Analyzing inductive expressions in a multilanguage optimizing compiler

Claim 3

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.

Claim 4

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.

Claim 5

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.

Claim 6

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.

Claim 7

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.

Claim 8

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.

Claim 9

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.

Claim 10

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.

Claim 11

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.