg.  

References merge_blocks(), and single_succ().

static bool tree_forwarder_block_p ( )
   Return true if basic block BB does nothing except pass control
   flow to another block and that we can safely insert a label at
   the start of the successor block.

   As a precondition, we require that BB be not equal to
     BB must have a single outgoing edge.  
         If PHI_WANTED is false, BB must not have any PHI nodes.
         Otherwise, BB must have PHI nodes.  
         BB may not be a predecessor of EXIT_BLOCK_PTR.  
         Nor should this be an infinite loop.  
         BB may not have an abnormal outgoing edge.  
     There should not be an edge coming from entry, or an EH edge.  
         If goto_locus of any of the edges differs, prevent removing
         the forwarder block for -O0.  
     Now walk through the statements backward.  We can ignore labels,
     anything else means this is not a forwarder block.  
             ??? For now, hope there's a corresponding debug
             assignment at the destination.  
         Protect loop latches, headers and preheaders.  

Variable Documentation

bitmap cfgcleanup_altered_bbs

CFG cleanup for trees. Copyright (C) 2001-2013 Free Software Foundation, Inc.

This file is part of GCC.

GCC is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version.

GCC is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with GCC; see the file COPYING3. If not see http://www.gnu.org/licenses/.

   The set of blocks in that at least one of the following changes happened:
   -- the statement at the end of the block was changed
   -- the block was newly created
   -- the set of the predecessors of the block changed
   -- the set of the successors of the block changed
   ??? Maybe we could track these changes separately, since they determine
       what cleanups it makes sense to try on the block.