The Machine-SUIF Control Flow Graph Library

Release version 2.00.11.30

Glenn Holloway and Michael D. Smith
{holloway,smith}@eecs.harvard.edu
Division of Engineering and Applied Sciences
Harvard University

Abstract

The Machine-SUIF CFG library provides an abstraction of control flow graphs with nodes containing lists of machine instructions. Such a CFG can be used to replace the linear instruction-list representation of the body of a procedure being compiled. The library supports translation of the linear representation to and from CFG form, transformation of a CFG by adding, reconnecting, and removing nodes, and fine-grained control over individual program instructions within CFG nodes. It allows precise control of program layout, to ensure that re-linearization leaves instructions in the desired order.

The CFG library is lean by design. We have factored tools for control-flow and data-flow analysis into separate libraries, based on this one. Users can pick the functionality they need, and they can build other CFG-based tools without incurring overhead for features they don't need.



Introduction

This document discusses the Machine-SUIF control flow graph (CFG) library. It discusses design goals, the interface to the library, implementation decisions, current status, and planned updates.

The Machine-SUIF CFG library presents an abstraction of control flow graphs built on the basic instr_list structure of the Machine-SUIF system. It allows you to build CFGs and to transform them by rearranging and reconnecting nodes. Machine SUIF provides companion libraries based on this one that support control-flow and data-flow analysis.

The CFG library consists of files in the cfg subdirectory of the machsuif distribution package. This document describes its design and use. It assumes that the reader is familiar with the SUIF and Machine-SUIF systems and has read the machine library documentation.

The following section discusses the development and research goals that drove the design of our CFG library. Section [->] describes our view of how the library will be used. Sections [->] through [->] describe the programming interface provided by the CFG library, and they provide some notes on its implementation.

Goals for the CFG Library

[*]

This section lists our goals for the CFG library.

The next section gives an overview of how we expect the library to be used.

Philosophy

[*]

One of the things that makes the control-flow graph such a useful compile-time abstraction is that it abstracts away from sequential code. It enables graph-based transformations and allows them to be expressed without reference to a linear code layout. By parceling a procedure's instruction list into basic blocks, we make it efficient to reorder the blocks without having to splice their instructions into a new linear order.

The question arises whether to manage the control instructions within a CFG so that at all times they correctly reflect some linear order. Take the code for a simple if ... then ... else ... statement. Suppose the code generator has produced a conditional branch instruction to the else statement, and has place the then statement after the if test. The then code ends with a jump around the else statement to the join point. Breaking basic blocks after control instructions will put a jump at the end of the node for the then statement, and none at the end of that for the else statement, which just falls through. If a transformation now inverts the sense of the conditional branch and adjusts CFG flow edges accordingly in the CFG, what should be done to the instruction lists in the then and else nodes? The natural re-linearization algorithm would remove the jump from the end of the then node and insert one at the end of the else node. Should such jump adjustments be made every time the CFG changes?

For many kinds of optimization, this isn't important, either because they change no control flow at all, or because they are not concerned with the precise schedule of instructions. For these, it's fine to leave the selection of a final layout and the adjustment of jump instructions to the library method that re-linearizes the flow graph.

For other optimizations, like instruction scheduling and code placement, it's essential to know exactly what instructions will appear where, and in fact, to be able to control such decisions precisely.

The Machine-SUIF CFG library allows you to control the layout relationship between any pair of nodes, but it also allows you to leave the layout of part or all of a CFG unspecified. At the time two nodes become constrained to be adjacent in memory order, the library may remove or insert a trailing jump instruction to achieve consistency with this constraint. But in the absence of layout constraints, it doesn't insert or remove jumps as the graph is transformed.

To implement this policy efficiently, the library must have quick access to the control-transfer instruction (CTI), if any, in each node. It relies on there being at most one CTI, and it keeps a record of its position. As a client, you're free to modify any of the instructions in a block, but when you change the CTI, you're expected to call a library method to reflect the change. [The library doesn't assume that CTI instruction, will be the last in the node; that would make it difficult, for example, to support scheduling for delayed-branch architectures by placing instructions after the control instruction of a block. The library doesn't provide full support for such architectures; e.g., when building a CFG, it always breaks basic blocks at control transfer instructions. Completing the support for target code with delay slots following branch instructions would require subclassing to perform the necessary construction-time adjustments.] Common operations on nodes, such as replacing one of its control successors, update the CTI directly, without requiring the programmer to make changes at both the graph level and the instruction level.

We have tried to keep the CFG library, and particularly its class interfaces, lean. Though Machine SUIF provides flow analysis libraries based on our CFG abstraction, users may want to write their own. We've tried to make it easy to do so.

Class Cfg

[*]

The two main classes used in the CFG library are Cfg and CfgNode. A Cfg represents an entire control flow graph for a procedure. The nodes within a Cfg have class CfgNode. The functions operating on class Cfg handle the graph overall: graph expansion and simplification, graph validation and printing, and the enumeration of nodes and edges. Those operating on CfgNode are about managing a node's relations with its control and layout neighbors and out keeping its code consistent with those relations.

CFG construction

To create a CFG or to normalize the form of an existing CFG, use the following functions:

<Cfg creation and normalization>= (U->)
Cfg* new_cfg(InstrList*, bool keep_layout    = false,
                         bool break_at_call  = false,
                         bool break_at_instr = false);
void canonicalize(Cfg*,  bool keep_layout    = false,
                         bool break_at_call  = false,
                         bool break_at_instr = false);
void remove_layout_info(Cfg*);

Creation function new_cfg parses a linear list of instructions into basic blocks. There are options for determining precisely what constitutes a basic block and whether the original linear layout of the instructions should be reflected by initial layout constraints. Given an existing CFG, you may want to ensure that it has the same form as if built from scratch with a given option set. The canonicalize function serves this purpose.

Each of the flags keep_layout, break_at_call, and break_at_instr is optional. When keep_layout is true, the CFG gets node-layout constraints reflecting the program's initial linear layout. When break_at_call is true, each procedure-call instruction is treated as a CTI, so that it terminates the CFG node that contains it. break_at_instr means that each node of the graph contains only one instruction.

The remove_layout_info function clears all layout constraints in a given graph.

CFG inspection

<Cfg inspectors>= (U->) [D->]
CfgNode* get_node(Cfg*, int pos);
CfgNode* get_node(Cfg*, CfgNodeHandle);
CfgNode* node_at(Cfg*, LabelSym*);

int nodes_size(Cfg*);
CfgNodeHandle nodes_start(Cfg*);
CfgNodeHandle nodes_last(Cfg*);
CfgNodeHandle nodes_end(Cfg*);

CfgNode* get_entry_node(Cfg*);
CfgNode* get_exit_node(Cfg*);

void set_entry_node(Cfg*, CfgNode*);
void set_exit_node(Cfg*, CfgNode*);

Their capsule summaries:

get_node(cfg, pos) returns the node at pos in cfg's node list, where pos is a position in the list (either a zero-based integer or a handle).
get_node(cfg, label) returns the node of cfg at label.
nodes_size(cfg) returns the number of nodes in cfg.
nodes_start(cfg) returns a handle on the first element of cfg.
nodes_last(cfg) returns a handle on the last element of cfg.
nodes_end(cfg) returns the sentinel handle for cfg.
get_entry_node(cfg) returns cfg's entry node.
get_exit_node(cfg) returns cfg's exit node.
set_entry_node(cfg, entry) sets cfg's entry node to entry.
set_exit_node(cfg, exit) sets cfg's exit node to exit.

Type CfgNodeHandle is an abbreviation for an STL iterator.

<CfgNodeHandle definition>= (U->)
typedef list<CfgNode*>::iterator CfgNodeHandle;

Nicknames
The nodes_ functions are so named because the sequence of node pointers in a Cfg is called nodes. These functions are frequently used; since a Cfg instance contains only one such sequence, we can provide shorter names for these handle-related functions without ambiguity.

<Cfg inspectors>+= (U->) [<-D]

inline int
size(Cfg *cfg)
{
    return nodes_size(cfg);
}

inline CfgNodeHandle
start(Cfg *cfg)
{
    return nodes_start(cfg);
}

inline CfgNodeHandle
end(Cfg *cfg)
{
    return nodes_end(cfg);
}

Graph simplification

The following functions perform graph simplifications. Each returns true if it succeeds in changing the graph. Since one kind of simplification can create the opportunity for others, you may want to repeat these transformations until no progress is made by any of them.

<Cfg simplification>= (U->)
bool remove_unreachable_nodes(Cfg*);
bool merge_node_sequences(Cfg*);
bool optimize_jumps(Cfg*);

The function remove_unreachable_nodes(cfg) removes nodes unreachable from the entry of cfg; it renumbers the CFG and thus may invalidate node numbers that you may have saved. The function merge_node_sequences(cfg) merges sequences of control-equivalent nodes in cfg, while optimize_jumps(cfg) eliminates jumps to jumps in cfg.

Graph printing

When debugging code that uses a CFG, it is often useful to be able to print an ASCII representation of the graph.

<Cfg printing>= (U->)
void fprint(FILE*, Cfg*, bool show_layout, bool show_code);

The optional flag show_layout causes the printout to indicate the layout predecessor and/or layout successor of any node that has such constraints. The optional flag show_code causes the instructions in each node to be printed. Both flags default to false.

Header file graph.h

Module graph has the following header file.

<graph.h>=
/* file "cfg/graph.h" -- Control Flow Graph */

<Machine-SUIF copyright>

#ifndef CFG_GRAPH_H
#define CFG_GRAPH_H

#include <machine/copyright.h>

#ifndef SUPPRESS_PRAGMA_INTERFACE
#pragma interface "cfg/graph.h"
#endif

#include <machine/machine.h>
#include <cfg/cfg_ir.h>

<CfgNodeHandle definition>

<Cfg creation and normalization>

<Cfg inspectors>

<Cfg simplification>

<Cfg printing>

#endif /* CFG_GRAPH_H */

Class CfgNode

The class CfgNode is the data structure used for the individual nodes in the CFG.

Node Creation

New CFG nodes are created with respect to a particular CFG and are immediately included in the graph's roster of nodes, even though they may have no connection to other nodes.

<CfgNode creation>= (U->)
CfgNode* new_empty_node(Cfg*);
CfgNode* insert_empty_node(Cfg*, CfgNode *tail, CfgNode *head);
CfgNode* insert_empty_node(Cfg*, CfgNode *tail, int succ_pos);
CfgNode* clone_node(Cfg*, CfgNode*);

The function new_empty_node(cfg) returns a new empty and isolated node of cfg. The function insert_empty_node(cfg, tail, head) inserts a new node along edge (tail, head), while insert_empty_node(cfg, tail, succ_pos) inserts a new node between tail and successor number succ_pos. Note that the sequence of successors of a node is an ordered multiset: there may be several edges from node tail to some node head. The second form of insert_empty_node allows you to say exactly which edge to split with a new node. Finally, clone_node(cfg, node) replicates node in cfg, leaving it disconnected.

Node Contents

The main content of a CFG node is a list of instructions. You can scan and edit this list using the same function calls that you'd use for the contents of an InstrList:

<CfgNode bodily functions>= (U->)
int instrs_size(CfgNode*);
InstrHandle instrs_start(CfgNode*);
InstrHandle instrs_last(CfgNode*);
InstrHandle instrs_end(CfgNode*);
InstrHandle prepend(CfgNode*, Instr*);
InstrHandle append(CfgNode*, Instr*);
void replace(CfgNode*, InstrHandle, Instr*);
InstrHandle insert_before(CfgNode*, InstrHandle, Instr*);
InstrHandle insert_after(CfgNode*, InstrHandle, Instr*);
Instr* remove(CfgNode*, InstrHandle);

where

instrs_size(node) returns the number of elements in node.
instrs_begin(node) returns a handle on the first element of node.
instrs_end(node) returns the sentinel handle for node.
prepend(node, instr) inserts instr at the beginning of node.
append(node, instr) inserts instr at the end of node.
replace(node, handle, instr) replaces the element at handle in node by instr.
insert_before(node, handle, instr) inserts instr before handle in node.
insert_after(node, handle, instr) inserts instr after handle in node.
remove(node, handle) removes and returns the instruction at handle in node.

As with InstrList, the instr_ functions have nicknames because the are so frequently used.

<CfgNode function nicknames>= (U->)
inline int
size(CfgNode *node)
{
    return instrs_size(node);
}

inline InstrHandle
start(CfgNode *node)
{
    return instrs_start(node);
}

inline InstrHandle
last(CfgNode *node)
{
    return instrs_last(node);
}

inline InstrHandle
end(CfgNode *node)
{
    return instrs_end(node);
}

Graph Node Identification

These OPI functions help identify a given CFG node, by giving a code-label symbol representing the start of the code within it or by printing an ASCII representation of it for debugging.

<CfgNode identification>= (U->)
Cfg* get_parent(CfgNode*);
int get_number(CfgNode*);
LabelSym* get_label(CfgNode*);
void fprint(FILE*, CfgNode*, bool show_code = false);

Function get_parent(node) returns the CFG that owns node. Function get_number(node) returns a non-negative integer identifier that is less than nodes_size(get_parent(node)). The function get_label(node) returns the label of node's leading label instruction, inserting a new label instruction if none was present. The fprint function is used to support printing of the overall CFG. It prints one line containing the node number and the nature of its control-termination instruction, if any, and then its lists of successor and predecessor numbers. If the boolean argument show_code is true, then it prints the node's instructions as well.

Control-Flow Relations

When a CFG is created from a list of instructions, its edges reflect the control paths implied by the control-flow instructions of that list. We call these normal edges. The CFG library also supports the creation of two other kinds of edges: exceptional and impossible.

The assumption usually made when the CFG creator encounters a procedure call instruction is that it returns exactly once for each time the call is executed. In many languages, it is possible to write procedures that may not return in the normal way, or that may return more than once for each call. To reflect such exceptional control paths, we provide a special kind of successor relation, used to describe flow of control from a call site directly to the exit node, or from the entry node to the point immediately after a call. We call such flow exceptional, since it is not the usual thing, and since it is often associated with the raising of an exception. We provide functions for creating and recognizing exceptional edges in a CFG.

Though unusual, exceptional edges are at least possible control paths. On the other hand, there are important analysis techniques that insert completely impossible edges in the CFG [cite bibmorgan]. They may require that every node lies on a path to the exit node, for example. To handle an infinite loop, they insert an impossible edge from one of the loop nodes to the exit. We also provide functions to create and recognize impossible edges.

When the CFG is constructed, impossible edges are included to be sure that every node is reachable from the entry, and every node has a path to the exit. However, no exceptional edges appear in the CFG when it is first constructed. To add them, you call set_exceptional_succ explicitly, as described below.

Finding Node Neighbors

Each CfgNode keeps a list of predecessors and a list of successors reachable by any kind of edge. You access these lists through the following functions.

<CfgNode control-neighbor functions>= (U->) [D->]
int succs_size(CfgNode*);
CfgNodeHandle succs_start(CfgNode*);
CfgNodeHandle succs_end(CfgNode*);

int preds_size(CfgNode*);
CfgNodeHandle preds_start(CfgNode*);
CfgNodeHandle preds_end(CfgNode*);

CfgNode *fall_succ(CfgNode*);
CfgNode *taken_succ(CfgNode*);

succs_size(node) and preds_size(node) return the number of nodes's successor and predecessor nodes, respectively.
succs_start(node) and preds_start(node) return a handle on nodes's first successor and predecessor, respectively.
succs_end(node) and preds_end(node) return the sentinel handle for nodes's successor and predecessor sequences, respectively.
fall_succ(node) and taken_succ(node) return node's first and second successors, respectively.

The sequence of predecessors represents an unordered set, while the sequence of successors corresponds to the target(s) of the node's terminating instruction. The predecessor list is unordered because we could not imagine a useful ordering to implement; it is a set (and not a multiset) because having a set seems a simpler abstraction. The successors sequence is ordered; positions 0 and 1 in the list have special meaning depending on the kind of control instruction that terminates the node. The same target can appear multiple times in the successors list, so that different cases of a multiway branch can jump to the same label. The number of successors of a CfgNode depends on the control instruction that ends the node:

All exceptional and impossible successors occur after the normal successors.

The functions returning a CfgNodeHandle permit individual nodes to be visited in the usual manner. The fall_succ and taken_succ functions are convenient ways to access successors 0 and 1, respectively.

Making new neighbors.

You set the normal successors of a node using one of the following.

<CfgNode control-neighbor functions>+= (U->) [<-D->]
CfgNode* get_pred(CfgNode*, int pos);
CfgNode* get_succ(CfgNode*, int pos);
void set_succ(CfgNode*, int pos, CfgNode *succ);
void set_fall_succ(CfgNode*, CfgNode *succ);
void set_taken_succ(CfgNode*, CfgNode *succ);

The function get_pred returns the n-th predecessor of the node to which it's applied (recall that the predecessors are not ordered). get_succ returns the node corresponding to the n-th successor edge, and different values of n may yield the same successor node.

Calling set_succ makes succ the new n-th successor of node. It disconnects the former n-th successor (if any), and updates the predecessor lists of the old and new successors. If necessary, it also modifies node's CTI, which should be consistent with the existing successor sequence, to reflect the change. For example, if node ends with a conditional branch, then

    set_succ(node, 1, succ);
unlinks any previous ``taken'' successor, sets it to succ, and replaces the target symbol of node's branch instruction with the first label of node succ.

Functions set_fall_succ and set_taken_succ are provided for mnemonic value; they are equivalent to calling set_succ with second argument 0 and 1, respectively.

Note that making node e a successor of the entry node has the effect of making e an entry point of the procedure.

Creating exceptional and impossible edges.

For exceptional and impossible edges, there are special ways of setting a successor node.

<CfgNode control-neighbor functions>+= (U->) [<-D->]
void set_exceptional_succ(CfgNode*, int n, CfgNode *succ);
void set_impossible_succ(CfgNode*, int n, CfgNode *succ);

If necessary, each of the above functions extends the successor sequence of the node to accommodate at least n+1 elements. An edge created by set_exceptional_succ (set_impossible_succ) is marked as exceptional (impossible). Detecting whether a particular numbered successor represents a normal, exceptional, or impossible edge is handled by the following predicates.

<CfgNode control-neighbor functions>+= (U->) [<-D]
bool is_normal_succ(CfgNode*, CfgNode *succ);
bool is_normal_succ(CfgNode*, int pos);
bool is_exceptional_succ(CfgNode*, CfgNode *succ);
bool is_exceptional_succ(CfgNode*, int pos);
bool is_possible_succ(CfgNode*, CfgNode *succ);
bool is_possible_succ(CfgNode*, int pos);
bool is_impossible_succ(CfgNode*, CfgNode *succ);
bool is_impossible_succ(CfgNode*, int pos);
bool is_abnormal_succ(CfgNode*, CfgNode *succ);
bool is_abnormal_succ(CfgNode*, int pos);

The versions of the above predicates that take an integer pos as their second argument test whether the pos-th successor edge is of the indicated kind. A ``possible'' successor is one that is either normal or exceptional, but not impossible. An ``abnormal'' one is either exceptional or impossible.

Testing the control-transfer instruction.

It's often necessary to test whether a node is terminated by a CTI, and if so, what kind.

<CfgNode control-instruction functions>= (U->) [D->]
bool ends_in_cti(CfgNode*);     // has CTI
bool ends_in_ubr(CfgNode*);     // has CTI satisfying is_ubr()
bool ends_in_cbr(CfgNode*);     // has CTI satisfying is_cbr()
bool ends_in_mbr(CfgNode*);     // has CTI satisfying is_mbr()
bool ends_in_call(CfgNode*);    // has CTI satisfying is_call()
bool ends_in_return(CfgNode*);  // has CTI satisfying is_return()

Adjusting the control-transfer instruction.

To this point, we have not described any way of changing the number of successors of a node. The set_succ function diverts an existing edge, and it updates the CTI accordingly. But sometimes, there's no substitute for altering the control instruction yourself, and then asking the library to adjust edges accordingly. Here are the relevant functions.

<CfgNode control-instruction functions>+= (U->) [<-D]
Instr* get_cti(CfgNode*);
InstrHandle get_cti_handle(CfgNode*);

void invert_branch(CfgNode*);

void reflect_cti(CfgNode*, Instr *cti, CfgNode *implicit_succ = NULL);

If the node has a control-transfer instruction, then it is returned by get_cti. Function get_cti_handle is similar, but returns a handle on the CTI's position in the instruction list, or a sentinel handle is there is no CTI.

Function invert_branch changes the polarity of a branch by changing its opcode to the opposite opcode (e.g. changing beq to bne). If the branch has a branch_edge_weights annotation, giving the edge frequencies of the node's two out-edges, then invert_branch swaps these as well.

Note that there is no corresponding set_cti function. To change the CTI, you must change the instruction list directly and then call reflect_cti. This function records the new CTI, which must already have been put into the node and changes the node's flow successors to reflect the new control instruction. Obviously, if that instruction can fall through, the function needs to know what to use as its implicit successor. In that case, pass it a second argument giving the implicit successor node.

reflect_cti preserves the exceptional and impossible successors of the node that it operates on, but it completely replaces the normal successors.. If its second argument is NULL, rather than a CTI instruction, it clears the node's CTI association and makes implicit_succ (its third argument) the node's only normal successor.

Layout relations

[*]

The CFG library supports basic block placement and layout by allowing the user to specify an ordering (partial or total) of the nodes in a CFG and its associated procedure. The ordering is specified by a doubly-linked list that runs through the CfgNodes of the graph. Null pointers in this linked list indicate ``don't care'' orderings, while connected components (this document calls them ``clumps'') of the list will be laid out in order. This gives you complete flexibility between specifying no order at all (all layout links null) to a total order on nodes.

Checking and changing layout constraints.

To read and modify layout information, the library provides the following functions.

<CfgNode layout functions>= (U->)
CfgNode *get_layout_pred(CfgNode*);
CfgNode *get_layout_succ(CfgNode*);

void clear_layout_succ(CfgNode*);
bool set_layout_succ(CfgNode*, CfgNode *succ);
bool set_layout_succ(CfgNode*, LabelSym *succ_label);

A nodes's layout predecessor is returned by get_layout_pred; its layout successor, by get_layout_succ.

All of the functions that change a layout constraint between two nodes are invoked on the first of the two in layout order. To remove a layout constraint, use clear_layout_succ. To establish a layout constraint, use set_layout_succ, supplying the new layout successor as its argument. (For convenience, you can supply a label of the successor node instead.)

The set_layout_succ function is careful about branch polarity and explicit jumps, inverting branches and inserting unconditional jumps when necessary. When it needs to invert a branch, this function calls invert_branch and then returns true.

Effects of layout constraints on code.

When a Cfg is constructed, no instructions are added or removed, whether or not layout constraints are specified. Thus a procedure can be transformed into CFG form and back to instruction-list form with no changes, whether or not the layout-retention option is selected.

When no layout links have been specified, the library treats the CFG as a general graph. You are free to transform it without having to ensure that there is a linearization of its current instructions that is in valid executable order.

On the other hand, the library insists that nodes with layout successors be valid standard basic blocks. This makes set_layout_succ picky about the conditions under which it allows a link to be set.

Bounding a node's computation.

It is sometimes useful to find the first ``real'' instruction in a node, i.e., the first one other than a label or some other kind of marker. It's occasionally useful to identify the last non-control-transfer instruction in a node. The following functions identify such ``boundary'' instructions within a node.

<CfgNode boundary functions>= (U->)
Instr *first_non_label(CfgNode*);
Instr *first_active_instr(CfgNode*);
Instr *last_non_cti(CfgNode*);

Function first_non_label returns the first instruction in a node that isn't a label. first_active_instr is similar, but skips pseudo-ops and null instructions in addition to labels. Function last_non_cti returns the last instruction before the terminating control instruction in the node. Each of these boundary-instruction functions returns the null pointer if there is no qualifying instruction.

Header file for module node.h

The CfgNode class and support data structures are defined in the module node, which has the following header file:

<node.h>=
/* file "cfg/node.h" -- Control Flow Graph Nodes */

<Machine-SUIF copyright>

#ifndef CFG_NODE_H
#define CFG_NODE_H

#include <machine/copyright.h>

#ifndef SUPPRESS_PRAGMA_INTERFACE
#pragma interface "cfg/node.h"
#endif

#include <machine/machine.h>
#include <cfg/cfg_ir.h>
#include <cfg/graph.h>

<CfgNode creation>

<CfgNode identification>

<CfgNode bodily functions>

<CfgNode function nicknames>

<CfgNode control-neighbor functions>

<CfgNode control-instruction functions>

<CfgNode layout functions>

<CfgNode boundary functions>

#endif /* CFG_NODE_H */

Utilities

[*]

This section lists a number of helper functions that don't fall into the traditional class hierarchies. These helper functions live in the util.h header file.

Enumeration

[*]

Depth-first walk

The following utility walks a subgraph of the CFG in depth-first order, performing a given action after all successors of a node have already been visited. It can walk either the forward or reverse graph. Its arguments are: the node at which to start; the graph orientation (true for forward); the set of nodes already visited; the node action to be performed; and an optional flag that causes impossible edges to be ignored.

<node enumeration utilities>= (U->)
extern void
    depth_first_walk(CfgNode *start, bool forward, NatSet *visited,
                     DepthFirstWalkAction&, bool not_impossible = false);

The action applied to each node in the walk is expressed using a subclass of the following little class:

<class DepthFirstWalkAction>= (U->)
class DepthFirstWalkAction {
  public:
    virtual ~DepthFirstWalkAction() { }
    virtual void operator()(CfgNode*) { }
};

This class can be used as it is, if no action is needed at each node. Otherwise you specialize it to perform whatever task is needed. The utility in the next section is implemented using a DepthFirstWalkAction that accumulates a list of nodes in reverse postorder.

Reverse postorder sequence

The following class presents the nodes of a CFG in reverse postorder [cite bibmorgan].

<class CfgNodeListRpo>= (U->)
class CfgNodeListRpo
{
  public:
    CfgNodeListRpo(Cfg *graph, bool forward = true);
    virtual ~CfgNodeListRpo();

    int size();
    CfgNodeHandle start();
    CfgNodeHandle end();

    void prepend(CfgNode*);
    void append(CfgNode*);
    void replace(CfgNodeHandle, CfgNode*);
    void insert_before(CfgNodeHandle, CfgNode*);
    void insert_after(CfgNodeHandle, CfgNode*);
    CfgNode* remove(CfgNodeHandle);

<CfgNodeListRpo protected part>
};

The interface of this class is exactly that of any other sequence in Machine SUIF. Its constructor takes a CFG and an optional boolean indicator of graph orientation. The constructor computes the node sequence; you can inspect and even modify the sequence using the usual set of sequence manipulation actions. By default, or when its second argument is true, the constructor operates on the normal forward graph; a false second argument causes it to use the reverse graph instead.

Typically, you bind a reverse-postorder it to a local variable and use the handle abstraction to scan it. The sequence is automatically reclaimed when the scope of the local variable ends. For example:

    {
        CfgNodeListRpo rpo(the_cfg);
        for (CfgNodeHandle h = rpo.start(); h != rpo.end(); ++h) {
            ...                 // use *h to access current node
        }
    }                           // list rpo is reclaimed here
The implementation of class CfgNodeListRpo uses a C++ list of node pointers:

<CfgNodeListRpo protected part>= (<-U)
  protected:
    list<CfgNode*> nodes;

Node Identification

The cfg_node annotation connects a label symbol back to the corresponding CFG node. The following helper function uses that note to return the node for a given label.

<function label_cfg_node>= (U->)
CfgNode *label_cfg_node(Sym *);

To reach the CFG node containing a given instruction, apply this function:

<function get_parent_node>= (U->)
CfgNode* get_parent_node(Instr*);

It returns the containing node, if there is one, or else NULL.

Node-Sequence Utilities

Sometimes it is useful to inquire if a node pred precedes another node in the CFG; has_pred provides this functionality.

<node-sequence utilities>= (U->)
extern bool has_pred(CfgNode *node, CfgNode *pred);

Header file for module

We declare module util with the following header file.

<util.h>=
/* file "cfg/util.h" -- Control Flow Graph Utilities */

<Machine-SUIF copyright>

#ifndef CFG_UTIL_H
#define CFG_UTIL_H

#include <machine/copyright.h>

#ifndef SUPPRESS_PRAGMA_INTERFACE
#pragma interface "cfg/util.h"
#endif

#include <machine/machine.h>

#include <cfg/cfg_ir.h>
#include <cfg/graph.h>

<class DepthFirstWalkAction>

<node enumeration utilities>

<class CfgNodeListRpo>

extern IdString k_cfg_node;

<function label_cfg_node>

<function get_parent_node>

<node-sequence utilities>

#endif /* CFG_UTIL_H */

Library initialization

Before you can start using the facilities of the cfg library, the library must initialize some parts of itself. In SUIF, this is performed by defining an init_libname routine.

<cfg library initialization>= (U->)
extern "C" void init_cfg(SuifEnv* suif_env);

The following string keys the annotation that connects a label to the corresponding CFG node.

<cfg string constants>= (U->)
extern IdString k_cfg_node;

The cfg library initialization header has the following layout:

<init.h>=
/* file "cfg/init.h" */

<Machine-SUIF copyright>

#ifndef CFG_INIT_H
#define CFG_INIT_H

#include <machine/copyright.h>

#ifndef SUPPRESS_PRAGMA_INTERFACE
#pragma interface "cfg/init.h"
#endif

<cfg library initialization>

<cfg string constants>

#endif /* CFG_INIT_H */

Header file for the CFG library

The following is the header file is for use by other libraries and passes that depend on the CFG library. It is never included in any implementation file within the machsuif/cfg directory.

<cfg.h>=
/* file "cfg/cfg.h" */

<Machine-SUIF copyright>

#ifndef CFG_CFG_H
#define CFG_CFG_H

#include <machine/copyright.h>

#include <cfg/cfg_ir.h>
#include <cfg/cfg_ir_factory.h>
#include <cfg/graph.h>
#include <cfg/node.h>
#include <cfg/util.h>
#include <cfg/init.h>

#endif /* CFG_CFG_H */

Hoof Specification of CFG Classes

This section contains the Hoof code that defines our Cfg and CfgNode IR objects in SUIF. We refer you to the SUIF documentation if you want to learn more about Hoof and these definitions.

Class Cfg

<class Cfg>= (U->)
concrete Cfg : AnyBody {
    list<CfgNode* owner> nodes;
    CfgNode* reference entry_node;
    CfgNode* reference exit_node;

    list<LabelSym* reference> noted_labels;

    bool keep_layout;
    bool break_at_call;
    bool break_at_instr;

    CPP_DECLARE
      public:
        list<CfgNode*>& nodes() { return _nodes; }
        InstrList* to_instr_list();
        list<LabelSym*>& noted_labels() { return _noted_labels; }
    CPP_DECLARE
    CPP_BODY
        extern InstrList* cfg_to_instr_list(Cfg*);
        InstrList*
        Cfg::to_instr_list()
        {
            return cfg_to_instr_list(this);
        }
    CPP_BODY
};

Class

<class CfgNode>= (U->)
concrete CfgNode : ScopedObject {
    int number;
    Instr* reference cti;
    list<Instr* owner> instrs;
    list<CfgNode* reference> preds;
    list<CfgNode* reference> succs;
    CfgNode* reference layout_pred;
    CfgNode* reference layout_succ;
    IInteger exc_succs; 
    IInteger imp_succs;

    CPP_DECLARE
      public:
        list<Instr*>& instrs() { return _instrs; }
        list<CfgNode*>& succs() { return _succs; }
        list<CfgNode*>& preds() { return _preds; }

        IInteger& exc_succs() { return _exc_succs; }
        IInteger& imp_succs() { return _imp_succs; }
    CPP_DECLARE
};

Overall hoof specification for module

The combined hoof grammar for the cfg module has the following layout.

<cfg_ir.hoof>=
# file "cfg_ir.hoof"
#
#    Copyright (c) 2000 The President and Fellows of Harvard College
#
#    All rights reserved.
#
#    This software is provided under the ter described in
#    the "machine/copyright.h" include file.


#include "machine/machine_ir.hoof"

module cfg_ir {

    include <bit_vector/bit_vector.h>;
    include <machine/machine.h>;

    import basicnodes;
    import machine;

<class Cfg>

<class CfgNode>
}

Copyright

All of the code is protected by the following copyright notice.

<Machine-SUIF copyright>= (<-U <-U <-U <-U <-U)
/*
    Copyright (c) 2000 The President and Fellows of Harvard College

    All rights reserved.

    This software is provided under the terms described in
    the "machine/copyright.h" include file.
*/

Summary

[*]

We have developed a CFG library that supports not only analysis of programs, but also CFG-level transformations, code layout, and fine-grained code motion. At Harvard, we have used this machine library to build compiler passes including dead code elimination, loop peeling and unrolling, static correlated branch prediction, code layout, register allocation, and a number of instruction schedulers.

Acknowledgments

This document is heavily based on the CFG library in Machine SUIF version 1.1.2 that was done by Cliff Young. Cliff traced the roots of this library all the way back to a CFG library that was written by Mike Smith (one of the co-authors of the current document). A version of this original library was modified by Bob Wilson at Stanford. Tony DeWitt brought the library to Harvard and helped Cliff to adapt the data structure constructors to the Machine-SUIF version 1.1.2 library. Gang Chen wrote the loop analysis code, and Mike built the unified data-flow analysis routines. Other members of the HUBE research group at Harvard contributed useful suggestions to the design. Tim Callahan of Synopsis, Inc. and Berkeley helped to uncover and fix several bugs.

This work is supported in part by an DARPA/NSF infrastructure grant (NDA904-97-C-0225), a NSF Young Investigator award (CCR-9457779), and a NSF Research Infrastructure award (CDA-9401024). We also gratefully acknowledge the generous support of this research by Advanced Micro Devices, Compaq, Digital Equipment, Hewlett-Packard, International Business Machines, Intel, and Microsoft.

References

[1] Cliff Young, David S. Johnson, David R. Karger, and Michael D. Smith. Near-optimal Intraprocedural Branch Alignment. Proc. ACM SIGPLAN 1997 Conference on Programming Language Design and Implementation, pp. 183-193, June 1997.

[2] Nikolas Gloy, Trevor Blackwell, Michael D. Smith, and Brad Calder. Procedure Placement using Temporal Ordering Information. Proc. 30th Annual IEEE/ACM Intl. Symp. on Microarchitecture, pp. 303-313, December 1997.

[3] Robert Morgan. Building an Optimizing Compiler. Digital Press, 1998.

[4] Cliff Young and Michael D. Smith. ``Improving the Accuracy of Static Branch Prediction Using Branch Correlation''. Proc. 6th Intl. Conf. on Architectural Support for Programming Languages and Operating Systems, pp. 232-241, October 1994.