An Overview of the SUIF2 System
Monica Lam
Stanford University
http://suif.stanford.edu/
Team Members
Stanford University
 Monica Lam
 Gerald Aigner
 Gerald Cheong
 Amer Diwan
 David Heine
 Amy Lim
 Vladimir Livshits
 Virendra Mehta
 Brian Murphy
 Costa Sapuntzakis
 Hansel Wan
 Chris Wilson
Harvard
 Michael Smith
 Erven Rohou
 Glenn Holloway
 Gang Chen
 Eric Feigin
 Nick Gloy
 Stuart Schechter
 Jason Simmons
 Omri Traub
 Cliff Young
UCSB
Portland Group
 Urs Hoelzle
 David Moore
 Andrew Duncan  Vince Shuster
 Bogdan Cocosel  Deb Caruso
 Holger Kienle
 David Wohlford
 Kristian Kvilekval  Bob Scollard
 Kevin O’gorman
 Prashanth Ramdas
 Radu Rugina
 Andrew Rutz
The SUIF System
PGI Fortran
Interprocedural Analysis
Parallelization
Locality Opt
C
EDG C++
SUIF2
Alpha
Java
Inst. Scheduling
Register Allocation
x86
Overview of SUIF Components
Basic Infrastructure
Extensible IR and utilities
Modular compiler system
Pass submodule
Backend Infrastructure
Optimization framework
HL Analysis Infrastructure
Graphs, sccs
Presburger arithmetic (omega)
Farkas lemma&“Gaussian Elimination”
Interprocedural framework
Garbage collection (Boehm’s)
PGI Fortran, EDG C/C++, Java FE
Standard IR, OSUIF (for OO lang)
Hoof: Suif object specification language
SUIF1 / SUIF2 translators
Statement dismantlers
Suifbrowser
Register allocation
Scheduling
Alpha and x86 backends
Call graph
Affine partitioning for parallelism & locality
Steensgaard’s alias analysis
Interprocedural parallelization:
array/scalar dependence/privatization
Motivation for Extensible IR
 Suif1 design
 A fixed set of flat C++ classes
 All codes must change if we add new IR nodes
e.g. printing objects, reading and writing to file
 Higher level semantic objects
 OSUIF (for object-oriented programming)
 Verilog event-driven control flow semantics
 Saturated arithmetic used in multimedia instruction sets
 Program analysis concepts
 phi nodes in SSA analysis
 synchronization operations introduced by parallelization
 Results of analysis
Concept I: Reflective IR
 Metaclass: captures representation of IR nodes in a data structure
 Enables common generic routines to implement
 Persistence: reading/writing to disk
 Cloning: routine that copies the data structure
 Printing: a generic routine to print out the information
 Walkers, iterators, visitors
Concept II:
Object hierarchy & virtual aggregate fields
ExecutionObject
Statement
get_child_statements
IfStatement
WhileStatement
get_then_part
get_else_part
get_body
 Abstract names to refer to fundamental concepts in subclasses
 e.g. get_child_statements
IfStatement: get_then_part and get_else_part, or
WhileStatement: get_body
Object hierarchy & virtual aggregate fields
 Maximize reuse and modular code development
 Write code that operates at the highest level of abstraction
 A pass works, without recompilation, for code with new subclasses
if new refinements are immaterial to a pass.
 The unknown information is maintained across a pass
 Example:
 Reuse an intraprocedural dead code elimination written for SUIF
on OSUIF (object-oriented SUIF) code
Concept III: Multiple Representations for
High-Level Analyses
 Multiple representations for different semantic levels
 e.g. FOR loops versus basic blocks in a control flow graph
 => Alternative representations
Mixture of high-level and low-level constructs
Dismantlers lower the representation
Concept IV: High-level object specification
Insulates user from details
Object Definition (.hoof)
SUIF Macro Generator
a general grammar-based tool
Interface for user (.h)
Implementation in
Meta-Class System (.cpp)
Meta-Class System
reading & writing to file in
machine-independent format
• Easy for the programmer
• Easy for the implementor to develop the system
Example of a Hoof Definition
C++
hoof
concrete New
{ int x; }
class New : public SuifObject
{
public:
int get_x();
void set_x(int the_value);
~New();
void print(…);
static const Lstring get_class_name();
…
}
 Uniform data access functions (get_ & set_)
 Automatic generation of meta class information etc.
Examples of Suif Nodes
abstract Statement : ExecutionObject
{
virtual list<Statement* owner> child_statements;
...
}
concrete IfStatement : Statement
{
Expression * condition in source_ops;
Statement * owner then_part in child_statements;
Statement * owner else_part in child_statements;
}
Motivation for a Modular Compiler System
 SUIF1:
 All passes read and write suif files: more modular and supportive of
experimentation
but it is slow
 Data between passes are written out as annotations
Annotations must be expressed as strings when written out
Requires manual pickling of annotations
 Nontrivial effort to support any interactivity
SUIF2
Concept I: A Modular Compiler Architecture
Executable
suifdriver
MODULES:
Passes
analyses
optimizations
Kernel
suifkernel
iokernel
IR
suifnodes
basicnodes
Components
 Kernel: provides all basic functionality
 iokernel: implements I/O
 suifkernel: hides iokernel and provides modules support, cloning,
command line parsing, list of factories, etc.
 Modules
 passes: provides a pass framework
 IR: basic program representations
 Suifdriver
 provides execution control over modules and passes
Memory/Memory vs File/File Passes
COMPILER
A series of stand-alone programs
Suif-file1
driver+module1
Suif-file2
driver+module2
Suif-file3
A driver that imports & applies
modules to program in memory
Suif-file1
Suifdriver
imports/executes
module1
module2
module3
driver+module3
Suif-file4
Suif-file4
Concept II: Dynamic Registration of Modules
 Compilation State: SuifEnv
 Keeps the loaded SUIF program
 All the registered modules
 A module is a C++ class
 that implements either a pass or a set of nodes in IR
 must have a unique module_name
 one or more modules make up a dll (dynamically linked library)
 each library includes a function (init_<dllname>) to register the
module dynamically
Dynamic Assembly of a Compiler
> suifdriver
suif> import basicnodes suifnodes
suif> import mylibrary
suif> load test.suif
suif> mylibrary_pass1
suif> print test.out
suif> save test.tsuif
 The Suifdriver:
 accepts a simple scripting language. (A version that accepts tcl/tk also
exists).
 pre-registered modules (import, load, print, save)
 imports libraries dynamically which can register new modules (new
commands)
 System can easily be used for demand-driven program analysis e.g. SUIF
explorer or a debugger
Concept III: Easy to write a new analysis:
subclass of a pass module
Executable
suifdriver
Passes
analyses
optimizations
Kernel
suifkernel
iokernel
IR
suifnodes
basicnodes
Example Pass: Count Statements in Procedures
class mypass: public Pass {
public:
mypass(SuifEnv *env, const Lstring &name): Pass(env, name) {}
virtual ~mypass() {}
Module *clone() const {return(Module*) this:}
void do_procedure_definition (ProcedureDefinition* proc_def)
{
cout << proc_def->get_procedure_symbol() ->get_name()
cout << object_iterator<ExecutionObject>(proc_def).length();
}
}
extern “C” void init_mypass (SuifEnv *suif_env) {
suif_env->get_module_subsystem()->register_module
(new mypass (suif_env, “mypass”));
}
Research Infrastructure: Support at 3 Levels
 I. Compose compiler with existing passes
 Dynamic composition of different passes
 II. Develop new passes
 User concentrates on algorithmic issues
 Infrastructure provides common functionalities
 Write code once, can run on SUIF program in memory or a file
 III. Develop new IR
 High-level specification of the IR nodes
 Old code works with new IR without recompilation
Deliverables




Infrastructure: basesuif system
Components: Front ends and passes
Testing
Documentation
Basic Infrastructure (Beta release)
 SUIF object system
 I/O, printing, cloning
 Visitors: dispatch method according to type
 Iterators: simple iteration of certain objects
 Walkers: user-controllable traversal of data structures
 Module subsystem
 Hoof language implementation
 Standard IR representation
 General data structures
 Infinite precision integers, strings
 lists, sets, hash maps
 assertion, error handling, command line parsing
 SuifBrowser
 Presents the source and SUIF code
Compiler components: C
EDG C
lcc C
SUIF +
cfenodes
SUIF1
SUIF
C
SUIF1
 EDG C front end -> SUIF
 Created cfenodes to represent C
constructs, which are then lowered
 SUIF -> C
 SUIF1 -> SUIF -> SUIF1
 Enables SUIF1 users to use old SUIF1
passes in the compiler
 Testing
 PGI unit tests
16 errors out of > 3000 tests
 Spec95int: all but gcc work.
(gcc not Ansi C, we are working on it).
Object-oriented Languages: C++ and Java
j2s
OSUIF
SUIF
C
EDG C++
 OSUIF: object-extension of SUIF (for Java and
C++)
 Types: classes, methods and fields
 Symbols: fields & methods
 SymbolTable: class-specific behavior for
member lookup
 Statements: exception handling & objectoriented instructions
 EDG C++ -> SUIF (on track)
Java
 no threads, exception handling, dynamic loading
 Java byte code-> OSUIF (Solaris)
 Class loading, resolution of the constant pool
 Control flow analysis (normal and exception handling)
 Data flow analysis (construct type information)
 OSUIF code generation
 OSUIF -> SUIF
 build-vtables: virtual tables
 lower-instance-variables: layout of class-types
 lower-methods: method dispatch
 lower-static-fields, lower-static-methods: put static in global scope
 Testing: compiled JDK 1.2beta2 javac compiler: 538 Java classes
Fortran front end
 Based on PGI’s proprietary front end
 PGI Fortran includes F90 and HPF;
only the F77 subset is translated to SUIF
 To be released only in binary form
 Alpha version delivered to Stanford, not yet released
Documentation
The SUIF2 Infrastructure Guide (an index to all documentation)
Overview of the SUIF Infrastructure
The SUIF Representation Guide
The Basic SUIF Programmer’s Guide
The SUIF System Programmer’s Reference Manual
 All the high-level interfaces documented (example)
 Generated from the .h files by Doxygen
All the cross references, e.g. class hierarchy information
 Web page: http://suif.stanford.edu/





The End
Overview of SUIF Components
Basic Infrastructure
Extensible IR and utilities
PGI Fortran, EDG C/C++, Java FE
Standard IR, OSUIF (for OO lang)
Hoof: Suif object specification language
SUIF1 / SUIF2 translators
Statement dismantlers
Suifbrowser
Modular compiler system
Pass submodule
Backend Infrastructure
Optimization framework
Register allocation
Scheduling
Alpha code generator, x86 code generator
HL Analysis Infrastructure
Graphs, sccs
Presburger arithmetic (omega)
Farkas lemma&“Gaussian Elimination”
Interprocedural framework
Garbage collection (Boehm’s)
Released
Call graph
Affine partitioning for parallelism & locality
Steensgaard’s alias analysis
Interprocedural parallelization:
array/scalar dependence/privatization
Pre-release
In progress
Comments




Will never implement such a system if not for the infrastructure project
Enables safe type casting of objects
All objects created via factories
Owner edges embed a tree into program representation
Descargar

powerpoint - SUIF - Stanford University