#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "ggc.h"
#include "tree.h"
#include "tm_p.h"
#include "basic-block.h"
#include "gimple-pretty-print.h"
#include "bitmap.h"
#include "gimple.h"
#include "gimple-ssa.h"
#include "tree-cfg.h"
#include "tree-phinodes.h"
#include "ssa-iterators.h"
#include "tree-ssanames.h"
#include "tree-dfa.h"
#include "tree-pass.h"
#include "domwalk.h"
#include "flags.h"
#include "langhooks.h"
#include "tree-cfgcleanup.h"

Include dependency graph for tree-ssa-dse.c:

Data Structures
class	dse_dom_walker

Functions
static bool	gate_dse (void)
static unsigned int	tree_ssa_dse (void)
static bool	dse_possible_dead_store_p ()
static void	dse_optimize_stmt ()
gimple_opt_pass *	make_pass_dse ()

Variables
static bitmap	need_eh_cleanup

Function Documentation

static void dse_optimize_stmt ( )

static

Attempt to eliminate dead stores in the statement referenced by BSI.

A dead store is a store into a memory location which will later be overwritten by another store without any intervening loads. In this case the earlier store can be deleted.

In our SSA + virtual operand world we use immediate uses of virtual operands to detect dead stores. If a store's virtual definition is used precisely once by a later store to the same location which post dominates the first store, then the first store is dead.

 If this statement has no virtual defs, then there is nothing
 to do.

 We know we have virtual definitions.  If this is a GIMPLE_ASSIGN
 that's not also a function call, then record it into our table.

 Don't return early on *this_2(D) ={v} {CLOBBER}.

     But only remove *this_2(D) ={v} {CLOBBER} if killed by
     another clobber stmt.

     If we have precisely one immediate use at this point and the
     stores are to the same memory location or there is a chain of
     virtual uses from stmt and the stmt which stores to that same
     memory location, then we may have found redundant store.

         If use_stmt is or might be a nop assignment, e.g. for
         struct { ... } S a, b, *p; ...
         b = a; b = b;
         or
         b = a; b = *p; where p might be &b,
         or
         *p = a; *p = b; where p might be &b,
         or
         *p = *u; *p = *v; where p might be v, then USE_STMT
         acts as a use as well as definition, so store in STMT
         is not dead.

         Then we need to fix the operand of the consuming stmt.

         Remove the dead store.

         And release any SSA_NAMEs set in this statement back to the
         SSA_NAME manager.

References bitmap_set_bit, dse_possible_dead_store_p(), dump_file, dump_flags, gimple_assign_lhs(), gimple_bb(), gimple_clobber_p(), gimple_get_lhs(), gimple_has_lhs(), gsi_remove(), gsi_stmt(), basic_block_def::index, operand_equal_p(), print_gimple_stmt(), ref_maybe_used_by_stmt_p(), release_defs(), stmt_kills_ref_p(), TDF_DETAILS, and unlink_stmt_vdef().

static bool dse_possible_dead_store_p ( )

static

A helper of dse_optimize_stmt. Given a GIMPLE_ASSIGN in STMT, find a candidate statement *USE_STMT that may prove STMT to be dead. Return TRUE if the above conditions are met, otherwise FALSE.

 Self-assignments are zombies.

 Find the first dominated statement that clobbers (part of) the
 memory stmt stores to with no intermediate statement that may use
 part of the memory stmt stores.  That is, find a store that may
 prove stmt to be a dead store.

     Limit stmt walking to be linear in the number of possibly
     dead stores.

         If we ever reach our DSE candidate stmt again fail.  We
         cannot handle dead stores in loops.

         In simple cases we can look through PHI nodes, but we
         have to be careful with loops and with memory references
         containing operands that are also operands of PHI nodes.
         See gcc.c-torture/execute/20051110-*.c.

                 Make sure we are not in a loop latch block.

                 We can look through PHIs to regions post-dominating
                 the DSE candidate stmt.

             Do not consider the PHI as use if it dominates the 
             stmt defining the virtual operand we are processing,
             we have processed it already in this case.

         If the statement is a use the store is not dead.

         If this is a store, remember it or bail out if we have
         multiple ones (the will be in different CFG parts then).

     If we didn't find any definition this means the store is dead
     if it isn't a store to global reachable memory.  In this case
     just pretend the stmt makes itself dead.  Otherwise fail.

 We deliberately stop on clobbering statements and not only on
 killing ones to make walking cheaper.  Otherwise we can just
 continue walking until both stores have equal reference trees.

Referenced by dse_optimize_stmt().

static bool gate_dse ( void )

static

gimple_opt_pass* make_pass_dse ( )

static unsigned int tree_ssa_dse ( )

static

Main entry point.

 We might consider making this a property of each pass so that it
 can be [re]computed on an as-needed basis.  Particularly since
 this pass could be seen as an extension of DCE which needs post
 dominators.

 Dead store elimination is fundamentally a walk of the post-dominator
 tree and a backwards walk of statements within each block.

 Removal of stores may make some EH edges dead.  Purge such edges from
 the CFG as needed.

 For now, just wipe the post-dominator information.

Variable Documentation

bitmap need_eh_cleanup

static

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/. This file implements dead store elimination.

A dead store is a store into a memory location which will later be overwritten by another store without any intervening loads. In this case the earlier store can be deleted.

In our SSA + virtual operand world we use immediate uses of virtual operands to detect dead stores. If a store's virtual definition is used precisely once by a later store to the same location which post dominates the first store, then the first store is dead.

The single use of the store's virtual definition ensures that there are no intervening aliased loads and the requirement that the second load post dominate the first ensures that if the earlier store executes, then the later stores will execute before the function exits.

It may help to think of this as first moving the earlier store to the point immediately before the later store. Again, the single use of the virtual definition and the post-dominance relationship ensure that such movement would be safe. Clearly if there are back to back stores, then the second is redundant.

Reviewing section 10.7.2 in Morgan's "Building an Optimizing Compiler" may also help in understanding this code since it discusses the relationship between dead store and redundant load elimination. In fact, they are the same transformation applied to different views of the CFG. Bitmap of blocks that have had EH statements cleaned. We should remove their dead edges eventually.

Data Structures

Functions

Variables

Function Documentation

Variable Documentation