domwalk.c File Reference

#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "basic-block.h"
#include "domwalk.h"
#include "sbitmap.h"

Include dependency graph for domwalk.c:

Functions
static int	cmp_bb_postorder ()

Variables
static int *	bb_postorder

---

Function Documentation

static int cmp_bb_postorder ( )

static

Place higher completion number first (pop off lower number first).

---

Variable Documentation

int* bb_postorder

static

Generic dominator tree walker Copyright (C) 2003-2013 Free Software Foundation, Inc. Contributed by Diego Novillo dnovi.nosp@m.llo@.nosp@m.redha.nosp@m.t.co.nosp@m.m

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 a generic walker for dominator trees.

To understand the dominator walker one must first have a grasp of dominators, immediate dominators and the dominator tree.

Dominators A block B1 is said to dominate B2 if every path from the entry to B2 must pass through B1. Given the dominance relationship, we can proceed to compute immediate dominators. Note it is not important whether or not our definition allows a block to dominate itself.

Immediate Dominators: Every block in the CFG has no more than one immediate dominator. The immediate dominator of block BB must dominate BB and must not dominate any other dominator of BB and must not be BB itself.

Dominator tree: If we then construct a tree where each node is a basic block and there is an edge from each block's immediate dominator to the block itself, then we have a dominator tree.

[ Note this walker can also walk the post-dominator tree, which is defined in a similar manner. i.e., block B1 is said to post-dominate block B2 if all paths from B2 to the exit block must pass through B1. ]

For example, given the CFG

             1
             |
             2
            / \
           3   4
              / \
 +———->5   6
 |          / \ /
 |    +—>8   7
 |    |   /    |
 |    +–9    11
 |      /      |
 +— 10 —> 12

We have a dominator tree which looks like

             1
             |
             2
            / \
           /   \
          3     4
             / / \ \
             | | | |
             5 6 7 12
             |   |
             8   11
             |
             9
             |
            10

The dominator tree is the basis for a number of analysis, transformation and optimization algorithms that operate on a semi-global basis.

The dominator walker is a generic routine which visits blocks in the CFG via a depth first search of the dominator tree. In the example above the dominator walker might visit blocks in the following order 1, 2, 3, 4, 5, 8, 9, 10, 6, 7, 11, 12.

The dominator walker has a number of callbacks to perform actions during the walk of the dominator tree. There are two callbacks which walk statements, one before visiting the dominator children, one after visiting the dominator children. There is a callback before and after each statement walk callback. In addition, the dominator walker manages allocation/deallocation of data structures which are local to each block visited.

The dominator walker is meant to provide a generic means to build a pass which can analyze or transform/optimize a function based on walking the dominator tree. One simply fills in the dominator walker data structure with the appropriate callbacks and calls the walker.

We currently use the dominator walker to prune the set of variables which might need PHI nodes (which can greatly improve compile-time performance in some cases).

We also use the dominator walker to rewrite the function into SSA form which reduces code duplication since the rewriting phase is inherently a walk of the dominator tree.

And (of course), we use the dominator walker to drive our dominator optimizer, which is a semi-global optimizer.

TODO:

Walking statements is based on the block statement iterator abstraction, which is currently an abstraction over walking tree statements. Thus the dominator walker is currently only useful for trees.

Referenced by dom_walker::walk().