Google Inc.

Googe Inc.

Christopher G Demetriou

Matthew Nicholas Papakipos

Brian K Grant

Morgan S McGuire

Cary Allen Coutant

Xinliang David Li

Nathan Daniel Tuck

Sriraman Tallam

Ian Lance Taylor

Noah L Gibbs

Jan Civlin

A method for debugging an application includes obtaining first and second fusible operation requests; if there is a break point between the first and the second operation request, generating a first set of compute kernels including programs corresponding to the first operation request, but not to the second operation request; and generating a second set of compute kernels including programs corresponding the second operation request, but not to the first operation request; if there is no break point between the first and the second operation request, generating a third set of compute kernels which include programs corresponding to a merge of the first and second operation requests; and arranging for execution of either the first and second, or the third set of compute kernels, further including debugging the first or second set of compute kernels when there is a break point set between the first and second operation requests.

Systems and methods for generating reference results using parallel-processing computer system

A runtime system implemented in accordance with the present invention provides an application platform for parallel-processing computer systems. Such a runtime system enables users to leverage the computational power of parallel-processing computer systems to accelerate/optimize numeric and array-intensive computations in their application programs. This enables greatly increased performance of high-performance computing (HPC) applications.

Application program interface of a parallel-processing computer system that supports multiple programming languages

A system and method for optimizing the object codes of a computer program is disclosed. The method includes receiving one or more object code units associated with an executable program; identifying, among the object code units, a first program entity and a first set of operations associated with the first program entity and a second program entity and a second set of operations associated with the second program entity, each program entity having an object code segment and an associated address; updating the object code units by inserting a predefined instruction before the first program entity's object code segment and causing the second set of operations to be associated with the predefined instruction if the first program entity's object code segment is identical to the second program entity's object code segment; and combining the updated object code units into the executable program.

Method and system for optimizing an executable program by generating special operations for identical program entities

A system and method for optimizing the object code of a computer program is disclosed. The method includes receiving one or more object code units from the memory; identifying first and second identical program entities in the one or more object code units, wherein each program entity has an object code segment and an associated address; updating the one or more object code units by merging the first identical program entity with the second identical program entity if neither of the first and second program entities is subject to an operation on the respective program entity's address; and linking the updated one or more object code units into an executable program.

Method and system for optimizing an executable program by selectively merging identical program entities

Systems and methods for compiling an application for a parallel-processing computer system

A method for optimizing the object code of a program is disclosed. A compiler generates, respectively, first and second object code segments for first and second source code segments of the program. If the two object code segments are determined to be identical, the compiler generates first and second debugging information entries in a compilation unit of the program and both entries include information for locating the first object code segment. The compiler inserts two entries into a call table in the compilation unit, each entry including information for locating a respective call site that invokes a respective source code segment within a call stack of the program and information for locating a respective debug information entry. The call table is used for associating an operation within the first object code segment with one of the first and second source code segments at runtime.

Method and system for optimizing the object code of a program

Systems and methods for caching compute kernels for an application running on a parallel-processing computer system

Systems and methods for determining compute kernels for an application in a parallel-processing computer system

A method for debugging the object code of a program is disclosed. While executing an object code segment on an information processing device, a debugger receives a location identifier associated with the object code segment and identifies a call site within a call stack of the program using the location identifier. The call site is responsible for invoking the execution of the object code segment. Next, the debugger scans a call table for an entry of the call site that includes information for locating a debugging information entry within a compilation unit of the program. The debugger retrieves the debugging information entry from the compilation unit using the location information and identifies a unique one of multiple different source code segments as the one corresponding to the object code segment using a source code segment identifier in the debugging information entry.

Method and system for debugging merged functions within a program

Systems and methods for debugging an application running on a parallel-processing computer system

Multi-thread runtime system

Runtime system for executing an application in a parallel-processing computer system

A runtime system implemented in accordance with the present invention provides an application platform for parallel-processing computer systems. Such a runtime system enables users to leverage the computational power of the parallel-processing computer systems to accelerate/optimize numeric and array-intensive computations in their application programs. A profiling tool is used to collect, analyze, and visualize the performance data of an application in connection with its execution on a parallel-processing computer system through the runtime system. This profiling tool greatly enhances an application developer's ability to understand how an application is executed on the parallel-processing computer system and fine-tune the application to achieve high performance.

Growing community of inventors

Redwood City, CA, United States of America

Christopher G Demetriou