US 5,577,253 AGrant
Analyzing Inductive Expressions in a Multilanguage Optimizing Compiler
Issue Date:1996-11-19
•9 Claims
•4 Drawing Sheets
Abstract
A compiler framework comprises a generic compiler back end which may be used by a plurality of front ends to generate object code for a target computer system. Each front end scans and parses a source module containing source code for a programming language, and generates an intermediate language representation that describes the source code. The intermediate language representation is input to the generic compiler back end which performs optimization and code generation for a plurality of target computer systems. A plurality of optimizations are performed by the generic compiler back end using induction variables. Induction variables are commonly used in performing optimizations such as loop unrolling. An efficient method for determining a set of induction variables for a programming loop construct uses effect and dependency information. Typically, effect and dependency information is used for another purpose besides detecting induction variables, for example, to validate other code optimizations. The method for determining the set of induction variables reuses this information as a shortcut to determine induction variables.
Metadata
Assignee
- Digital Equipment Corporation
Inventor
- David S. Blickstein
Application Information
Application Number:US 4001729
Filing Date:1995-03-06
Priority Date:1991-02-27
Art Unit:237
Classifications
IPC:
G06F 945
Field of Search:
395200
Patent Drawings (4 sheets)
Description
RELATED CASES
This application discloses subject matter also disclosed in the following copending applications, filed herewith and assigned to Digital Equipment Corporation, the assignee of this application:
Ser. No. 07/662,461, filed Feb. 27, 1991 abandoned in favor of Ser. No. 08/249,670, filed May. 26, 1994, now U.S. Pat. No. 5,493,625, by Robert Neil Faiman, Jr., David Scott Blickstein and Steven Hobbs, for "INTERFACE FOR REPRESENTING EFFECTS IN A MULTILANGUAGE OPTIMIZING COMPILER"; (PD90-0364)
Ser. No. 07/662,725, filed Feb. 27, 1991 abandoned in favor of Ser. No. 08/243,560, filed May 16, 1994; by Robert Neil Faiman, Jr., for "METHOD OF CONSTRUCTING A CONSTANT-FOLDING MECHANISM IN A MULTILANGUAGE OPTIMIZING COMPILER"; (PD90-0365)
Ser. No. 07/662,477, filed Feb. 27, 1991 abandoned in favor of Ser. No. 08/243,615 filed May 16, 1994 and Ser. No. 08/364,432 filed Dec. 27, 1994; by Dennis Joseph Murphy and Robert Neil Faiman, Jr., for "INTERFACE FOR SYMBOL TABLE CONSTRUCTION IN A MULTILANGUAGE OPTIMIZING COMPILER"; (PD90-0366)
Ser. No. 07/662,464, filed Feb. 27, 1991 abandoned in favor of Ser. No. 08/234,441 filed Apr. 20, 1994, now allowed; by Caroline Sweeney Davidson, Richard Barry Grove and Steven Hobbs, for "MULTILANGUAGE OPTIMIZING COMPILER USING TEMPLATES IN MULTIPLE PASS CODE GENERATION". (PD90-0369)
BACKGROUND OF THE INVENTION
This invention relates to compilers for digital computer programs, and more particularly to a compiler framework that is adapted to be used with a number of different computer languages, to generate code for a number of different target machines.
Compilers are usually constructed for translating a specific source language to object code for execution on a specific target machine which has a specific operating system. For example, a Fortran compiler may be available for generating code for a computer having the VAX architecture using the VMS operating system, or a C compiler for a 80386 computer executing MS/DOS. Intermediate parts of these language- and target-specific compilers share a great deal of common structure and function, however, and so construction of a new compiler can be aided by using some of the component parts of an existing compiler, and modifying others. Nevertheless, it has been the practice to construct new compilers for each combination of source language and target machine, and when new and higher-performance computer architectures are designed the task of rewriting compilers for each of the commonly-used source languages is a major task.
The field of computer-aided software engineering (CASE) is heavily dependent upon compiler technology. CASE tools and programming environments are built upon core compilers. In addition, performance specifications of computer hardware are often integrally involved with compiler technology. The speed of a processor is usually measured in high-level language benchmarks, so therefore optimizing compilers can influence the price-performance factor of new computer equipment.
In order to facilitate construction of compilers for a variety of different high-level languages, and different target computer architectures, it is desirable to enhance the commonality of core components of the compiler framework. The front end of a compiler directly accesses the source code module, and so necessarily is language-specific; a compiler front end constructed to interpret Pascal would not be able to interpret C. Likewise, the code generator in the back end of a compiler has to use the instruction set of the target computer architecture, and so is machine-specific. Thus, it is the intermediate components of a compiler that are susceptible to being made more generic. Compiler front end usually functions to first translate the source code into an intermediate language, so that the program that was originally written in the high-level source language appears in a more elemental language for the internal operations of the compiler. The front end usually produces a representation of the program or routine, in intermediate language, in the form of a so-called graph, along with a symbol table. These two data structures, the intermediate language graph and the symbol table, are the representation of the program as used internally by the compiler. Thus, by making the intermediate language and construction of the symbol table of universal or generic character, the components following the front end can be made more generic.
After the compiler front end has generated the intermediate language graph and symbol table, various optimizing techniques are usually implemented. The flow graph is rearranged, meaning the program is rewritten, to optimize speed of execution on the target machine. Some optimizations are target-specific, but most are generic. Commonly-used optimizations are code motion, strength reduction, etc. Next in the internal organization of a compiler is the register and memory allocation. Up to this point, data references were to variables and constants by name or in the abstract, without regard to where stored; now, however, data references are assigned to more concrete locations, such as specific registers and memory displacements (not memory addresses yet). At this point, further optimizations are possible, in the form of register allocation to maintain data in registers are minimize memory references; thus the program may be again rearranged to optimize register usage. Register allocation is also somewhat target machine dependent, and so the generic nature of the compiler must accommodate specifying the number, size and special assignments for the register set of the target CPU. Following register and memory allocation, the compiler implements the code generation phase, in which object code images are produced, and these are of course in the target machine language or instruction set, i.e., machine-specific. Subsequently, the object code images are linked to produce executable packages, adding various run-time modules, etc., all of which is machine-specific.
In a typical compiler implementation, it is thus seen that the structure of the intermediate language graph, and the optimization and register and memory allocation phases, are those most susceptible to being made more generic. However, due to substantive differences in the high-level languages most commonly used today, and differences in target machine architecture, obstacles exist to discourage construction of a generic compiler core.
SUMMARY OF THE INVENTION
In accordance with one embodiment of the invention, a compiler framework is provided which uses a generic "shell" or control and sequencing mechanism, and a generic back end (where the code generator is of course target-specific). The generic back end provides the functions of optimization, register and memory allocation, and code generation. The shell may be executed on various host computers, and the code generation function of the back end may be targeted for any of a number of computer architectures. A front end is tailored for each different source language, such as Cobol, Fortran, Pascal, C, C++, Ada, etc. The front end scans and parses the source code modules, and generates from them an intermediate language representation of the program expressed in the source code. This intermediate language is constructed to represent any of the source code languages in a universal manner, so the interface between the front end and back end is of a standard format, and need not be rewritten for each language-specific front end.
The intermediate language representation generated by the front end is based upon a tuple as the elemental unit, where each tuple represents a single operation to be performed, such as a load, a store, an add, a label, a branch, etc. A data structure is created by the front end for each tuple, with fields for various necessary information. Along with the ordered series of tuples, the front end generates a symbol table for all references to variables, routines, labels, etc., as is the usual practice. The tuples are in ordered sequences within blocks, where a block is a part of the code that begins with a routine or label and ends in a branch, for example, where no entry or exit is permitted between the start and finish of a block. Each block is also a data structure, or node, and contains pointers to its successors and predecessors (these being to symbols in the symbol table). The interlinked blocks make up a flow graph, called the intermediate language graph, which is the representation of the program used by the back end to do the optimizations, register and memory allocations, etc.
One of the features of the invention is a mechanism for representing effects and dependencies in the interface between front end and back end. A tuple has an effect if it writes to memory, and has a dependency if it reads from a location which some other node may write to. Various higher level languages have differing ways of expressing operations, and the same sequence may in one language allow a result or dependency, while in another language it may not. Thus, a mechanism which is independent of source language is provided for describing the effects of program execution. This mechanism provides a means for the compiler front end to generate a detailed language-specific information to the multilanguage optimizer in the compiler back end. This mechanism is used by the global optimizer to determine legal and effective optimizations, including common subexpression recognition and code motions. The intermediate language and structure of the tuples contain information so that the back end (optimizers) can ask questions of the front end (obtain information from the intermediate language graph), from which the back end can determine when the execution of the code produced for the target machine for one tuple will affect the value computed by code for another tuple. The interface between back end and front end is in this respect language independent. The back end does not need to know what language it is compiling. The advantage is that a different back end (and shell) need not be written for each source language, but instead an optimizing compiler can be produced for each source language by merely tailoring a front end for each different language.
Another feature of one embodiment of the invention is the use in the optimization part of the compiler of a method for analyzing induction variables. A variable is said to be an induction variable if it increments or decrements once every time through the loop, and is executed at most once each time through the loop. In addition to finding induction variables, this optimization finds inductive expressions, which are expressions that can be computed as linear functions of induction variables. The object of this optimization is generally to replace multiplications with additions, which are cheaper execute faster on most architectures); this is known as strength reduction. Detection of induction variables requires the use of "sets" of potential induction variables; doing this dynamically for each loop is an expensive and complicated operation, so the improvement here is to use the side effects sets used to construct IDEF sets.
An additional feature of one embodiment of the invention is a mechanism for "folding constants" (referred to as K-folding or a KFOLD routine), included as one of the optimizations. This mechanism is for finding occurrences where expressions can be reduced to a constant and calculated at compile time rather than a more time-consuming calculation during runtime. An important feature is that the KFOLD code is built by the compiler framework itself rather than having to be coded or calculated by the user. The KFOLD builder functions as a front end, like the other language-specific front ends, but there is no source code input; instead, the input is in intermediate language and merely consists of a listing of all of the operators and all of the data types. The advantage is that a much more thorough KFOLD package can be generated, at much lower cost.
A further feature of one embodiment is the type definition mechanism, referred to a the TD module. This module provides mechanisms used by the front end and the compiler of the back end in constructing program type information to be incorporated in an object module for use by a linker or debugger. The creation of "type information" takes place in the context of symbol table creation and allows a front end to specify to the back end an abstract representation of program type information. The TD module provides service routines that allow a front end to describe basic types and abstract types.
In addition, a feature of one embodiment is a method for doing code generation using code templates in a multipass manner. The selection and application of code templates occurs at four different times during the compilation process: (1) The pattern select or PATSELECT phase does a pattern match in the CONTEXT pass to select the best code templates; (2) The TNASSIGN and TNLIFE tasks of the CONTEXT pass use context actions of the selected templates to analyze the evaluation order to expressions and to allocate temporary names (TNs) with lifetimes nonlocal to the code templates; (3) The TNBIND pass uses the binding actions of the selected templates to allocate TNs with lifetimes local to the code templates; (4) Finally, the CODE pass uses code generation actions of the selected templates to guide the generation of object code.
BRIEF DESCRIPTION OF THE DRAWINGS
The novel features believed characteristic of the invention are set forth in the appended claims. The invention itself, however, as well as other features and advantages thereof, will be best understood by reference to the detailed description of specific embodiments which follows, when read in conjunction with the accompanying drawings, wherein:
FIG. 1 is a schematic representation of a compiler using features of the invention;
FIG. 2 is an electrical diagram in block form of a host computer upon which the methods of various features of the invention may be executed;
FIG. 3 is a diagrammatic representation of code to be compiled by the compiler of FIG. 1, in source code form, intermediate language form, tree from, and assembly language form;
FIG. 4 is a diagrammatic representation of the data structure of a tuble used in the compiler of FIG. 1;
FIG. 5 is a logic flow chart of the operation of the shell of FIG. 1;
FIG. 6 is an example listing of code containing constants; and
FIG. 7 is a diagram of data fields and relationships (pointers) for illustrating type definition according to one feature of the invention.
DETAILED DESCRIPTION OF SPECIFIC EMBODIMENTS
Referring to FIG. 1, the compiler framework 10 according to one embodiment of the invention is a language-independent framework for the creation of portable, retargetable compilers. The compiler framework 10 consists of a portable operating system interface referred to as the shell 11 and a retargetable optimizer and code generator 12 (the back end). The shell 11 is portable in that can be adapted to function with any of several operating systems such as VAX/VMS, Unix, etc., executing on the host computer. The shell operates under this host operating system 13 executing on a host computing system as seen in FIG. 2, typically including a CPU 14 coupled to a main memory 15 by a system bus 16, and coupled to disk storage 17 by an I/O controller 18. The shell 11 and compiler 12 may be combined with a front end 20 to create a portable, retargetable compiler for a particular source language. Thus, a compiler based on the framework 10 of the invention consists of three basic parts: a shell 11 which has been tailored for a particular host operating system 14--this determines the host environment of the compiler; a front end 20 for a particular source language (e.g., C, C++, Pascal, Fortran, Ada, Cobol, etc.)--this determines the source language of the compiler; and a back end 12 for a particular target machine (i.e., a particular architecture such as VAX, RISC, etc.)--this determines the target machine of the compiler.
Since the interfaces between the shell 11, the front end 20, and the back end 12 are fixed, the individual components of a compiler produced according to the invention may be replaced freely. That is, the front end 20 may consist of a number of interchangeable front ends, e.g., one for Fortran, one for Cobol, one for Pascal, one for C, etc. Likewise, a shell 11 tailored for running under VMS on a VAX computer may be replaced by a shell 11 running under the Unix operating system on a RISC workstation, while the front end 20 and back end 12 remain the same.
The shell 11 provides a fixed interface between the host operating system 13 and the rest of the compiler. The shell provides several advantages according to the invention. First, the shell 11 provides a portable interface to basic features of the operating system 13. For example, the front end 20 need not know details of the file system, command parsing, or heap storage allocation under the host operating system 13, since all these services are accessed through shell routines, and the shell is tailored to the operating system 13 being used. Second, the shell 11 eliminates duplication of effort by providing a single implementation of some common compiler components, such as command line parsing, include-file processing, and diagnostic file generation. Third, the use of these common components also guarantees consistency among compilers created using the framework 10; all compilers created using this framework 10 will write listing files in the same format, will treat command line qualifiers the same, will issue similar-looking error messages, etc. Fourth, having common shell utilities in the shell 11 improves the internal integration of the compiler, since the front and back ends 20 and 12 use the same shell functions. For example, use of the shell locator package means that source file locations can be referred to consistently in the source listing, front-end generated diagnostics, back-end generated diagnostics, the object listing, and the debugger information.
The front end 20 is the only component of a compiler created by the framework 10 which understands the source language being compiled. This source language is that used to generate the text of a source code file or files (module or modules) 21 which define the input of the compiler. The front end 20 performs a number of functions. First, it calls the shell 11 to obtain command line information and text lines from the source files 21. Second, the front end 20 calls the shell 11 to control the listing file, write diagnostic messages, and possibly to write other files for specific languages. Third, the front end 20 does lexical, syntactic, and semantic analysis to translate the source text in file 21 to a language-independent internal representation used for the interface 22 between the front end 20 and the back end 12. Fourth, the front end 20 invokes the back end 12 to generate target system object code 23 from the information in the internal representation. Fifth, the front end 20 provides routines which the back end 12 calls via call path 24 to obtain language-specific information during back end processing. Not included in the compiler framework of FIG. 1 is a linker which links the object code modules or images 23 to form an executable image to run on the target machine 25.
The target machine 25 for which the back end 12 of the compiler creates code is a computer of some specific architecture, i.e., it has a register set of some specific number and data width, the logic executes a specific instruction set, specific addressing modes are available, etc. Examples are (1) the VAX architecture, as described in (2) a RISC type of architecture based upon the 32-bit RISC chip available from MIPS, Inc., as part number R2000 or R3000 and described by Lane in "MIPS R2000 RISC Architecture", Printice-Hall, 1987, and (3) an advanced RISC architecture with 64-bit registers as described in copending application Ser. No. 547,589, filed Jun. 29, 1990. Various other architectures would be likewise accommodated.
In general, the front end 20 need not consider the architecture of the target machine 25 upon which the object code 23 will be executed, when the front end 20 is translating from source code 15 to the internal representation of interface 22, since the internal representation is independent of the target machine 25 architecture. Some aspects of the front end 20 may need to be tailored to the target system, however; for example, aspects of the data representation such as allocation and alignment, might be customized to fit the target machine 25 architecture better, and routine call argument mechanisms may depend on the target system calling standard, and further the runtime library interface will probably be different for each target system.
The back end 12 functions to translate the internal representation 22 constructed by the front end 20 into target system object code 23. The back end 12 performs the basic functions of optimization 26, code generation 27, storage and register allocation 28, and object file emission 29. The optimization function is performed on the code when it is in its internal representation. The back end 12 also includes utility routines which are called by the front end 20 to create a symbol table 30 and intermediate language data structures.
When the user (that is, a user of the computer system of FIG. 2, where the computer system is executing the operating system 13) invokes the compiler of FIG. 1 (though a callable interface, or some other mechanism), the shell 11 receives control. The shell 11 invokes the front end 20 to compile an input stream from source file 15 into an object file 23. The front end 20 invokes the back end 12 to produce each object module within the object file 23. The front end 20 may invoke the back end 12 to create code for each individual routine within an object module 23, or it may call a back end driver which will generate code for an entire module at once.
The front end 20 parses the source code 21 and generates an intermediate language version of the program expressed in the source code. The elemental structure of the intermediate language is a tuple. A tuple is an expression which in source language performs one operation. For example, referring to FIG. 3, the expression
as represented in source language is broken down into four tuples for representation in the intermediate language, these being numbered $1, $2, $3 and $4. This way of expressing the code in IL includes a first tuple $1 which is a fetch represented by an item 31, with the object of the fetch being a symbol J. The next tuple is a literal, item 32, also making reference to a symbol "1." The next tuple is an Add, item 33, which makes reference to the results of tuples $1 and $2. The last tuple is a store, item 34, referencing the result of tuple $3 and placing the result in symbol I in the symbol table. The expression may also be expressed as a logic tree as seen in FIG. 3, where the tuples are identified by the same reference numerals. This same line of source code could be expressed in assembly for a RISC type target machine, as three instructions LOAD, ADD integer, and STORE, using some register such as REG 4 in the register file, in the general form seen in FIG. 3. Or, for a CISC machine, the code emitted may be merely a single instruction, ADD #1,J,I as also seen in the Figure.
A tuple, then, is the elemental expression of a computer program, and in the form used in this invention is a data structure 35 which contains at least the elements set forth in FIG. 4, including (1) an operator and type field 36, e.g., Fetch, Store, Add, etc., (2) a locator 37 for defining where in the source module 21 the source equivalent to the tuple is located, (3) operand pointers 38 to other tuples, to literal nodes or symbol nodes, such as the pointers to I and #1 tuples $1 and $2 in FIG. 3. A tuple also has attribute fields 39, which may include, for example, Label, Conditional Branch, Argument (for Calls), or SymRef (a symbol in the symbol table). The tuple has a number field 40, representing the order of this tuple in the block.
The front end 20 parses the source code to identify tuples, then to identify basic blocks of code. A block of code is defined to be a sequence of tuples with no entry or exit between the first and last tuple. Usually a block starts with a label or routine entry and ends with a branch to another label. A task of the front end 20 is to parse the source code 21 and identify the tuples and blocks, which of course requires the front end to be language specific. The tuple thus contains fields 41 that say whether or not this tuple is the beginning of a block, and the end of a block.
As discussed in more detail below, one feature of the invention is a method of representing effects. A tuple has effects if it stores or writes to a memory location (represented at the IL level as a symbol), or is dependent upon what another tuple writes to a location. Thus, in the example given in FIG. 3, tuple $4 has an effect (store to I) and tuple $1 has a dependency (content of J). Thus the tuple data structure as represented in FIG. 4 has fields 42 and 43 to store the effects and dependencies of this tuple.
A single execution of a compiler of FIG. 1 is driven by the shell 11 as illustrated in the flow chart of FIG. 5. As indicated by the item 45 of FIG. 5, the shell 11 receives control when the compiler of FIG. 1 is invoked by the user via the operating system 13. The user in a command line specifies a "plus-list" or list of the modules 21 to be operated upon. The next step is calling by the shell 11 of a front-end routine GEM$XX.sub.-- INIT, which does any necessary initialization for the front end, indicated by the item 46. This front end routine GEM$XX.sub.-- INIT is described in the Appendix. Next, the shell 11 parses the global command qualifiers and calls the front end routine GEM$XX.sub.-- PROCESS.sub.-- GLOBALS, as indicated by the item 47. Then, for each "plus-list" (comma-separated entity) in the command line used at the operating system 13 level to involve the compiler, the shell executes a series of actions; this is implemented by a loop using a decision point 48 to check the plus-list. So long as there is an item left in the plus-list, the actions indicated by the items 49-52 are executed. These actions include accessing the source files 21 specified in the command line and creating an input stream for them, indicated by the item 49, then parsing the local qualifiers (specific to that plus-list), calling GEM$XX.sub.-- PROCESS.sub.-- LOCALS to do any front-end determined processing on them, and opening the output files specified by the qualifiers, indicated by the item 50. The actions in the loop further include calling the front-end routine GEM$XX.sub.-- COMPILE to compile the input stream, indicated by the item 51, then closing the output files, item 52. When the loop falls through, indicating all of the plus-list items have been processed, the next step is calling the front end routine GEM$XX.sub.-- FINI to do any front-end cleanup, indicated by item 53. Then, the execution is terminated, returning control to the invoker, item 54.
The shell 11 calls GEM$XX.sub.-- COMPILE to compile a single input stream. An input stream represents the concatenation of the source files or modules 21 specified in a single "plus list" in the compiler command line, as well as any included files or library text. By default, compiling a single input stream produces a single object file 23, although the compiler does allow the front end 20 to specify multiple object files 23 during the compilation of an input stream.
Before calling GEM$XX.sub.-- COMPILE, the shell 11 creates the input stream, parses the local qualifiers, and opens the output files. After calling GEM$XX.sub.-- COMPILE, it closes all the input and output files.
The front end 20 (GEM$XX.sub.-- COMPILE and the front-end routines that are called from it) reads source records 21 from the input stream, translates them into the intermediate representation of interface 22 (including tuples, blocks, etc. of the intermediate language graph, and the symbol table) and invokes the back end 12 to translate the intermediate representation into object code in the object file 23.
An object file 23 may contain any number of object modules. Pascal creates a single object module for an entire input stream (a MODULE or PROGRAM). FORTRAN (in one embodiment) creates a separate object module for each END statement in the input stream. BLISS creates an object module for each MODULE.
To create an object module 23, the front end 20 translates the input stream or some subsequence thereof (which we can call a source module 21 ) into its internal representation for interface 22, which consists of a symbol table 30 for the module and an intermediate language graph 55 for each routine. The front end 20 then calls back end routines to initialize the object module 23, to allocate storage for the symbols in the symbol table 30 via storage allocation 28, to initialize that storage, to generate code for the routines via emitter 29, and to complete the object module 23.
The compiler is organized as a collection of packages, each of which defines a collection of routines or data structures related to some aspect of the compilation process. Each package is identified by a two-letter code, which is generally an abbreviation of the package function. The interface to a package is defined by a specification file. If a package is named ZZ, then its specification file will be GEM$ZZ.SDL.
Any symbol which is declared in a package's specification file is said to be exported from that package. In general, symbols exported from package ZZ have names beginning with GEM$ZZ.sub.--. The specific prefixing conventions for global and exported names are set forth in Table 1.
The shell 11 is a collection of routines to support common compiler activities. The shell components are interrelated, so that a program that uses any shell component gets the entire shell. It is possible, however, for a program to use the shell 11 without using the back end 12. This can be a convenient way of writing small utility programs with production-quality features (input file concatenation and inclusion, command line parsing, diagnostic file generation, good listing files, etc.) Note that the shell 11 is actually the "main program" of any program that uses it, and that the body of the application must be called from the shell 11 using the conventions described below. To use a shell package ZZ from a BLISS program, a user does a LIBRARY `GEM$ZZ`. To use the shell from other languages, a user must first translate the shell specification files into the implementation language.
The shell packages are summarized in the following paragraphs; they are documented in their specification files (the GEM$ZZ.SDL files) in the Appendix. Most shell routine arguments (e.g., integer, string, etc.) fall into one of the categories set forth in Table 2.
The interface from the shell 11 to the front end 20 has certain requirements. Since the shell 11 receives control when a compiler of FIG. 1 is invoked, a front end 20 must declare entry points so the shell 11 can invoke it and declare global variables to pass front end specific information to the shell 11. The front end 20 provides the global routines set forth in Table 3, in one embodiment. These routines have no parameters and return no results.
The Virtual Memory Package (GEM$VM): The virtual memory package provides a standard interface for allocating virtual memory. It supports the zoned memory concept of the VMS LIB$VN facility; in fact, under VMS, GEM$VM is an almost transparent layer over LIB$VM. However, the GEM$VM interface is guaranteed to be supported unchanged on any host system.
The Locator Package (GEM$LO): A locator describes a range of source text 21 (starting and ending file, line, and column number). The text input package returns locators for the source lines that it reads. Locators are also used in the symbol table 30 and intermediate language nodes 43 to facilitate message and debugger table generation, and are used for specifying where in the listing file the listing package should perform actions. A locator is represented as a longword. The locator package maintains a locator database, and provides routines to create and interpret locators. There is also a provision for user-created locators, which allow a front end to create its own locators to describe program elements which come from a non-standard source (for example, BLISS macros or Ada generic instantiation).
The Text Input Package (GEM$TI): The text input package supports concatenated source files 21, nested (included) source files 21, and default and related files specs, while insulating the front end 20 from the I/O architecture of the underlying operating system 13. Text of a source file 21 is read a line at a time. The text input package GEM$TI colludes with the locator package GEM$LO to create a locator describing each source line it reads.
The Text Output Package (GEM$TX): The text output package supports output to any number of output files 44 simultaneously. Like the text input package, it insulates its caller from the operating system 13. It will write strings passed by reference or descriptor. It provides automatic line wrapping and indentation, page wrapping, and callbacks to a user-provided start-of-page routine.
The Listing Package (GEM$LS): The listing package will write a standard format listing file containing a copy of the source files 21 (as read by the text input package GEM$TI), with annotations provided by the front end 11 at locations specified with locators. The listing file is created as a GEM$TX output file 44, which the front end 20 may also write to directly, using the GEM$TX output routines.
The Internal Representation
The internal representation of a module 21 comprises a symbol table 30 for the module and a compact intermediate language graph 55 or CILG, for each routine in source module 21. These are both pointer-linked data structures made up of nodes.
Nodes, according to the framework of FIG. 1, will be defined. Almost all data structures used in the interface between the front and back ends 20 and 12 (and most of the data structures used privately by the back end 12) are nodes. A node as the term is used herein is a serf-identifying block of storage, generally allocated from the heap with GEM$VM.sub.-- GET. All nodes have the aggregate type GEM$NODE, with fields GEM$NOD.sub.-- KIND and GEM$NOD.sub.-- SUBKIND. Kind is a value from the enumerated type GEM$NODE.sub.-- KINDS which identifies the general kind of the node. Subkind is a value from the enumerated type GEM$NODE.sub.-- SUBKINDS which identifies the particular kind of the node within the general class of nodes specified by kind. Any particular node also has an aggregate type determined by its kind field. For example, if kind is GEM$NODE.sub.-- K.sub.-- SYMBOL, then the node has type GEM$SYMBOL.sub.-- NODE. Note that the types associated with nodes do not obey the naming conventions described above. The interface node types and their associated enumerated type constants are defined in the files set forth in Table 4.
The compiler framework of FIG. 1 supports a simple tree-structured symbol table 30, in which symbol nodes are linked together in chains off of block nodes, which are arranged in a tree. All symbolic information to be used by the compiler must be included in this symbol table 30. There are also literal nodes, representing literal values of the compiled program; frame nodes, representing storage areas (PSECTs and stack frames) where variables may be allocated; and parameter nodes, representing elements in the parameter lists of routine entry points. The symbol table structure and the contents of symbol table nodes are described below.
The intermediate language is the language used for all internal representations of the source code 21. The front end 20 describes the code of a routine to be compiled as a compact intermediate language graph 55, or CILG. This is simply a linked list of CIL tuple nodes 35 of FIG. 4 (also referred to as tuple nodes, or simply as tuples), each of which represents an operation and has pointers 38 to the tuple nodes representing its operands. Tuple nodes may also contain pointers 38 to symbol table nodes. The intermediate language is described in more detail below.
The front end 20 must create the internal representation 22 of the module 21 one node at a time, and link the nodes together into the symbol table 30 and IL data structures 55. The routines and macros of Table 5, also documented in the Appendix, are used to create and manipulate the data structures of the internal representation 22.
The back end 12 makes no assumptions about how the front end 20 represents block and symbol names. Instead, the front end 20 is required to provide a standard call-back interface that the back end 12 can use to obtain these names.
Every symbol node has a flag, GEM$SYM.sub.-- HAS.sub.-- NAME, and every block node has a flag, GEM$BLK.sub.-- HAS.sub.-- NAME. When the front end 20 initializes a symbol or block node, it must set its has name flag to indicate whether a name string is available for it. (Some symbols and blocks, such as global and external symbols and top level module blocks, must have names.)
There is a global variable, GEM$ST.sub.-- G.sub.-- GET.sub.-- NAME, in the ST package. Before invoking the back end, the front end must set this variable to the address of a callback routine which fits the description set forth in Table 5.
To compile a source module using the GEM$CO.sub.-- COMPILE.sub.-- MODULE interface, a front end (that is, the routing GEM$XX.sub.-- COMPILE) does (in order) each of the activities described in the following paragraphs.
1. Create the Internal Representation
The first task of the front end 20 is to create the internal representation 22 of the source module. To begin with, it must call GEM$ST.sub.-- INIT to initialize the symbol table 30 and associated virtual memory zones. It must then read the source module 21 from the input stream, using the GEM$TI package; do lexical, syntactic, and semantic analysis of the source module 21; and generate the symbol table 30 and the intermediate language graphs 55 for the module as described above, using the GEM$ST and GEM$IL routines which are documented in the Appendix.
In addition, the module's source listing may be annotated with calls to the GEM$LS shell package, and error's in the module may be reported with calls to the GEM$MS package.
If the source module 21 contains errors severe enough to prevent code generation, then the front end 20 should now call GEM$LS.sub.-- WRITE.sub.-- SOURCE to write the listing file and GEM$ST.sub.-- FINI to release all the space allocated for the internal representation 22. Otherwise, it must proceed with the following steps.
2. Specify the Callback Routines
Before calling the back end 12 to compile the module 21, the front end 20 must initialize the following global variables with the addresses of routines that will be called by the back end 12.
(1) GEM$ST.sub.-- G.sub.-- GET.sub.-- NAME must be initialized to the address of a routine that will yield the names of symbol and block nodes in the symbol table 30, as described above.
(2) The GEM$SE.sub.-- G global variables must be initialized to the addresses of routines that will do source-language defined side effect analysis, as described below. The compiler provides a predefined collection of side effect routines, suitable for use during the early development of a front end 20, which can be selected by calling GEM$SE.sub.-- DEFAULT.sub.-- IMPLEMENTATION.
(3) GEM$ER.sub.-- G.sub.-- REPORT.sub.-- ROUTINE contains the address of the front end's routine for reporting back end detected errors, as described below.
3. Do the Compilation
When the internal representation is complete, the front end 20 can call GEM$CO.sub.-- COMPILE.sub.-- MODULE (described below) to translate it into target machine object representation 23. The front end should then call GEM$LS.sub.-- WRITE.sub.-- SOURCE to list the input stream in the listing file. It may also call GEM$ML.sub.-- LIST.sub.-- MACHINE.sub.-- CODE to produce an assembly code listing of the compiled module 23.
Note that normally, GEM$LS.sub.-- WRITE.sub.-- SOURCE has to be called after GEM$CO.sub.-- COMPILE.sub.-- MODULE so that the source listing 21 can be annotated with any error messages generated during back end processing. However, it is a good idea for the front end 20 to provide a debugging switch which will cause GEM$LS.sub.-- WRITE.sub.-- SOURCE to be called first. This will make it possible to get a source listing even if a bug causes the compiler to abort during back end processing.
4. Clean Up
When compilation is complete, the front end 20 must call GEM$CO.sub.-- COMPILE.sub.-- MODULE to release the space used for back end processing, and then GEM$ST.sub.-- FINI to release the space used for the internal representation.
The back end 12 is able to detect conditions during compilation which are likely to represent conditions in the source program 21 which ought to be reported to the user, such as uninitialized variables, unreachable code, or conflicts of static storage initialization. However, a particular front end 20 may need to customize which of these conditions will be reported, or the precise messages that will be issued.
To allow this, the back end 12 reports all anomalous conditions that it detects by calling the routine whose address is in the global variable GEM$34.sub.-- G.sub.-- REPORT.sub.-- ROUTINE, with the argument list described below. This routine is responsible for actually issuing the error message.
There is a default error reporting routine set forth in the Appendix named GEM$ER.sub.-- REPORT.sub.-- ROUTINE, whose address will be in GEM$ER.sub.-- G.sub.-- REPORT.sub.-- ROUTINE unless the front end has stored the address of its own report routine there. This default routine has three uses:
(1) The default routine provides reasonable messages, so the front end developers are not obliged to provide their own routine unless and until they need to customize it.
(2) When the front end developers do choose to write a report routine, they can use the default routine as a model.
(3) The front end's routine can be written as a filter, which processes (or ignores) certain errors itself, and calls the default routine with all others.
Interface for Representing Effects
As an essential step in detecting common subexpressions (CSEs), invariant expressions, and opportunities for code motion, the optimizer 26 in the back end 12 must be able to determine when two expression tuples are guaranteed to compute the same value. The basic criterion is that an expression B computes the same value as an expression A if:
1. A and B are literal references to literals with the same value, CSE references to the same CSE, or symbol references to the same symbol; or
2. a. A. is evaluated on every control flow path from the start of the routine to B, and
b. A and B have the same operator and data type, and
c. the operands of B compute the same values as the corresponding operands of A (obviously a recursive definition), and
d. no tuple which occurs on any path from an evaluation of A to an evaluation of B can affect the value computed by B.
The optimizer 26 of FIG. 1 can validate criteria 1, 2a, 2b, and 2c by itself; but criterion 2d depends on the semantics of the language being compiled, i.e., the language of source code module 21. But since the compiler 12 in the back end must be language-independent, a generic interface is provided to the front end 20 to convey the necessary information. When can the execution of one tuple affect the value computed by another tuple? The interface 22 must allow the optimizer 26 to ask this question, and the compiler front end 20 to answer it.
The model underlying this interface 22 is that some tuples have effects, and that other tuples have dependencies. A tuple has an effect if it might change the contents of one or more memory locations. A tuple has a dependency on a memory location if the value computed by the tuple depends on the contents of the memory location. Thus, the execution of one tuple can affect the value computed by another tuple if it has the effect of modifying a memory location which the other tuple depends on.
Given the ramifications of address arithmetic and indirect addressing, it is impossible in general to determine the particular memory location accessed by a tuple. Thus we must deal with heuristic approximations to the sets of memory locations which might possibly be accessed.
The actual interface 22 provides two mechanisms for the front end 20 to communicate dependency information to the optimizer 26. These are the straight-line dependency interface and the effects-class interface.
In the straight-line dependency interface, to determine dependencies in straight-line code, the optimizer 26 will ask the front end 20 to (1) push tuples on an effects stack and pop them off again, and (2) find the top-most tuple on the effects stack whose execution might possibly affect the value computed by a specified tuple.
The straight-line mechanism is not appropriate when the optimizer 26 needs to compute what effects might occur as a result of program flow through arbitrary sets of flow paths. For this situation, the front end 20 is allowed to define a specified number (initially 128) of effects classes, each representing some (possibly indeterminate) set of memory locations. A set of effects classes is represented by a bit vector. For example, an effects class might represent the memory location named by a particular variable, the set of all memory locations which can be modified by procedure calls, or the set of memory locations which can be accessed by indirect references (pointer dereferences).
For the effects-class interface, the optimizer will ask the front end to (1) compute the set of effects classes containing memory locations which might be changed by a particular tuple, and (2) compute the set of effects classes containing memory locations which a particular tuple might depend on.
Using this effects-class interface, the optimizer can compute, for each basic block, a bit-vector (referred to as the LDEF set) which represents the set of effects classes containing memory locations which can be modified by some tuple in that basic block.
The optimizer will also ask the front end to (3) compute the set of effects classes which might include the memory location associated with a particular variable symbol.
This information is used by the split lifetime optimization phase (see below) to compute the lifetime of a split candidate.
The optimizer 26 uses these interfaces as follows. Remember that the reason for these interfaces is to allow the optimizer 26 in back end 12 to determine when "no tuple which occurs on any path from an evaluation of A to an evaluation of B can affect the value computed by B." If A and B occur in the same basic block, this just means "no tuple between A and B can change the value computed by B." This can be easily determined using the straight-line dependency interface.
If the basic block containing A dominates the basic block containing B (i.e., every flow path from the routine entry node to the basic block containing B passes through the basic block containing A), then the optimizer finds the series of basic blocks X1, X2, . . . Xn, where X1 is the basic block containing A, Xn is the basic block containing B, and each Xi immediately dominates X(i+1). Then the test has two parts:
1. There must be no tuple between A and the end of basic block X1, or between the beginning of basic block Xn and B, or in any of the basic blocks X2, X3, . . . X(n-1), which can change the value computed by B. This can be easily determined using the straight-line dependency interface.
2. There must be no flow path between two of the basic blocks Xi and X(i+1) which contains a tuple which can change the value computed by B. The optimizer tests this with the effects-class mechanism, by computing the union of the LDEF sets of all the basic blacks which occur on any flow path from Xi to X(i+1), computing the intersection of this set with the set of effects classes containing memory locations that B might depend on, and testing whether this intersection is empty.
The structure of the interface will now be described. The interface routines are called by the back end 12. The front end 20 must make its implementation of the interface available before it invokes the back end 12. It does this by placing the addresses of its interface routine entry points in standard global variables. The optimizer 26 can then load the routine address from the appropriate global variable when it invokes one of these routines. The interface routines are documented below with names of the form GEM.sub.-- SE.sub.-- xxx. The front end must store the entry address of each corresponding implementation routine in the global variable named GEM.sub.-- SE.sub.-- G.sub.-- xxx.
Tuples that have effects and dependencies are of interest to this interface. Only a few of the IL tuples can have effects and dependencies. (Roughly speaking, tuples that do a store can have effects; tuples that do a fetch can have a dependency; tuples that do a routine call can have both.)
More specifically, each tuple falls into one of the following categories:
1. The tuple does not have any effects, nor is it dependent on any effects. (Example: ADD). Tuples that fall into this class are NOT pushed on the effects stack. Nor are such tuples ever passed to GEM.sub.-- SE.sub.-- EFFECTS.
2. The tuple may have effects, but has no dependencies. (Example: STORE).
3. The tuple may have dependencies, but does not cause any affects. (Example: FETCH).
4. The tuple both may have effects (out-effects) and a separate set of dependencies (in-effects). (Example: procedure calls)
5. The tuple may have both effects and dependencies. The effects it depends on are identical to the effects it produces. (Example: PREINCR).
A particular tuple called the DEFINES tuple is provided to allow a front end 20 to specify effects which are not associated with any tuple. One possible use of the DEFINES tuple would be to implement the BLISS CODECOMMENT feature, which acts as a fence across which optimizations are disallowed. The translation of CODECOMMENT would be a DEFINES tuple that has all effects, and therefore invalidate all tuples.
Argument passing tuples (such as ARGVAL and ARGADR) have effects and dependencies. However, the effects and dependencies of a parameter tuple are actually considered to be associated with the routine call that the parameter tuple belongs to. For example, in the BLISS routine call F(X,.X+Y), the parameter X would have the effect of changing X. However, this would not invalidate a previously computed value of .X+.Y, since the effect does not actually occur until F is called.
The data structure of FIG. 4 representing a tuple is accessed by both front end 20 and back end 12, and some fields of this structure are limited to only front end or only back end access. Every tuple 35 which can have effects or dependencies will contain one or more longword fields 42 or 43, typically named GEM.sub.-- TPL.sub.-- xxx.sub.-- EFFECTS or GEM.sub.-- TPL.sub.-- xxx.sub.-- DEPENDENCIES. The field names used for particular tuples are described in the section on The Intermediate Language. No code in the back end will ever examine or modify these fields--they are reserved for use by the front end. They are intended as a convenient place to record information which can be used to simplify the coding of the interface routines. There is a similar longword field named GEM.sub.-- SYM.sub.-- EFFECTS in each symbol node of symbol table 30, which is also reserved for use by the front end 20.
For the straight-line dependency interface, a description of the routines will now be given. The front end provides an implementation of the following routines:
GEM.sub.-- SE.sub.-- PUSH.sub.-- EFFECT(EIL.sub.-- TUPLE:in GEM.sub.-- TUPLE.sub.-- NODE)--Pushes the EIL tuple whose address is in the EIL.sub.-- TUPLE parameter onto the effects stack.
GEM.sub.-- SE.sub.-- PUSH.sub.-- EFFECT(EIL.sub.-- TUPLE :in GEM.sub.-- TUPLE.sub.-- NODE)--Pops the topmost EIL tuple from the effects stack. This is guaranteed to be the tuple whose address is in the EIL.sub.-- TUPLE parameter. Of course, this means that the parameter is redundant. However, it may simplify the coding of the POP procedure for a front end that doesn't use a single-stack implementation for the effects stack (see the implementation discussion below).
Returns the most recently pushed tuple whose GEM.sub.-- TPL.sub.-- EXPR.sub.-- COUNT field is greater than MIN.sub.-- EXPR.sub.-- COUNT (see below), and whose execution may change the results produced by EIL.sub.-- TUPLE. Returns null (zero) if no tuple on the stack affects EIL.sub.-- TUPLE. May also return the same tuple specified in the parameter.
Returns the most recently pushed tuple whose GEM.sub.-- TPL.sub.-- EXPR.sub.-- COUNT field is greater than MIN.sub.-- EXPR.sub.-- COUNT (see below), and whose execution may modify the value of variable VVAR.sub.-- SYM. Returns null (zero) if no tuple on the stack affects EIL.sub.-- TUPLE. May also return the same tuple specified in the parameter.
GEM.sub.-- SE.sub.-- PUSH.sub.-- EFFECT and GEM.sub.-- SE.sub.-- POP.sub.-- EFFECT will be called only with tuples which can have effects. GEM.sub.-- SE.sub.-- FIND.sub.-- EFFECT will be called only with tuples which can have dependencies.
There is an order of invocation. Every EIL tuple has a field called GEM.sub.--TPL EXPR.sub.-- COUNT. This field contains the index of the tuple in a walk of the EILG in which basic blocks are visited in dominator tree depth-first preorder. If the back end 12 calls GEM.sub.-- SE.sub.-- PUSH.sub.-- EFFECT with a tuple A, and subsequently calls GEM.sub.-- SE.sub.-- PUSH.sub.-- EFFECT or GEM.sub.-- SE.sub.-- FIND.sub.-- EFFECT with a tuple B, without having called GEM.sub.-- SE.sub.-- POP.sub.-- EFFECT with tuple A in the interim, then it is guaranteed that either tuple A precedes tuple B in the same basic block, or the basic block containing tuple A properly dominates the basic block containing tuple B. Therefore, the EXPR.sub.-- COUNT values of tuples on the effects stack decreases with increasing stack depth (i.e., more recently pushed tuples have higher EXPR.sub.-- COUNTs than less recently pushed tuples). This means that the FIND.sub.-- EFFECT routine can cut short its search of the effects stack as soon as it encounters a tuple T whose EXPR.sub.-- COUNT is less than or equal to the MIN.sub.-- EXPR.sub.-- COUNT argument. This is because all tuples stacked deeper than T are guaranteed to have EXPR.sub.-- COUNTs that are less than MIN.sub.-- EXPR.sub.-- COUNT.
The mechanism actually used for the implementation of the effects stack is entirely up to the front end 20, as is the rule that it uses to determine when the execution of one tuple might affect the value computed by another tuple. A naive stack implementation is certainly possible, though it would probably be inefficient. A more sophisticated implementation might be built around a hash table, so that multiple small stacks (possibly each concerned with only one or a few variables) would be used instead of a single large stack.
The effects-class interface will now be described. Recall that an effects set is a bit vector representing a set of effects classes, and that an effects class represents some arbitrary set of memory locations. Typically, an effects class will represent one of the following:
1. A single named variable. For effective optimization, each simple (i.e., non-aggregate) local variable which is used frequently in a routine should have an effects class dedicated to it.
2. A set of named variables with some common property; for example, in FORTRAN, all the variables in a particular named common block.
3. A set of memory locations which may not be determined until runtime, but which have some common property; for example, all the memory locations which are visible outside this routine (and which might therefore be modified by a routine call); or, in Pascal, all the memory locations which will be dynamically allocated with NEW calls and which have a particular type.
The literal GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS is exported by the GEM.sub.-- SE package. It is the maximum number of distinct effects classes that the front end 20 may define. It will be 128 in the initial implementation. The GEM.sub.-- SE.sub.-- EFFECTS.sub.-- SET type is exported by the GEM.sub.-- SE package. It is a macro which expands to BITVECTOR[GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS]. Thus, given the declaration X: GEM.sub.-- SE.sub.-- EFFECTS.sub.-- SET, the following constructs are all natural (where 0.ltoreq.N.ltoreq.GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS-1):
The interface routines for the effects-class interface will now be described. The front end 20 must provide an implementation of the following routines:
The union of the effects of tuple EIL.sub.-- TUPLE and EFFECTS.sub.-- BV is written into EFFECTS.sub.-- BV.
Writes the set of effects classes that EIL.sub.-- TUPLE depends on into EFFECTS.sub.-- BV.
Writes into EFFECTS.sub.-- BV the set of effects classes that might include the memory associated with variable SYMBOL.
GEM.sub.-- SE.sub.-- EFFECTS will be called only with tuples which can have effects. GEM.sub.-- SE.sub.-- DEPENDENCIES will be called only with tuples which can have dependencies.
The compiler may provide implementations for the interface routines mentioned above, but these routines are not intended for use in a production compiler. They are inefficient, and their rules for when one tuple invalidates another probably will not coincide exactly with the semantics of any particular language. However, they allow useful default optimizations to occur while other components of a front end 20 being implemented.
The EFFECTS field of each symbol node is treated as an effects class number, between 32 and GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS. When the address expression of a fetch or store tuple has a base symbol, the EFFECTS field of the symbol is checked. If it zero, then it is set to a new value between 32 and GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS.
For computing effects sets, using the effects class implementation as described above, the front end must call GEM.sub.-- SE.sub.-- INIT.sub.-- EFFECTS.sub.-- CLASSES before invoking the GEM.sub.-- IL.sub.-- BUILD phase.
This implementation provides information about effects by defining a simple model for effects:
1. No variables are overlaid:
2. Data access operations not in canonical form (as defined in CT.006) have (for stores) or depend on (for fetches) effect 0.
3. Calls have effects 32 through GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS. ARGADR parameters are treated as if the call writes into their address operands.
Effects classes 0 and 32 through GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS are reserved. Effect 0 represents references to memory such that the variables referenced can't be identified (pointer dereferences, parameters, etc.)
When a variable is first referenced using a data access operator in canonical form it is assigned an effects class number n in the range 32 to GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS. The number is recorded in the EFFECTS field of the symbol node. The reference and all subsequent references to that variable will have effect or dependency n.
The implementation includes some hooks for experimentation, testing, etc:
1. Tuples that may have effects or dependencies have one or more "effects fields" (EFFECTS, DEPENDENCIES, EFFECTS.sub.-- 2, etc.) reserved to the front end to record the effects and dependencies of the tuple. The compiler-supplied effects class callbacks interprets an effects field as a bitvector of length 32 representing the first word of a GEM.sub.-- SE.sub.-- EFFECTS.sub.-- SET. That is, if bit n of the field is true, the routines add effects class n to the computed effects of the tuple.
2. The front end can choose the effects class for a variable by writing the effects class number between 1 and GEM.sub.-- SE.sub.-- K.sub.-- MAX.sub.-- EFFECTS into the effects field of the variable's symbol node. The effects class routines do not assign an effects class if the EFFECTS field is not zero.
3. Effects classes 1 through 32 are reserved for use by the front end. It may assign any interpretation to those effects classes.
To use the straight-line dependency implementation discussed above, the front end must call GEM.sub.-- SE.sub.-- INIT.sub.-- EFFECTS.sub.-- STACK before invoking the GEM.sub.-- DF.sub.-- DATAFLOW phase. This implementation uses the information provided by the GEM.sub.-- SE.sub.-- EFFECTS and GEM.sub.-- SE.sub.-- DEPENDENCIES callbacks to determine invalidations. That is, GEM.sub.-- SE.sub.-- FIND.sub.-- EFFECT(X) returns the most recently pushed tuple Y such that the intersection of GEM.sub.-- SE.sub.-- EFFECTS(Y) and GEM.sub.-- SE.sub.-- DEPENDENCIES(X) is non-null.
Induction Variables
According to one feature of the invention, an improved method of treating induction variables in a compiler is provided. First, the definition and detection of induction variables and inductive expressions will be discussed.
An integer variable V is said to be an induction variable of loop L if each store to V that occurs in L:
1. increments (or decrements) V by the same amount each time it is executed.
2. is executed at most once in every "complete trip" through the loop. A trip is "complete" if it flows back to the loop top.
For example, the following code illustrates an induction variable V:
In the compile function, in addition to finding induction variables, we are also interested in inductive expressions. Inductive expressions are expressions that can computed as linear functions of induction variables.
Consider the following program:
The expressions "I*8," "I-4," "T" and "T*4" are all inductive expressions in that they can be recomputed as linear functions of I.
As a brief illustration of some of the optimizations based on induction variables, consider the following program example:
This is a straightforward DO loop, I being the loop control variable. Notice that the inductive expression I*4 increases by 4 on each trip through the loop. By introducing a new variable, I2, we can replace the multiplication with an addition, which is a less expensive operation. This is optimization known as strength reduction, used in optimizing compilers for a long time:
Note that we now have two variables (I and I2) where we used to have one. We can eliminate the original loop control variable completely by recasting the uses of I to be in terms of I2:
This optimization is known as induction variable elimination.
These optimizations (strength reduction and induction variable elimination) operate directly on induction variables. In addition to these optimizations, induction variable detection provides information to other optimizations such as auto-inc/dec, vectorization, loop unrolling, etc.
In the model used in the compiler of FIG. 1, induction variables may be incremented more than once during the loop. Furthermore, the number of changes can even differ with each iteration. In fact, the number of changes can be zero for a particular iteration. The loop invariant increment value may differ between individual stores, but each individual store must increment the variable by the same amount whenever it is executed.
There are several different categories of inductive variables, with different properties, including basic induction variables, inductive expressions, and pseudo induction variables.
Basic induction variables are the simplest form of induction variable. They have known properties that apply throughout the loop. All other induction variables and expressions are always built up as linear functions of a basic induction variables. Basic induction variables are generally modified in the form I=I+q or I=I-q where "q" is loop invariant. More generally, however, the requirement is that the assignment be of the form I=f(I) where f(I) is a linear function of I with a coefficient of 1.
The basic induction variables of a particular loop are represented by a set in the loop top. In addition to this set, we also maintain the set of basic induction variables in the loop that have conditional stores that may not be executed on every trip through the loop. This inhibits vectorization and can make strength reduction more "desirable."
An inductive expression is either a reference to an induction variable or a linear function of another inductive expression. Inductive expressions must be in one of the following forms:
where f(I) and g(I) are inductive expressions derived from basic induction variable I with respect to loop L and E is invariant in loop L. If there are no stores to I between f(I) and the arithmetic operator of which it is an operand, then the arithmetic operator is an inductive expression derived from basic induction variable I with respect to loop L.
The other category is pseudo induction variables. Under certain conditions, a variable may behave like an induction variable on all but the first trip through the loop. These can be turned into induction variables (and thus vectorized) by peeling the first iteration of the loop. Such variables are referred to as "pseudo induction variables." This occurs when a fetch within the loop is reached only by two stores, one within the loop that defines a derived induction variable, and another store whose value flows in through the loop top. Additionally, it must be guaranteed that all stores within the loop are executed once per trip.
On the first trip through the loop, D has the value 50 at the assignment to I. On subsequent trips, D has the value 5,6,7, etc. By unrolling the loop once, the subsequent trips can be vectorized. Note that the algorithms given herein do not find induction variables that are pseudo induction variables.
In order to identify a basic induction variable the compiler must be able to recognize all stores to it. The absence of the "has aliased stores" attribute guarantees this and thus we only recognize basic induction variables that do not have "has aliased stores."
Detection of basic induction variables requires the use of "sets" of potential induction variables. Doing this dynamically for each loop is an expensive and complicated operation. Instead, we will use the side effect sets used to construct IDEF sets.
A variable "X" is said to be "in" IDEF set S if the all the effects that fetch's of X depend on are in S. That is, X is in IDEF set S only if GEM.sub.-- SE.sub.-- DEPENDENCIES(X) is a subset of S.
Note that the presence of X in a basic induction set implies only that:
(a) X is a basic induction variable or
(b) X is loop invariant and shares IDEF bits with at least one variable that is a basic induction variable.
The algorithm descriptions used may take the following liberties (perhaps more) in the interest of keeping the algorithm description simple: (1) The collection of the constant pans of the linear function cannot cause an overflow. (2) All stores completely redefine the variable.
The algorithm starts out by assuming that all variables modified in the loop are basic induction variables. Each loop top has a basic induction variable set. As we find stores that don't satisfy the requirements for basic induction variables, we eliminate variables from the basic IV set of the loop top.
Since inductive expressions and derived induction variables are always functions of basic IVs, we might say that fetches of basic IVs are the atomic forms of inductive expressions. That is, for an expression to have the inductive property it either has inductive operands, or it is a fetch of a basic induction variable.
Using the rules given earlier, we build up inductive expressions from simpler inductive expressions based on assumptions about basic IVs. The basic IV of an inductive expression is always retained with the expression. Thus, after the algorithm has run, we can tell whether the expression is truly inductive by checking to see that the basic IV from which it is derived is still in the basic IV set of the loop.
The FIND.sub.-- IV algorithm will become part of the DATAFLOW phase which does a depth first dominator tree walk.
The following is a summary overview of the tuple processing that is done: select TUPLE[OPCODE}
[FETCH]
If base symbol is still a basis IV candidate
then
mark this tuple as being inductive.
[STORE]
Let V be the base symbol of the store.
If the value being stored is not inductive or.sub.-- else
the basic IV of the inductive value being stored is not V or.sub.-- else
the coefficient of the stored value is not 1
remove V from the basic IV set of the loop top
then
remove V from the basic IV set of the loop top
then
mark the store as being inductive
[ADD, SUB, MUL, etc.]
If one operand is inductive and other operand is loop invariant
then
mark this tuple as being inductive
The fields added to the tuple data structure, and fields added to the flow nodes, to accommodate induction variable detection, are set forth in Table 6a.
Automatic Creation of Kfold Routine
As previously discussed, the programming language compiler of FIG. 1 translates programs written in a source language into the machine language of a target machine 25. The compiler includes a front end 20, which incorporates knowledge of the source language in module 21 being compiled, and a back end 12, which incorporates knowledge of the machine language of the target machine 25. The front end translates programs from the source language into the intermediate language of the ILG 55, and the back end translates programs from the intermediate language into the target machine language.
The intermediate language generally specifies a collection of operators (for example, add, shift, compare, fetch, store, or tangent), a collection of data types (for example, "signed 32-bit integer," "IEEE S-format floating point," or "character string"), and a representation for values of those data types.
One of the optimizations included in the optimizer 26 is a constant expression evaluation routine. An example of a source code listing that may be related to a constant expression is shown in FIG. 6, where A and B are found to be constants, so A+B is a constant, then I and J are both equal to the same constant. The compiler can do the calculation (A+B), and save the fetch of A and B separately at run time, as well as saving the ADD operation. The I=A+B and J=A+B expressions of the code of FIG. 6 are thus both represented as merely STORE #9,I or STORE #9,J. This is known as "constant folding" because the constants are detected, calculated at compile time, and "folded" into the object code image. The mechanism for doing this is part of the optimizer 26, referred to as a Kfold routine.
The compiler of FIG. 1 incorporates a Kfold routine for evaluating expressions of the intermediate language to find these constant expressions. In general, given an operator of the intermediate language and the values of its operands, this routine will yield the same value which is computed by that operator when applied to those values. Such a constant expression evaluation routine has many applications in a compiler. For example,
(a) The execution speed of the machine code which is generated for a program may be improved if some expressions of the program can be evaluated by the compiler itself rather than when the program is executed.
(b) Some source languages may allow the use of expressions with constant operands to represent constant values. Compilation of a program in such a language requires the evaluation of such expressions by the compiler.
(c) If the repertoire of operations provided in the intermediate language is richer than the set of operations provided by the programming language or environment in which the compiler is implemented, the most convenient way to perform some computation in the compiler may be to represent it as an expression in the intermediate language and submit it to the constant expression evaluation routine.
The implementation of a constant expression evaluation routine may be a matter of considerable difficulty. The IL may have dozens of operations (e.g., ADD, SUBT, COSINE, etc.), and when distinct data types are considered (e.g., INT32, NINT64, FLOAT& etc.), an intermediate language may have hundreds or thousands of distinct operators. The evaluator must be able to apply each of the operations to each of the data types correctly, lest the compiler fail to perform its function fully or correctly. Particularly when floating-point types are involved, it is likely that not all of the operations which can be represented in the intermediate language will be directly available in the programming language in which the compiler is implemented. Consequently, a constant expression evaluation routine is liable to be extremely long, containing hundreds of distinct cases, and be highly error-prone.
According to an important feature of one embodiment of the invention, the crucial point is that the one language in which the precise meaning of an operator of the intermediate language can always be specified both tersely and precisely is the intermediate language itself. That is, the compiler back end itself must be capable of generating code which correctly implements any operator of the intermediate language. Another way to say this is that compiler back end already embodies the knowledge of the sequences of machine code instructions necessary to realize the effect of each intermediate language operator, and it would be redundant to have to encode this same knowledge again in a different form in the constant expression evaluation routine.
Based upon this concept, according to the invention, the mechanical generation of a constant expression evaluation routine becomes straightforward: The first step is to create a new compiler of FIG. 1, which uses the same back end 12 as the regular compiler, but replaces its front end 20 with the special front end described below. (Equivalently, provide a special mode for the compiler in which it operates as described below.)
Second, the special front end 20 or special mode of operation does not read and translate a source program 21. Instead, it generates the intermediate language for the constant expression evaluation routine, as follows:
(a) The routine does a conditional branch to select a case based on the intermediate language operator specified in the argument list.
(b) Each case contains the code for a single operator. It fetches the operand values from the routine's argument list, applies the operator to them, and returns the result.
(c) Since the routine is being generated directly in the intermediate language, the code for each case simply consists of intermediate language operators to fetch the operands from the argument list, then the intermediate language operator for the particular case, and then the intermediate language operators to return the result.
Third, when this intermediate language graph is submitted to the compiler's back end, it will generate machine code for the constant expression evaluation routine.
In the special front end just described, the front end can contain a list of all the operators for which cases must be generated, and can mechanically generate the intermediate language for each case.
However, the process can be further simplified if, as may often occur, the compile back end already contains a table of operator information. (For example, such a table may be used to check the correctness of the intermediate language graph generated by the front end.) It is then possible for the special front end to use this table, already provided by the back end, to determine which cases to be generated.
Type Definition
The compiler of FIG. 1 uses a type definition module referred to as the GEM.sub.-- TD module. GEM.sub.-- TD provides the mechanisms used by a front end 20 and back end 12 in constructing program type information to be incorporated in an object module for use by a linker or debugger. It is intended that this type specification service will allow a front end 20 to describe program symbols and their associated type information to the object module builder 29 in a manner independent of target object file requirements. This type specification service acts as a procedural "grammar of types" so that the compiler may associate abstract type specifications and program symbols. The type specification interfaces are defined below, and a number of examples of the use of the GEM.sub.-- TD services are referenced.
The creation of type information takes place in the context of symbol table 30 creation and allows a front end 20 to specify an abstract representation of program type information. The object module builder 29 will later use this information in constructing Debug symbol table information.
The GEM.sub.-- TD module provides service routines that allows a front end 20 to describe basic types and derived types. These routines typically construct internal data structures describing the specified type information. A new compiler node type, GEM.sub.-- TDI, will be defined to manage this type information. The definition of the type node data structure is private to the compiler 12 and may not be altered or examined by the front end 20. When defining a type, the front end 20 is returned a "handle" to the type node by the GEM.sub.-- TD routine defining the type. The handle allows a front end to associate a type with a program symbol but prohibits it from altering or examining the fields of the data structure.
Type nodes will be created and managed by scope, that is, when transmitting type information, a front end 20 will specify the block node that a type is to be declared within, and the shell will be responsible for the management of the type nodes within that scope. The shell will manage type nodes in a list rooted in the block node in which the type is defined. The block node data structure will be expanded to define the fields TYPE.sub.-- LIST.sub.-- HEAD and TYPE.sub.-- LIST.sub.-- TAIL.
A front end 20 may choose to make on-the-fly calls to the type specification service routines or may choose to make a pass over the entire symbol table to generate the type information.
After defining a type the front end must associate this type information with the symbols of that type. Symbol nodes will have a new field DST.sub.-- TYPE.sub.-- INFO used to associate a symbol with its type. A symbol's DST.sub.-- TYPE.sub.-- INFO field will contain the address of the type node handle returned by a GEM.sub.-- TD service. A symbol node with a DST.sub.-- TYPE.sub.-- INFO value of null will have the target specified behavior for symbols not having type information.
Referring to FIG. 7, the data fields and relationships are illustrated for the function:
A block node 60 for toy-proc contains fields 61 and 62 (decl list pointers) pointing to the entries 63, 64 and 65 in the symbol table 30. Also, it contains fields 66 and 67 functioning as type list pointers, pointing to the entries 68 and 69 in the type list for int and float. The entries 63, 64 and 65 also have pointers 70, 71 and 72 pointing to the entries 68 and 69, for int and float, as the case may be.
The GEM.sub.-- TD type specification service consists of routines to allow a front end 20 to define standard and derived types and to associate those types with program symbols. The compiler back end 12 will use the resulting type definitions and their symbol node associations to generate target specified Debug Symbol tables. Note that boolean is not considered a basic type. Compilers for languages such as Pascal should define boolean as an enumeration containing the elements true and false.
Action Language for Multipass Code Generator
A method for doing code generation in the back end 12 by code generator 29 using code templates will now be described. The selection and application of code templates occurs at four different times during the compilation process.
1. The PATSELECT phase does a pattern match in the CONTEXT p ass to select the best code templates. (During this pattern match the UCOMP and DELAY optimization tasks are done in parallel as part of the pattern matching process.)
2. The TNASSIGN and TNLIFE tasks of the CONTEXT pass use context actions of the selected templates to analyze the evaluation order to expressions and to allocate TNs with lifetimes nonlocal to the code templates.
3. The TNBIND pass uses the binding actions of the selected templates to allocate TNs with lifetimes local to the code templates.
4. Finally, the CODE pass uses code generation actions of the selected templates to guide the generation of object code.
A template is used at different times during a compilation. It consists of three major components:
1. ILG Pattern--which guides the template selection process that matches templates to applicable ILG structures.
2. Undelayed Actions--which determine the processing of matched ILG structures during the CONTEXT, TNBIND and CODE passes. The undelayed actions are performed when the template is first processed in each pass. As a result, the template actions for each ILG node are processed three different times--once for each pass. Some of the actions will have meaning for only one pass and will be ignored in the other passes. Other actions will have meanings in more than one pass but the required processing will be different in each pass.
3. Delayed Actions--which also determine the processing of matched ILG structures during the CONTEXT, TNBIND and CODE passes. The delayed actions are performed each pass when the result computed by the template is first processed as the leaf of another template. Delayed actions are useful on target machines like a VAX that have address modes. Simple register machines like a RISC would probably not make heavy use of delayed actions.
An ILG pattern of a code generation template consists of four pieces of information:
1. A result value mode (see the examples given in the Appendix) which encodes the representation of a value computed by the template's generated code.
2. A pattern tree which describes the arrangement of ILG nodes that can be coded by this template. The interior nodes of the pattern tree are IL operators; the leaves of the pattern tree are either value mode sets or IL operators with no operands.
3. A sequence of Boolean tests. All of these must evaluate to true in order for the pattern to be applicable.
4. An integer that represents the "cost" of the code generated by this template.
The pattern matches or PATSELECT phase matches an ILG subtree with the pattern of a template. If more than one template pattern can be applied at an ILG node then the pattern matcher delays choosing between the alternative templates until it knows which one leads to the lowest estimated code cost.
There are three different action interpreters--the CONTEXT interpreter, the TNBIND interpreter and the CODE interpreter. The actions of each template are performed in three different passes of the compiler by the appropriate interpreter. Although the identical template is used in all three passes, the semantics of the actions are phase dependent so that different things are done each pass. Many actions have meanings in only one of the three passes and they do nothing in the other two passes. Other actions have meanings in more than one pass but the semantics of an action in one pass are often very different from the semantics of the same action in a different pass. However, having only one action sequence in a template makes it very easy to understand and to maintain the dependencies between the various passes.
The action sequence for each template consists of two parts--the undelayed actions and the delayed actions. When a pattern of selected ILG nodes is first processed the undelayed actions are interpreted. When the ILG pattern is later used as the leaf of another ILG pattern then the delayed actions are interpreted.
At the start of interpreting the undelayed actions a table of operand variables is created. An operand variable can contain a temporary name (TN), a literal or a target specific address mode.
Temporary names are each partitioned into one of three classes: (1) permanent TNs, (2) delayed TNs and (3) local TNs. The class of a TN is determined by its lifetime and usage.
Each TN must have an allocation lifetime. The allocation lifetime is begun by the appropriate template action and extends along all flow paths leading to the last use of the TN. The TNs in the permanent class can have a lifetime that ends some arbitrarily large amount of code into the future after creation of the TN. The life of a delayed class TN must begin n a delayed action of a template and terminate shortly afterwards when the TN is used as a leaf. The life of a local TN never extends beyond the interpretation of a single pattern.
The class of a TN determines how it is processed. Permanent class TNs are created once in the CONTEXT pass and the same TN data structure is kept through all three passes and is used to store the complicated lifetime description of the TN. Delayed class and local class TNs have lifetimes of very restricted duration so they do not need a permanent data structure to track this information. As a result, the TN data structure for delayed class and local class TNs are built each pass when interpreting the actions and deleted immediately after their last use in each pass. Interpreting the same action sequence in each pass guarantees identical TN data structures are built in each pass for TNs of these classes.
There will be a large list of different template actions. Some of the actions will be target machine dependent. The Appendix contains a list of proposed or example template actions, so that a user can by these code template examples-determine for a particular embodiment what will be needed.
The Intermediate Language Representation
The internal representation used in the compiler framework 10 of FIG. 1 comprises the symbol table 30 and intermediate language graph 55, which are the data structures created by the front end 20 to represent the structure, data, and code of a source module 21. The following describes the nodes which are the primitive components of these data structures, including a specification of the symbol table 30 and intermediate language used in the IL graph 55. In a compiler as described with reference to FIG. 1, the front end 20 generates a symbol table 30 to describe the blocks, routines, variables, literal values, etc. of a program contained in source module 21, and one or more intermediate language graphs 55, to describe the executable code. The following describes these internal data structures.
The design of the compiler of FIG. 1 in general, and of the intermediate language and symbol table in particular, is intended to address a variety of architectures ranging from "Complex Instruction Set Computers" (CISC) such as VAX to "Reduced Instruction Set Computers" (RISC) such as PRISM, MIPS (a 32-bit RISC machine), or an advanced 64-bit RISC architecture. This design does assume that the architecture of target machine 25 has certain basic features. First byte organization and addressability are assumed and Twos-complement binary arithmetic, with "Little-endian" bit ordering. "Reasonable" address representation is also assumed, i.e., that an address fits in a register.
In general, the front end 20 can be oblivious to the details of the target architecture 25 when creating the intermediate representation of a program. Most constructs of the intermediate representation have a well-defined meaning which is independent of the target architecture 25. There are some issues that must be resolved in implementing the front end 20, however. First, not all data types will be available on all architectures, as explained below. Second, arithmetic overflow behavior and the representation of "small integer" arithmetic may vary on different architectures, again, as discussed below. Third, the behaviors of some operators (such as the arithmetic shift operators) are defined only for subranges of the operand values for which the underlying machine instructions are defined on particular architectures. For operand values outside this specified range, such operators may be well behaved for any particular machine, but may have different behaviors on different machines. Lastly, calling conventions will be different on different target systems 25, requiring the front end 20 to generate different intermediate representations for the same source language constructs in some cases.
The phrase "Intermediate Language" refers to an abstract language for specifying executable code. An "Intermediate Language Graph" (ILG) 55 is a particular program expressed in this language.
The intermediate language in graph 55 is really a language of data structures in memory, with pointers providing the syntactic structure. However, there is also an approximate textual representation for ILGs, used for IL dumps written by the compiler as a debugging aid.
The primitive concept of the IL is the tuple as described above with reference to FIG. 4--an ILG 55 is made up of tuples 35 representing the operations to be executed. These tuples 35 are tied together by pointers (e.g., operand pointers 38) which represent various relations. The most important relations are the operator-operand relation (a pointer 38 from an operator to each of its operands) and the linear ordering on all the tuples in each basic block of the ILG, which provides a nominal execution order. This linear order is represented by the tuple number 40 within a block, and by the pointers linking all the blocks of a routine or module.
The computation defined by an ILG 55 is as follows:
(1) Start at the BEGIN tuple of the ILG.
(2) Evaluate each tuple in linear order: fetch the saved results of its operands, compute and save its result, and perform any secondary action that may be defined for it. (There are exceptions to this simple evaluation rule for "flow boolean" and "conditional selection" operators.)
(3) After evaluating a branch tuple, continue evaluation at the label tuple selected by that branch tuple.
It should be understood that these rules define the "meaning" of an IL graph 55. The code generator 29 is allowed to rearrange the actions indicated by the ILG, so long as it preserves their dependencies, as specified by the following rules:
(1) If the ILG 55 contains an expression, and a statement whose execution might affect the value computed by evaluating the expression, then the generated code for the expression and the generated code for the statement must be executed in the same order that the statement and the expression occurred in the ILG.
(2) If the ILG 55 contains two statements whose execution might affect the value computed by evaluating some common expression, then the generated code for the two statements must be executed in the same order that the statements occurred in the ILG.
question of when the execution of a statement might affect the value computed by the evaluation of an expression is resolved by reference to the side effects mechanism described below.
The ILG 55 constructed by the front end 20 is not the same as the ILG processed by the back end 12. The front end 20 generates a Compact IL Graph, while the back end 12 processes an Expanded IL Graph. When the back end 12 generates code for a routine, the first thing it does is to expand that routine's CILG into an EILG. The differences between the two forms are several. First, the CIL provides "shorthand" tuples, which are expanded into sequences of lower-level tuples in the EIL. Second, the nodes which represent EIL tuples have many more fields than the nodes which represent CIL tuples. The additional fields contain information which is used by the back end 12, but which can be computed by the IL expander (or by other back end phases) from the fields in the CIL nodes. Third, there are different structural restrictions on the CILG and the EILG. This description is directed to the compact IL, although this information generally pertains to both the CIL and the ElL.
The structure of a symbol table 30 represents the structure of the module 21 being compiled. At the heart of the table 30 is a tree of block nodes representing the blocks, routines, and lexical scopes of the module 21; the tree structure represents their nesting relationship. Associated with each block node is a list of the symbol nodes which are declared in that block. Associated with each routine block is an ILG 55 representing the code for that routine. A symbol node represents a symbolic entity in the module, such as a variable, label, or entry point. Constant values in the module 21 being compiled are represented by literal nodes. Literal nodes may be referred both from the symbol table 30 and from ILGs 55. The term literal table is also used to refer to the collection of all literal nodes that have been created in a compilation. Frame nodes represent areas of storage in which code and data can be allocated. Generally, these are either the stack frames of routines or PSECTs. Parameter nodes are used to build parameter lists, which are associated with entry point symbols. Each parameter node relates a parameter symbol in a routine with a location in the argument list of an entry point.
Data Types
The intermediate representation used in graph 55 describes a program for an abstract machine 25, which has only a small set of types, the data types which are described in the following list. These data types are distinct from the data types of the source language of module 21, which are relevant only to the front end 20. It is the responsibility of the front end 20 to determine, for each target machine 25, the data types to be used to represent each source language data type.
Data Types
Null
Representational
Scalar
Address
Signed Integer
Unsigned Integer
Floating Point
Complex
Boolean
The null data type is a special data type, which is the type of tuples that do not compute a value. A representational data type is a type whose values have a specific representation in the target machine architecture. The representational data types are divided into scalar data types and aggregate data types. A scalar data type is one whose values can be represented in a small fixed number of memory locations or registers. The scalar data types are subdivided into the address data type and the arithmetic data types. Note that the arithmetic types may be used to represent any other kind of data than can fit in the appropriate number of bits. In particular, source language character and logical data types must be represented with integer data types. There is a single address data type, ADDR. A value of type ADDR is represented as a binary integer with 32 or 64 bits.
There are signed integer data types INT8, INT16, INT32, and INT64, where a value of type INT.sup.x-1 is represented as a signed binary integer with .sup.x-1 bits, and is therefore in the range--(2.sup.x-1) . . . (2.sup.x-1 -1). The type INT8 may also be referred to as IBYTE. The type INT16 may also be referred to as IWORD. The type INT32 may also be referred to as ILONG. The type INT64 may also be referred to as IQUAD. The integer type with the same number of bits as an address may also be referred to as lADDR. The largest signed integer type supported for the target architecture (INT32 or INT64) may also be referred to as IMAX. Any binary scaling (as in PL/I) must be provided by the front end--there are no IL provisions for a scaled binary data type.
There are unsigned integer data types UINT8, UINT16, UINT32, and UINT64, where a value of type UINT.sup.x-1 is represented as a signed binary integer with .sup.x-1 bits, and is therefore in the range 0 . . . (2.sup.x -1). The type UINT8 may also be referred to as UBYTE or as CHAR8. The type UINT16 may also be referred to as UWORD or as CHAR16. The type UINT32 may also be referred to as ULONG. The type UINT64 may also be referred to as UQUAD. The unsigned integer type with the same number of bits as an address may also be referred to as UADDR. The largest unsigned integer type supported for the target architecture (UINT32 or UINT64) may also be referred to as UMAX.
The floating point data types are the VAX floating point types, REALF, REALD, REALG, and REALH, and the IEEE floating point types, REALS, REALT, REALQ, and REALE. Not all of these will necessarily be supported on any particular target architecture.
The complex data types are CMPLXF, CMPLXD, CMPLXG, CMPLXS, and CMPLXT. A complex value is represented as a pair of values of the corresponding real type, which represent the real and imaginary pans of the complex value. Only complex types which correspond to supported floating point types will be supported on a particular target architecture.
A value of an aggregate data type consists of a sequence of contiguous elements. An aggregate value is characterized by its body, the actual sequence of elements, and length, the number of elements in the sequence. The aggregate types are:
(a) Character strings, type STR8, which have elements of type CHAR8.
(b) Extended character strings, type STR16, which have elements of type CHAR16.
(c) Bit strings, type BITS, whose elements are single bits, packed as tightly as possible.
(d) PL/I and COBOL decimal strings, type DECIMAL, whose elements are decimal digits (represented as four-bit BCD digits, packed two per byte, with a leading sign digit). (The DECIMAL value is characterized by its precision, the number of digits it contains (not counting the leading sign digit), and its scale, the number of those digits which are regarded as coming after the decimal point.
The elements of an aggregate value are numbered starting at zero. (Note that this will require many front ends to subtract one when translating a source program string index to an IL string index.)
There is no limit on the number of elements which may be processed in a string operation. A flag might be introduced in the future to allow the front end to indicate character string expressions whose lengths were guaranteed not to exceed 65535 characters, and which could therefore be computed efficiently with the VAX character string instructions.) The length word of a varying-length string in memory will still be only 16 bits. Decimal strings are limited to 31-digits (plus the sign digit) on all target architectures.
An example of the details of the representational type system for the various target architectures is indicated in Table 6.
There is a single Boolean data type, BOOL. This is the type of logical values computed during the execution of a program. It does not have a specified physical representation. For example, a Boolean value might be represented by the value of a binary integer, the value of a processor condition code, or the value of the processor program counter. In particular, type BOOL does not correspond to any logical or Boolean data types that may be present in a source language. These must be represented as INT or UINT values, and convened to and from type BOOL as necessary.
The general features that are common to all tuples in the intermediate language, and the structural characteristics of ILGs 55 (routines in the intermediate language) will now be described.
An ILG 55 is made up of IL tuple nodes (usually just called tuples). All tuples contain the fields listed in Table 7. Other fields, known as attributes, occur only in particular kinds of tuples.
Unlike symbol table nodes, which may be allocated with an arbitrary amount of space reserved for use by the front end 20, CIL tuple nodes will contain only the fields specified here. EIL tuple nodes will contain additional fields, located at a negative offset from the tuple node address, which are private to the back end 12.
Structure of the ILG
One tuple in an ILG can refer to another tuple in two different ways: as an operand or as an attribute. When only the operator-operand relation is considered, a CILG is directed acyclic graph (DAG), while an EILG is a forest (i.e., a collection of trees).
Attribute pointers 39 create additional structure on the ILG, and also allow references from the ILG to the symbol table 30. The most important structural relation is the linear order of the ILG, defined by the next tuple and prey tuple attribute pointers. All of the tuples in a CILG occur in a single list defined by the linear order. The tuples of an EILG occur in a collection of circular lists, one for each basic block.
The following rules apply to the structure of an ILG. If a front end 20 creates a CILG which violates these rules, the results are unpredictable, although the back end will attempt, where convenient, to detect violations and terminate compilation:
(a) A tuple whose result type is NULL is referred to as a statement tuple, and a tuple whose result type is not NULL is referred to as an expression tuple.
(b) In the CIL:
(i) A scalar or Boolean expression tuple may be an operand of one or more other tuples. An aggregate expression tuple must be used as an operand of exactly one other tuple, which must be in the same basic block (see below).
(ii) An operand may be an expression tuple, a symbol node, or a literal node.
(iii) A symbol node used as an operand always has type ADDR. A literal node used as an operand has the data type of the literal.
(iv) A symbol representing a variable which is allocated to a register does not have an address, in the normal sense. However, such a symbol may be used as the address operand of a tuple which reads from or writes to memory (a FETCH or STORE), in which case the tuple will access the indicated register.
(v) If a symbol represents a variable in a stack frame, then that stack frame must be associated with the current routine or one of its ancestors in the symbol table block tree; otherwise, there would be no way of finding the stack frame at execution time.
(c) In the EIL operands must be expression tuples, and every expression tuple must be an operand of exactly one other tuple.
(d) No statement tuple may be an operand of any other tuple.
(e) A tuple which is an operand of another tuple must precede that tuple in the linear ordering of the ILG. (In an EILG, this means that the operand and the operator must occur in the same basic block.)
(f) An expression tuple must dominate every tuple which it is an operand of. That is, it must be impossible to get from an entry point of a routine to a tuple without encountering every operand of that tuple on the way.
Subsequent paragraphs in this section describe the sons of operations that are available in the intermediate language and the operators that are used to represent them. The individual operators are all collected in a data structure called <REFERENCE>(part.sub.-- tuple.sub.-- dictionary), the tuple dictionary. Each operator in the dictionary is documented using a structured format. Table 8 discusses the main categories in this format, the information presented under each, and the format used to present the information.
The format section of a tuple specifies the number of operands and the allowed operator, operand, and result types in a single line of the form:
op.type(type-1, . . . , type-n): result
where op is the name of the tuple operator, and type specifies the allowable operator types. If ".type" is omitted, then the operator type must be NULL. Otherwise, type must be on eof the following:
(a) A specific type name (ADDR, BOOL, BITS, IADDR, etc.) indicates that only the specified type is allowed.
(b) INT, UINT, REAL, CMPLX, or STR indicates that any type belonging to the specified family is legal. For example, CMPLX means that CMPLXF, CMPLXD, CMPLXG, CMPLXS, and CMPLXT are all allowed; STR means that STR8 and STR16 are allowed.
(c) ALL indicates that any type other than NULL is legal.
(d) A string of the letters I, U, R, C, A, S, and B indicates that any type belonging to a family represented by one of the letters is allowed, as follows:
The expressions Type-1, . . . , Type-n" specify the allowable types of the tuple's operands. If the parenthesized list is omitted, then the operator takes no operands. Otherwise, the tuple must have one operand for each type in the list. Each type-i must be one of the following:
(a) T means that the operand type must be the same as the operator type.
(b) A specific type name (ADDR, BOOL, BITS, IADDR, etc.) means that the operand must have the specified type.
(c) A string of the type code letters I, U, R, C, A, S, and B has the same meaning that it does for the type specifier. Note that operands with the type specifier IU, which means "any integer," are generally converted to type IMAX in the generated code. Program behavior is therefore undefined if the actual value of such an operand cannot be converted to type IMAX.
(d) If the operator and operand type specifiers are REAL and CMPLX or STR and CHAR, then the actual operator and operand types must be consistent. For example, the type specification "CADD.CMPLX(T, REAL): T" indicates that the second operand must have type REALF if the operator type is CMPLXF, REALS if the operator type is CMPLXT, etc. If the operator type is SB, i.e., character string or bit string, and an operand type specifier is CHAR, then the operand type must be CHAR8 if the operator type is STR8, CHAR16 if the operator type is STR16, and IMAX if the operator type is BITS. That is, IMAX is treated as the character type corresponding to the string type BITS.
The actual operands of the tuple must be tuple nodes whose result types are consistent with the types specified by the operand type list. In the CIL, they may also be symbol nodes, which are always treated as having type ADDR, or literal nodes, which are treated as having the types specified by their data type fields.
The expression "Result" specifies the allowable result types. If it is omitted, then the operator is a statement operator and the tuple's result type must be NULL. Otherwise, it is interpreted exactly the same way as the operand type specifiers.
Addresses and Memory References
An address expression is one of the references in the intermediate language. The simplest form of address expression is a symbol. That is, an operand field of a tuple node may contain the address of a symbol node, to represent the memory address (or the register) associated with that symbol. An address value can also be obtained by fetching it from memory (a "pointer variable"), by casting an arithmetic value, or by evaluating a preincrement tuple, a postincrement tuple, or one of the tuples of the following list:
A data access tuple is a tuple which causes a value to be loaded from or stored into memory. (The word "memory" here includes registers in a register set of the target CPU 25. The only difference between a register and a normal memory location of the CPU 25 is that the "address" of a register can only be used in a data access tuple.) The data access operators are listed in Table 9.
In every data access tuple, the first operand is an address expression. Every data access tuple also has an offset attribute which contains a longword integer. The address of the memory location to be accessed is the sum of the run-time address operand and the compile-time constant offset attribute.
All data access tuples will have some or all of the attributes listed in Table 10. The uses of the effects, effects2, and base symbol attributes are discussed in more detail below in the section Interface for Representing Effects.
Another type of reference is the Array Reference. The APLUS and AMINUS tuples are sufficient for all address computations. However, they do not provide any information about the meaning of an address computation. In particular, they don't provide any information about array references and subscript expressions that might have been present in the source code. This information is needed for vetorization. Therefore, the IL has tuples which specifically describe array references.
For example, given a BLISS vector declared as local X: vector[20,long], a reference to .X[.I] could be represented as
Given a Pascal array declared as var Y:packed array [1 . . . 10, 1 . . 10] of 0 . . . 255, an assignment Y[I, J]:=Z could be represented as
The basic array reference operators are AREF and SUBSCR. AREF yields the address of a specified element in an array. SUBSCR computes the offset of an array element.
The first operand or an AREF tuple is an address expression representing the base address of the array, and its second operand is a SUBSCR tuple which computes the byte offset from the base address to an element of the array. The AREF tuple adds the value of the SUBSCR tuple to the base address to compute the address of the indexed element. In fact, the code for AREF(origin, subscript) is identical to the code for APLUS(origin, subscript).
A SUBSCR tuple computes the offset of an element along one dimension in an array. Its operands are:
(a) The element index. Individual indices in a subscript expression are not normalized for a zero origin. Instead, an origin offset to account for non-zero lower bounds in the array declaration should be added into the address operand of the AREF tuple or the offset field of the tuple that uses the element address.
(b) The stride. This is the difference between the addresses of consecutive elements along the dimension. For a simple vector of longwords, the stride would be a literal 4, but for multidimensional arrays, the "elements" of the higher dimensions rows (or larger cross-sections) of the array.
(c) An expression for the remainder of the subscript expression (that is, for the remaining indices in the subscript expression). This must be either another SUBSCR expression or a literal node representing the integer constant zero.
The code for SUBSCR(index, stride, remainder) is identical to the code for ADD(MUL(index, stride), remainder).
A SUBSCR tuple also has a position attribute, which indicates the position of the index in the subscript list of the array reference. It is required that a position number identify the same subscript position in all references to a given array. For the most effective vectorization, it is recommended that position 1 ought to be the most rapidly varying subscript, position 2 the next most rapidly varying, etc.
There are several tuple operators that don't really fit in any other section; These miscellaneous operators are the following:
Arithmetic Tuples
The arithmetic tuples are used to manipulate "arithmetic" values--integers, real numbers, and complex numbers. This includes fetching, storing, and conversions, as well as traditional arithmetic operations such as addition and multiplication.
The shift instructions in the VAX and RISC architectures are so different from one another that a fully abstract IL shift operator would be certain to generate inefficient code on one or both architectures. On the other hand, the IL has to support shifting, since many source languages have some sort of shift operators. As a compromise, the IL provides the following operators (None of the shift operators will ever cause an arithmetic overflow exception.):
(a) SHL, SHR, and SHRA do a left shift, a logical right shift, and an arithmetic right shift, respectively, and require a positive shift count. (That is, their behavior is undefined if the shift count is negative.) These support the C shift operators, and map directly into the RISC architecture shift instructions.
(b) SH does a left shift if its operand is positive, or an arithmetic right shift if its operand is negative. This supports the BLISS shift operator, and maps directly into the VAX shift instruction.
(c) ROT is the rotate operator. Although it is described differently in the VAX and RISC architectures, the actual behavior in all cases can be characterized as a left rotation whose count is specified by the least significant n bits of the count operand, where n is the base-two logarithm of the register size. (For example, on VAX and MIPS the rotate count is the least significant five bits of the count operand.)
Integer overflow is another feature to consider. Them is a problem in attempting to specify the sizes for integer arithmetic in the IL so that, for all target machines, code will be generated that will satisfy the semantics of the source language and will be as efficient as possible subject to the constraints imposed by those semantics. In particular, some machines (such as VAX) will happily do byte and word arithmetic, while RISC machines typically do only longword arithmetic. Doing all the size conversions would be wasteful on a VAX, but emulating true byte or word arithmetic would be inefficient on a RISC machine.
The following rules are intended to allow the code generator sufficient flexibility to generate reasonable code for all target machines (Everything that is said about INT types below applies equally to UINT types.):
(a) If the result type of an expression is INT.sup.x-1, the compiler may actually perform the indicated computation with y-bit arithmetic, where y.gtoreq.x. This might produce a y-bit result with more than x significant bits, if the original x-bit computation would have overflowed. For example, an ADD.INT16 might be implemented with a 32-bit add. 20000+30000 results in an overflow when done as a 16-bit add, but produces the legal 32-bit number 50000 when done as a 32-bit add.
(b) Every arithmetic operator has a suppress overflow flag (which is only meaningful when the tuple result type is INT or UINT). If this flag is set, then the code generated for a tuple must not report any sort of overflow condition, regardless of the results of the computation, and may ignore the possible presence of extraneous high-order bits in the result (except when the result is used as the operand of an XCVT tuple). Note that the suppress overflow flag is defined in tuples (such as IAND) for which overflow could never occur anyway. Suppressing overflow for these tuples will be particularly easy. The suppress overflow flag is intended for situations where it would be semantically incorrect for an operation to overflow. It may result in more costly code on some architectures. On VAX, for example, extra code is required to suppress overflow detection. Therefore, if it is immaterial whether an operation overflows, or if the front end knows that a particular operation can never overflow, then this flag should be cleared to allow the compiler to generate the most efficient code.
(c) The routine block node has a detect overflow flag. If this flag is clear, then the back end is not required to generate code to detect overflows in integer arithmetic operations. It is free, however, to generate code that will detect overflows if this is more efficient--mandatory suppression of overflow detection can be accomplished only by setting the suppress overflow flag in a particular tuple.
(d) If the detect overflow flag is set in the routine block node, then the generated code must guarantee, for each expression tree, that either the result computed for that expression is valid, or an integer overflow exception is signalled. This is not a requirement that overflow be detected in every possible subexpression of an expression. For example, suppose that A, B, C, and X are 16-bit variables, and that A is 32767 and B and C are 1. In the assignment X:=A+B-C, the generated code might compute A+B-C using 32-bit arithmetic and then check whether the result is a 16-bit result before storing it. This would store the correct answer 32767, even though the same expression, if computed with 16-bit arithmetic, would result in an integer overflow error. The assignment X:=A+B, on the other hand, would compute the value 32768 correctly, but would then generate an overflow exception when it attempted to store it into X. The collection of places where overflows must be detected is not clear, but certainly includes right-hand sides of stores and arguments in routine calls.
(e) Notice also the XCVT conversion operator, which returns the value of its operand, forcing any extraneous high-order bits of the representation to be consistent with the sign of the actual operand. For example, if E is a UINT8 expression which is evaluated using 32-bit arithmetic, then XCVT. UINT8(E :INT16) will be a 16-bit integer whose high-order 8 bits are guaranteed to be zero. In general, if E is an expression of type T, then XCVT. T(E:T) can be used to force the representation of a value to be consistent with its nominal size.
(f) If the representation of an integer operand in some expression contains high-order significant bits beyond the nominal size of the operand, then the generated code is free to use either the full represented value or the value at the nominal size. When this is not acceptable, the front end must generate an XCVT tuple to discard unwanted high-order bits from the representation.
There is not any mechanism in the IL to disable the detection of floating-point overflow exceptions. A floating-point overflow will always result in the signalling of an exception. The signalling of floating-point underflow is controlled only at the routine level. Routine block nodes have a detect underflow flag. If it is set, the compiler is required to generate code which will detect and report any floating-point underflows which occur in that routine; otherwise, the generated code must ignore floating-point underflows.
The conversion operators will compute a value of one arithmetic type that is related to a value of another arithmetic type. The ROUND and TRUNC operators for real-to-integer conversions, the CMPLX operator for real-to-complex conversions, and the REAL and IMAG operators for complex-to-real conversions are all familiar. (ROUND and TRUNC are also defined with a real result type.)
CVT is the general purpose conversion operator. It will do conversions between any two arithmetic types. It is important to be aware, though, that the only conversions that are done directly are UNIT-INT, INT-REAL, and REAL-CMPLX (and of course conversions within a type, such as INT16-INT32). This means, for example, that a CMPLXG-to-UINT16 conversion will actually be done as the series of conversions CMPLXG-to-REALG, REALG-to-INT32, INTB2-to-UINT16. This is not the behavior of VAX Pascal, which has direct real-to-unsigned conversions.
XCVT is a special operator which deals only with integer types. Like CVT, it yields the value of its result type which is arithmetically equal to its operand. However, it has the special feature that it will first change the high-order bits of the representation of the operand so that the operand's representation is arithmetically equal to its value.
For example, consider the expression
If the expression is computed with 32-bit arithmetic, then the result of the ADD might be a register containing %X00000101 (257). The XCVT would then discard the high-order bits, leaving %X00000001 (1), which would already be a valid 16-bit signed integer.
CAST is not really a conversion operator, since it deals with bit patterns, not values. A CAST tuple yields the value of its result type which has the same bit pattern as its operand (truncating or concatenating zero bits if necessary).
Another type is Variable Modification Operators. The operators with names of the form OPMOD, where OP is ADD, IAND, etc., all have an address operand and a value operand. They fetch an arithmetic value from the specified address, perform the indicated operation between it and the value operand, and store the result back at the same address. They also yield the computed value. They are intended to implement C;s op=operators. For example, the code sequence
will have the same effect as
These operators also have OPMODA and OPMODX forms, which fetch, update, and replace a value in a packed array element or a bit field.
The PREINCR, PREINCRA, and PREINCRX operators are essentially the same as ADDMOD, ADDMODA, and ADDMODX, except that instead of a value operand, they have an attribute field containing a compile-time constant increment value. They can be applied to addresses (pointer variables) as well as arithmetic variables. They are intended to implement C's preincrement and predecrement operators.
The POSTINCR, POSTINCRA, and POSTINCRX operators are the same as the PREINCR, and PREINCRX tuples, except that the value of the tuple is the value that the memory location held before it was updated, rather than the value that was stored back into it. They are intended to implement C's postincrement and postdecrement operators.
Strings
The string (or aggregate) types of the compiler are types whose values are sequences of values from a base type. These types are:
STR8, a sequence of eight-bit characters (type CHAR8).
STR16, a sequence of sixteen-bit characters (type CHAR16).
BITS, a sequence of single bits.
DECIMAL, a sequence of decimal digits and an associated precision.
The elements in a character or bit string sequence are numbered from 0 to n-1, where n is the string length. If an eight-bit character string is represented in memory at address A, then the byte at address A contains the first character of the string, the byte at address A+1 contains the second character of the string, and so on through the byte at address A+n-1, which contains the last character of the string. If a sixteen-bit character string is represented in memory at address A, then the word at address A contains the first character of the string, the word at address A+2 contains the second character of the string, and so on through the word at address A+2(n-1), which contains the last character of the string. If a bit string is represented in memory at address A, then the first eight bits of the string are the least significant through the most significant bits of the byte at address A+1, etc.
Aggregate values in general must be represented somewhere in memory, unlike scalar values which can occur in registers, or even as literal operands in machine instructions. However, the semantic model of the intermediate language is that strings can be fetched, manipulated, and stored just like scalars. The compiler is responsible for allocating temporaries to hold intermediate string values.
Note that the code generated for string operations must be consistent with this model, even when there is overlap between the operands. For example, the IL statement STOREF.STR8(A+1,[20], FETCHF.STR8(A,[20]) moves a twenty character string up one position in memory. It must not simply make twenty copies of the character at A.
A string is said to be empty if its length is zero. Let head be a function that returns the first element of a non-empty string, tail be a function that returns the string containing all elements except the first of a non-empty string, and empty be a function that is true if a string is empty and false otherwise. Then the relation between two strings X and Y, as tested by the standard comparison operators (EQL, NEQ, LSS, LEQ, GTR, GEQ), is defined as follows:
If empty(X).LAMBDA.empty(Y) then X=Y.
If empty(X)Vempty(Y) then X<Y.
If empty(X).LAMBDA.empty(Y)then X>Y.
If empty(X).LAMBDA.empty(Y).LAMBDA.head(X)<head(Y)then X<Y.
If empty(X).LAMBDA.empty(Y).LAMBDA.head(X)>head(Y)then X>Y.
If empty(X).LAMBDA.empty(Y).LAMBDA.head(X)=head(Y) then rel(X,Y)=rel(tail(X),tail(Y)).
The string comparison operators in some languages (such as Pascal) operate only on equal-length strings, padding the shorter string in a comparison to the length of the longer string. Therefore, the IL also has padded string comparison operators, EQLP, NEQP, LSSP, LEQP, GTRP, and GEQP.
All of the string operators are listed in Table 12.
Booleans
Unlike the representational data types, the Boolean data type does not have a unique representation. During program execution, Boolean values may be represented explicitly by the value of some bit in a binary integer, or implicitly by the particular code path that is taken. Since there is no unique representation, it is not possible to have Boolean variables in the IL. However, most source languages provide for the logical interpretation of representational values, and many allow the declaration of logical or Boolean variables. Therefore, operators are needed to convert between Boolean values and their source language binary representations.
The LBSET operator interprets an integer as a Boolean by testing its least significant bit, and the NONZERO operator interprets an integer as a Boolean by testing whether the whole integer is zero or not. The LSBIT operator represents a Boolean value as an integer with the bit pattern <00 . . . 00> or <00 . . . 01>, and the ALLBITS operator represents a Boolean value as an integer with the bit pattern <00 . . . 00 > or <11 . . . 11>. These operators support the binary representation of Boolean values in the various source languages as follows:
Even though Boolean values do not have a representation, and therefore cannot be represented with normal literal nodes, it is very desirable to be able to apply all the regular IL transformations to Boolean expressions. Therefore, the back end 12 provides two special literal nodes, whose addresses are contained in the global variables GEM$ST.sub.-- G.sub.-- TRUE and GEM$ST.sub.-- G.sub.-- FALSE. These literal nodes cannot be used for static storage initialization, but they can be used as operands in an ILG.
Boolean expressions involving AND and OR operators can be evaluated in two different ways, full evaluation and flow or short-circuit evaluation. In full evaluation, both operands are fully evaluated, yielding real mode values, which are then used as operands to an AND or OR instruction to yield a real mode result. In flow or short-circuit evaluation, the first operand is evaluated. If the value of the expression is determined by the value of the first operand, then the second operand is skipped; otherwise, the second operand is evaluated and the value of the expression is the value of the second operand.
Some source languages require full evaluation of AND and OR expressions; others require (or have special operators for) short-circuit evaluation; and still others do not specify the kind of evaluation, leaving the choice to the compiler. Three sets of operators are provided for these three cases:
(a) LANDC and LORC ("Logical AND Conditional" and "Logical OR Conditional") are the flow Boolean operators. They evaluate their first operands and then may bypass the evaluation of their second operands.
(b) LANDU and LORU ("Logical AND Unconditional" and "Logical OR Unconditional") are the full evaluation Boolean operators. They behave like normal binary operators, computing a result value from two fully evaluated operand expressions.
(c) LAND and LOR ("Logical AND" and "Logical OR") are CIL operators which do not specify either the kind of evaluation or the order of the operands. During IL expansion, they may be replaced either by LANDC and LORC or by LANDU and LORU tuples. Furthermore, when they are replaced by LANDC and LORC tuples, their operands may be interchanged if the cost of evaluating their first operands appears to be greater than the cost of evaluating their second operands.
The back end 12 must be able to identify the tuples belonging to each operand of a LAND, LOR, LANDC, or LORC tuple. In the CIL, the FLOWMARK tuple is used for this purpose. All of the tuples associated with the first operand of one of these tuples must immediately precede all of the tuples associated with the second operand, which must immediately precede the Boolean operator tuple itself. The first tuple associated with any operand of one of these tuples must be immediately preceded by a FLOWMARK tuple.
For example,
The selection operators will select one of two values of any type, depending on the value of a Boolean operand. Like the logical OR and AND tuples, there are three selection tuples:
(a) SELC will evaluate only its second or its third operand, depending on whether its first operand is true or false.
(b) SELU will always evaluate all three of its operands, and then will select the value of either its second or third operand.
(c) SEL is a CIL operator which does not specify the kind of evaluation. It is replaced by either a SELC or a SELU operator during IL expansion.
Also like the logical AND and OR tuples, SEL and SELC require that the tuples associated with their operands be contiguous, in operand order, and preceded with FLOWMARK tuples.
For example
All of the Boolean operators are listed in Table 13.
Runtime Checking
Checking operators verify that some condition is true during the execution of the program, and cause an exception if the condition is not true. Except for the ASSERT operator, all of the checking operators return the value of their first operand. Every checking tuple has a condition field, which specifies the exception to be signalled if the condition is not true, and a can be continued field, which indicates whether control might be returned after the exception is signalled. If control returns to a checking tuple after an exception, then the checking tuple will return the same value that it would have returned if the exception had not occurred. The checking operators are listed in Table 14.
Flow Control
An ILG 55 is made up of basic blocks. A basic block is a sequence of tuples beginning with a branch target tuple and ending with a branch tuple or a flow termination tuple. A basic block is entered only at its beginning, and in principle all code in it is then executed before control passes out of it at its end (but see the discussion of conditional evaluation above).
In a CILG, the basic blocks are concatenated end to end. The branch tuple at the end of a basic block may be omitted if control flows from it into following basic block, which must begin with a LABEL tuple. Similarly, the LABEL tuple at the beginning of a basic block may be omitted if there are no branches to it. (That is, if the back end sees a LABEL tuple which is not preceded by a branch tuple, then it inserts a BRANCH to it; if it sees a branch tuple which is not followed by a branch target tuple, then it inserts a LABEL tuple with a synthesized label symbol.) The IL expansion phase produces a circular list of tuples for each basic block, with a separate flow graph data structure to represent the relations between them.
Within a basic block, flow implicitly follows the linear tuple ordering. Because all flow between basic blocks is represented with explicitly flow control tuples, the basic blocks of an ILG may be arranged in any order without affecting the meaning of a routine.
The branch target tuple at the beginning of each basic block contains a pointer to a label symbol or entry symbol node in the symbol table. Control flow between basic blocks is represented by a destination list which is an attribute of a branch tuple. Each node in a destination list points to a label symbol or entry symbol node which is also pointed to by some branch target tuple in the same routine, to indicate that control might be transferred to the basic block that begins with that basic block.
A branch target tuple marks the start of a basic block. All branch target tuples have the following attributes:
A branch tuple marks the end of a basic block and specifies its successors. All branch tuples have the following attributes:
A destination list is a list of destination nodes, linked together by their next fields. The destination list field of a branch tuple contains a pointer to the first destination node in such a list. (Note that a destination node can occur in only one destination list, and a destination list can be pointed to by only one branch tuple. Even if two branches have the same destinations, they still must have distinct, albeit identical, destination lists.) Every destination node has a target field, which contains a pointer to a label or entry symbol node. A destination node represents a potential transfer of control to the basic block whose branch target tuple's label symbol field contains a pointer to the same symbol node. There are two kinds of destination nodes. Most kinds of branch tuples use simple destination nodes, and choose a destination based on its position in the destination list. BRSEL tuples, however, use selector destination nodes, and choose the destination whose selector matches the tuple's operand value. A selector destination node has additional fields low test and high test, both longword integers. It matches an operand value if the operand value falls between the destination's low test and high test values.
Unlike the regular branch operators, which specify a set of destinations with a destination list and then select one of them based on a tuple operand, the indirect branch operators (JUMP and JUMPLOCAL) cause control to be transferred to the address specified by an address expression (usually a label variable). These would be the operators used for a FORTRAN assigned goto or a PL/I got through a label variable.
The back end still needs to know the possible destinations of an indirect branch tuple so that it can build the routine flow graph correctly. Therefore, indirect branch tuples have a destination list, just like regular branch operators. However, their destination list contains only a single destination (which is optional for JUMP tuples). The target label of this destination node identifies a VLABEL tuple which is immediately followed by a VBRANCH tuple. The destination list of the VBRANCH tuple then lists all of the actual possible destinations in this routine of the indirect branch.
This combination of a VLABEL tuple and a VBRANCH tuple is referred to as a virtual basic block. No code is ever generated for it (which is why there must not be any other tuples between the VLABEL and the VBRANCH). It represents the fact that control can pass from the indirect branch to any of the successors of the virtual block. This has the advantage that if many indirect branches have the same set of possible destinations, a single virtual basic block can represent the possible destinations of all of them.
There is one other virtual basic block in every routine. This is the block which consists of the BEGIN and ENTRYPTR tuples. No code is generated for it, since execution always begins at an ENTRY tuple, but it identifies all the entry points of the routine for the back end.
A basic block may end with a branch tuple or with a flow termination tuple. When control reaches a flow termination tuple, it leaves the current routine completely. Since flow termination tuples do not transfer control to a destination in the current routine, they do not have destination list and target symbol attributes.
Note that the JUMP operator is effectively a flow termination operator if it does not have a destination list, since that means that it does not have any possible destinations in the current routine. JUMPSYMBOL is a flow termination operator which is used to represent a non-local goto to a known label in the CIL; in the EIL it is replaced by such a non-local JUMP.
All of the flow control operators are listed in Table 15.
Routine Calls and Parameter Passing
There are three types of linkage conventions: control, parameter, return value. The phrase "linkage conventions" refers to all the rules about the generated code which allow a calling routine and a called routine to "talk to each other" properly. Some of these rules are built in to the code generator 29. In other cases there are choices, which must be made consistently for a calling and called routine. Some of these choices will be made by the shell (when it has access to both routines); others must be made by the front end 20, and encoded in the symbol table 30 and ILG 55.
A control linkage convention defines the instructions which must be executed to pass control from a calling to a called routine, to establish the execution context of the called routine, and to return control to the calling routine. Control linkage conventions are determined by the INITCALL and CALL tuples in the calling routine and the entry symbol node for the called routine.
A CALL tuple whose operand is a reference to an entry symbol node which isn't an external reference is an identified call, and there is complete freedom to select the linkage for it, even to the extent of compiling the called routine in line or generating a customized copy of the called routine. For unidentified calls, the calling convention field of the INITCALL tuple must specify the control linkage convention to use for the call. The value of this field must come from the enumerated type GEM$CALLING.sub.-- CONVENTION, those constants are defined in the following list:
A routine block node has a standard entry field which specifies what control linkage convention to use for the copy of this routine that will be called by unidentified calls to this routine. The value of this field must come from the enumerated type GEM$ENTRY.sub.-- CONVENTION, whose constants are defined in the following list:
Parameter Linkage Conventions are another type. A routine call makes an argument list available to the called routine. The argument list is a collection of scalar values (often addresses) in locations which are known to both the calling and the called routine by agreement (registers, or location sin a block of memory whose address is contained in some standard register).
A formal parameter of a called routine is represented by a variable symbol node whose is a parameter flag set. The address associated with a parameter symbol is either a storage location specified by the calling routine or a local storage location which contains a copy of the data passed by the calling routine. (Remember that an "address" may actually specify a register.) It is derived from the argument list and from the mechanism and the semantic flags of the parameter symbol, as described below.
A parameter has bind semantics if the address associated with the parameter variable is the address of the storage location which was passed by the calling routine (the actual storage location). It has copy semantics if the compiler allocates storage for it in the called routine (the local storage location) and generates copies between the actual and local storage locations as needed. (The local storage location of a parameter with bind semantics is the same as its actual storage location.)
The compiler will choose whether to use bind or copy semantics for a parameter based on the usage pattern of the parameter within the routine and on the flags listed in Table 10-3. ("Alias effects" are discussed in CT0.70, Data Access Model. Briefly, they are ways that the actual storage location might be accessed, other than through the parameter symbol. This includes direct reference to a non-local variable which might be the actual storage location, dereference effects, and calls to other routines which might access the actual storage location.)
Table 17 illustrates the use of the parameter semantic flags as they would be set for various source languages.
A parameter mechanism specifies the relationship between what the calling routine wishes to pass to the called routine and what is actually stored in the argument list. A parameter symbol has a mechanism field which specifies the mechanism which is used to pass a value to this parameter, and an argument tuple has a mechanism field which specifies the mechanism by which this argument is to be passed. The values of these fields must come from the enumerated type GEM$MECHANISM, whose constants are listed in Table 18.
If a parameter variable's unknown size flag is false, then the size of the parameter is known at compile time, and is specified by its size field. If unknown size is true, then the size of the parameter is not known at compile time. The size of an unknown size parameter can be determined at run time if it has the array, string, or address and length (reference with associated length parameter) mechanism. When a separate length word is passed with the address and length mechanism, and the parameter has an aggregate data type, the length argument is interpreted as the parameter size in elements (bits or characters), not in bytes. Furthermore, if the parameter is a character string whose string representation is varying or asciz, then the size is a maximum size, not the string's current size, and applies only to the test part of the string, and not to the space that is required for the string length word or null terminator. Note that a parameter cannot have copy semantics unless the compiler know how much to copy. If the actual parameter size is neither known at compile time nor computable by the compiler at run time, then the front end must set the parameter's must bind flag to force the use of bind semantics.
Another type is Return Value Linkage Conventions. A called routine can return information to its caller in two ways. The first is by using an output parameter. This is a variable which is passed with a mechanism other than value, so that the called routine can store a value into it. The second way is with a return value. A return value is a value which is computed by the called routine and "returned" to the caller, where it becomes available as an expression value through a special result tuple.
Scalar values can be returned in registers. For example, almost all of our languages return arithmetic function values in a standard register; and the BLISS "output parameter" feature allows a routine to return values in arbitrary registers.
For a routine to return a string, there must be tuples in the argument list to allocate a temporary buffer for the return value and to pass its address to the called routine, tuples in the called routine to store the return value into the buffer, and tuples in the caller to make the retrieve the value from the buffer.
When the size of a returned string is determined by the called routine, the caller cannot just allocate space for the result, since it does not know in advance how big the result will be. The mechanisms listed in Table 19 provide for this possibility. These mechanisms are provided through special tuples. However, their availability depends on the calling standard of the target environment.
The caller may: (a) require that the called routine return a value by fixed buffer; (b) require that the called routine return a value on the stack; (c) request that the called routine return a value by dynamic string, but accept a string returned on the stack if the called routine so chooses. The called routine must always be prepared to return a dynamic-size result by fixed buffer or on the stack if the caller requires it. It must also be prepared to return a result either by dynamic string or on the stack when the caller requests a result by dynamic string. Representation of routine calls in the CIL will now be considered. Them are many distinct operations involved in calling a procedure or function. Any of the following steps may be necessary.:
(a) Allocate space for the argument list.
(b) Allocate space for pass-by-value operand expression.
(c) Allocate space for descriptors.
(d) Create argument descriptors.
(e) Create argument descriptors.
(f) Allocate space for result values. (A result value, or output argument, is an argument which does not exist until after the call. In the IL, a function will be treated as a procedure with a result value.)
(g) Create the argument list.
(h) Call the routine.
(i) Release space that was allocated for arguments, descriptors, and the argument list.
(j) Get the result values from the call.
(k) Free the space that was allocated for the result values.
The general strategy taken in the IL is to provide separate operators for the different operations involved in doing a call, but to require that these be tied together in a specified fashion. A routine call in the IL consists of:
1. An INITCALL statement, which flags the beginning of the series of actions which will make up the call.
2. A series of argument and temporary allocation statements which will construct the argument list.
3. A call statement (CALL or BPCALL) which actually effects the transfer of control to the called routine.
4. A series of result tuples which make the call's return values accessible.
The INITCALL and call statements are mandatory; the argument list and result tuples are optional. All of the tuples involved in a call must occur in the same basic block, and any result tuples must follow the call tuple immediately, with no intervening tuples. There are no other restrictions, though, on what tuples may occur between the INITCALL and the call. The IL for a routine call may even be contained within the argument list IL for another call.
Constructing the argument list involves allocating space for the argument list itself, for addresses and descriptors of arguments, for temporaries to hold values being passed, and for output arguments. It may also involve initializing the allocated space. These activities are specified in the IL with argument tuples. All argument tuples have names beginning with ARG, and have the attributes listed in Table 20.
When the calling routine has a value to pass, it uses one of the argument tuples whose names begin with ARGVAL. With these tuples, the actual argument value is specified as an operand of the argument tuple. Note that this does not necessarily mean that the argument is passed using the value mechanism. If the mechanism is value, the operand value is stored directly into the argument list; otherwise, a temporary is allocated, the operand value is stored into the temporary, and the temporary is passed by reference or descriptor. (This is like %REF in BLISS.) The value mechanism will only be supported with the ARGVAL tuple with scalar types, and with the ARGVALA tuple with a compile-time constant size.
When the calling routine has the address of an existing store location to pass, it uses on eof the argument tuples whose names begin with ARGADR. With these tuples, the address of the actual storage location is specified as an operand of the argument tuple. Thus, the value mechanism cannot be used with these tuples. Since the occurrence of one of these tuples in an argument list can cause the called routine to read from or write to a storage location known to the current routine, these tuples can have dependencies and side effects, and therefore have the offset, effects, effects2, and base symbol fields that are used in all memory reference tuples, as well as the special flags parm is read and parm is written, which indicate whether the compiler should assume that the called routine might read from and/or write to the storage location.
When an argument tuple specifies the general mechanism, a code is generated to allocate space for the descriptor and to fill in its base address field. The front end must explicitly specify any other fields that are to be initialized in the descriptor. It does this using DSCFIELD tuples, which refer back to a preceding argument tuple with the general mechanism and specify a value to be stored into a field in the descriptor that was allocated for that argument.
Constructing an Argument Block
Some RTL linkages may require that a collection of arguments be passed in an argument block, whose address is passed to the RTL routine like an ordinary reference parameter. This is accomplished using three special tuples.
(a) ARGBLOCK is an argument tuple which allocates a block of a specified size on the stack and passes its address to the called routine. The block can be initialized using BLKFIELD tuples.
(b) A BLKFIELD tuple is like a DSCFIELD tuple, except that it refers back to a preceding ARGBLOCK tuple instead of to an arbitrary tuple with the general descriptor mechanism. It stores a value into a field of the argument block.
(c) ARGDEFINES is like an argument tuple, except that it doesn't generate any code. It allows the front end to specify argument-like side effects which are not associated with a regular argument tuple. In particular, it can be used to indicate the effects associated with arguments which have been passed through an argument block.
For a routine to return an aggregate value, it must store that value into a location that has been allocated by its caller. The tuples whose names begin with ARGTMP will allocate a block of storage of a specified size and pass its address to a called routine. They are the same as the ARGADR tuples, except that the ARGADR tuples pass the address of an existing block of storage, and the ARGTMP tuples pass the address of a temporary that has been allocated especially for the call.
The ARGBUF, ARGSTK, and ARGDYN tuples will allocate the temporaries and pass the special descriptors necessary to obtain a dynamic string return value. These tuples have all the usual argument tuple attributes, but their mechanism attribute is ignored, since the mechanism is implied by the use of the dynamic return value mechanism.
The tuples whose names begin with RESULT will make the return values from a routine call accessible in the calling routine. Their effect is to move the output parameters from the temporary locations or registers where they have been returned by the called routine into more lasting temporaries. The value of a result tuple is simply the value of the return value that it has retrieved. All the result tuples for a call must immediately follow the call tuple.
Bound Procedure Calls
A bound procedure value, or BPV, represents the information needed to call an unknown routine. Since routines may contain uplevel references to stack allocated variables in other routines, a bound procedure value must incorporate not only the code address of the routine to be called, but also sufficient information to construct a static link for it.
Unfortunately, BPVs are handled very differently in different software architectures--how they are created, how they are represented, how they are called, and even how big they are. Therefore, the compiler will not attempt to provide a consistent representation. Instead, the front end will be expected to generate differing code, depending on the target software architecture.
(a) In the VAX and MIPS software architectures, a BPV is simply a code address and a context value, and a bound procedure call is done by loading the context value into a specific register and then doing a call to the code address. Therefore, the front end will be responsible for representing a BPV as a pair of independent address values. The code address is obtained with a BPLINK tuple. A call to a BPV should be represented as a CALL whose address operand is the code address value, with the context value passed by value as a special register argument in the architecture's static link register.
(b) On RISC machines as referred to, all procedures are represented by descriptors which contain a code address along with some additional information, and a BPV is simply the address of a special descriptor, constructed at run time, which contains a context pointer and the address of an RTL routine to load the context pointer and call the real routine. The front end will have to allocate space for such a descriptor itself, and use the BPVAL tuple to fill it in. Then the BPV is represented by the address of the descriptor, and a call to the BPV should be represented by a call to that address.
It is necessary for the back end 12 to know what the parameters are for each entry point in a routine. The front end 20 accomplishes this by setting the param list and param list tail fields of each entry symbol node to point to the first and last nodes in a list of parameter nodes (linked by their next fields) that represents the parameter list of that entry point.
Each parameter node has a symbol field which points to a parameter symbol node of the routine that contains the entry point, and arg location, pass by register, and special register fields which have the same meaning that they do in argument tuples (see Table 20). Thus, the list of parameter nodes identifies all the parameters of an entry point and where they occur in that entry point's argument list.
Note that a parameter symbol may occur in more than one parameter list, possibly with a different arg location in each. Parameter nodes do not have a mechanism field, however, since the mechanism is regarded as an attribute of a parameter symbol rather than of its occurrence in a particular argument list.
The RETURNREG tuple returns a scalar value in a specified register, and the RETURNSTK and RETURNDYN tuples return a string value using one of the dynamic string return mechanisms provided in the PRISM calling standard. Note that no special tuples are needed for a called routine to return a value through an argument temporary, since there is no difference between returning a value through an argument temporary and storing a value into an ordinary output parameter.
The address associated with a parameter symbol is the address of the parameter's local storage location. The called routine can obtain the address of the descriptor for a parameter with the general descriptor mechanism by using the DESCADDR tuple. It can obtain the actual size of an unknown size parameter using the SIZE tuple, provided that the size is available in the argument list (either in a descriptor or in a separate size parameter).
All of the operators involved in routine calls are listed in Table 21.
Storage Allocation and Scoping
A lexical block is a range of a source program over which a set of declarations is valid--for example, a routine, subroutine, function, or begin-end block. In the symbol table, the lexical structure of a routine is represented by a tree of scope block nodes whose root is the routine block node. Each basic block in the ILG contains code belonging to a single lexical block. The branch target tuple at the start of a basic block has a scope block field which points to the corresponding block node in the symbol table. Every lexical block in a routine must have a unique scope entry basic block, which is the only basic block in the lexical block to which control can pass from any basic block outside that lexical block. This scope entry basic block is identified by the block entry flag in the branch target tuple.
A reference to a variable symbol in the CIL always yields the address of a storage location (or the name of a register):
1. A static variable is one whose storage class is static, global ref, or preserved. Static variables are located in some PSECT at compile time, so that every reference to such a variable will refer to the same location.
2. A local variable is one whose storage class is automatic, stacklocal, register, or register preferred, and whose unknown size flag is false. Local variables exist only during a single execution of their lexical scope, and may have multiple instances if multiple instances of their lexical scope may be executing simultaneously. They are allocated at compile time to registers or to known locations in their routine's stack frame.
3. A dynamic variable is one with the same storage class as a local variable, but whose unknown size flag is true. Like local variables, dynamic variables exist only during a single execution of their lexical scope, and may have multiple instances if multiple instances of their lexical scope may be executing simultaneously. They are allocated on the stack at run time by a CREATE tuple, and are accessed through an associated pointer variable which is created by the back end.
4. Parameters with copy semantics behave like local or dynamic variables, depending on the setting of their unknown size flag.
5. Parameters with bind semantics are not allocated in the called routine at all. They are accessed through an associated pointer variable which is created by the back end to hold the actual storage location address.
A tuple in a lexical block may refer to any variable which is declared in that lexical block, or in any of its ancestors in the symbol table block tree. There are no problems referring to variables in the current routine, of course. Static variables of another routine can be referred to directly. Local and dynamic variables of other routines require a "static chain" to locate the stack frame in which the variable is declared. However, the back end 12 is completely responsible for generating the code for creating and using static chains, provided that the front end correctly annotates the routine blocks and variables.
There are several kinds of dynamic stack allocation:
1. The stack storage for a dynamic variable is allocated by a CREATE tuple. It exists from the execution of the CREATE tuple until control passes into a basic block which is not in the same lexical block as the CREATE tuple. (This means that the CREATE tuple for a dynamic variable must be allocated in a basic block whose scope block is the block in which the variable is declared; otherwise, its dynamic storage will be released while the variable is lexically still in scope.)
2. Code to allocate the stack storage for an unknown size copy parameter is generated immediately following the ENTRY tuple. Since ENTRY tuples must be in the main routine block, this storage exists until the routine returns.
3. A dynamic temporary may be created by the back end to hold the value of an aggregate expression. It exists from the execution of the tuple which creates the value at least until the execution of the tuple which uses that vale.
4. Stack space is allocated to hold the argument value for an aggregate ARGVALx tuple. It exists from the execution of the ARGVAL.sub.-- x tuple until the execution of the CALL tuple.
5. Stack space is allocated to hold a return value for an ARGTMPx tuple. It exists from the execution of the ARGTMPx tuple until the evaluation of the RESULTx tuple which fetches the return value.
While this invention has been described with reference to specific embodiments, this description is not meant to be construed in a limiting sense. Various modifications of the disclosed embodiments, as well as other embodiments of the invention, will be apparent to persons skilled in the art upon reference to this description. It is therefore contemplated that the appended claims will cover any such modifications or embodiments as fall within the true scope of the invention. ##SPC1##
Claims
What is claimed is:
1. A method executed in a computer system, the method comprising the steps of: generating a flow graph for a source program that is being compiled using a compiler front end which performs syntactic and semantic processing of a programming language, said flow graph comprising tuples, each tuple representing a single expression in said source program, said flow graph being a representation of said source program in an intermediate language; determining, by said compiler front end for each tuple which can have an effect on another tuple and which is included in said flow graph, an effect indicator that represents an effect said each tuple has on said another tuple, said effect indicator being associated with said each tuple; determining, by said compiler front end for each tuple which can have a dependency on another tuple and which is included in said flow graph, a dependency indicator that represents a dependency said each tuple has on said another tuple, said dependency indicator being associated with said each tuple; detecting a programming loop construct in said flow graph, said programming loop construct comprising one or more blocks in said flow graph, each of said blocks corresponding to a portion of said source program and including a sequence of tuples with no entry or exit between a first tuple add a last tuple of said sequence; determining for each of said blocks comprising said programming loop construct, a set of effects by examining effect indicators that are associated with tuples in the sequence included in a corresponding block, each of said set of effects being associated with the corresponding block representing memory locations that may be modified by one of said tuples in said each corresponding block; determining for said programming loop construct a set of induction variable candidates by examining said sets of effect and dependency indicators, said dependency indicators being associated with tuples that are included in said programing loop construct, said set of induction variable candidates including variables whose dependencies are included in said set of candidates and representing memory locations that may be modified by one of said tuples in said programming loop construct; and optimizing said flow graph using a compiler back end and said set of induction variable candidates.
2. The method of claim 1, wherein said optimizing further comprises updating said set of induction variable candidates by examining tuples in said sequence and eliminating a variable from said set of induction variable candidates when said variable does not conform to predetermined rules that define an induction variable; and using said set of induction variable candidates to perform one or more code optimizations.
3. The method of claim 2, wherein said predetermined rules include basic induction variables, inductive expressions and pseudo-inductive expressions, a basic induction variable being a variable represented as a linear function of said variable, an inductive expression being a reference to another inductive variable or is a linear function of another inductive expression, and a pseudo-induction variable having properties of either an induction variable or an inductive expression for iterations through said programming loop construct other than a first iteration.
4. The method of claim 2 further comprising using effect indicators and dependency indicators to validate a first code optimization and to perform a second code optimization that affects tuples included in said programming loop construct.
5. The method of claim 2, wherein said code optimization performed using said set of induction variable candidates is one of common subexpression elimination, strength reduction, variable elimination, vectorization, and loop unrolling.
6. The method of claim 2 wherein, after executing said updating step by examining each tuple in said sequence, said set of induction variable candidates includes only those variables that conform to said predetermined rules.
7. The method of claim 1, wherein said compiler back end is a common compiler back end used by a plurality of compiler front ends to perform optimizations and generate object code, said compiler front end is a first compiler front end, said source program is a first source program, and the method further comprises compiling a second source program using a second compiler front end and said compiler back end, said second compiler front end performing syntactic and semantic processing of a second programming language different from said first programming language, said second source program including source statements in said second programming language.
8. A memory that is used in a computer system for compiling source programs, the memory comprising: means for generating a flow graph for a source program that is being compiled using a compiler front end which performs syntactic and semantic processing of a programming language, said flow graph comprising tuples, each tuple representing a single expression in said source program, said flow graph being a representation of said source program in an intermediate language; first determining means for determining, by said compiler front end for each tuple which can have an effect on another tuple and which is included in said flow graph, an effect indicator that represents an effect said each tuple has on said another tuple, said effect indicator being associated with said each tuple; second determining means for determining, by said compiler front end for each tuple which can have a dependency on another tuple and which is included in said flow graph, a dependency indicator that represents a dependency said each tuple has on said another tuple, said dependency indicator being associated with said each tuple; means for detecting a programming loop construct in said flow graph, said programming loop construct comprising one or more blocks in said flow graph, each of said blocks corresponding to a portion of said source program and including a sequence of tuples with no entry or exit between a first tuple and a last tuple of said sequence; third determining means, coupled to said detecting means and said first determining means, for determining a set of effects for each of said blocks comprising said programming loop construct by examining effect indicators that are associated with tuples in the sequence included in a corresponding block, each of said set of effects being associated with the corresponding block representing memory locations that may be modified by one of said tuples in said each corresponding block; fourth determining means, coupled to said third determining means and said second determining means, for determining a set of induction variable candidates for said programming loop construct by examining said sets of effects and dependency indicators, said dependency indicators being associated with tuples that are included in said programming loop construct, said set of induction variable candidates including variables whose dependencies are included in said set of candidates and representing memory locations that may be modified by one of said tuples in said programming loop construct; and means for optimizing said flow graph, coupled to said generating means and said fourth determining means, using a compiler back end and said set of induction variable candidates.
9. The memory of claim 8 wherein said optimizing means further comprises: means for updating said set of induction variable candidates by examining tuples in said sequence and eliminating a variable from said set of induction variable candidates when said variable does not conform to predetermined rules that define an induction variable; and means for using said first set to perform code optimizations.
Patent Citations (12)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US4642765 | 1987-02-01 | Cocke et al. | |
| US4667290 | 1987-05-01 | Goss et al. | |
| US4802091 | 1989-01-01 | Cocke et al. | |
| US4807126 | 1989-02-01 | Gotou et al. | |
| US5067068 | 1991-11-01 | Iwasawa et al. | |
| US5109331 | 1992-04-01 | Ishida et al. | |
| US5170465 | 1992-12-01 | McKeeman et al. | |
| US5175856 | 1992-12-01 | Van Dyke et al. | |
| US5182806 | 1993-01-01 | McKeeman et al. | |
| US5193191 | 1993-03-01 | McKeeman et al. | |
| US5201050 | 1993-04-01 | McKeeman et al. | |
| US5493675 | 1996-02-01 | Faiman, Jr. et al. |
Non-Patent Literature (2)
- Aho, A., et al., Compilers, Principles, Techniques, and Tools, Addison-Wesley, pp. 528, 534-535, 546-554, 592-648 (1986).
- Aho et al, "Compilers, Principles, Techniques and Tools", Addison Wesley, Publisher, 1986, pp. 1-24, 463-472, 513, 528-534, 585-591.