Systems and methods for compiling an application for 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 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.
1. A computer-implemented method, configured to be performed at a parallel-processing computer system with first and second processing elements, each having a different instruction set architecture, comprising:
at run-time, while the application is executing,
receiving one or more operation requests from an application;
preparing an intermediate representation for the operation requests;
dynamically preparing a set of compute kernels for the intermediate representation, by:
compiling one or more compute kernels for a previously uncompiled first portion of the intermediate representation for execution on the first processing element;
searching a set of previously compiled compute kernels to determine whether a second portion of the intermediate representation has been previously compiled; and
selecting one or more previously compiled compute kernels for execution on the second processing element, the selected compute kernels having been compiled for execution on the second processing element; and
dynamically executing the set of compute kernels on the first and second processing elements,
wherein the first and second processing elements are selected from a group consisting of: single-core central processing units, multi-core central processing units, graphics processing units, single-core co-processors and multi-core co-processors.
2. The method of claim 1 , wherein the operation requests correspond to an application program interface function call in the application.
3. The method of claim 1 , wherein the operation requests correspond to a code segment written in a specific programming language in the application.
4. The method of claim 1 , further comprising writing results of the execution into shared system memory accessible to the first and second processing elements and the application.
5. A computer-implemented method, configured to be performed at a parallel-processing computer system with first and second processing elements, each having a different instruction set architecture, comprising:
at compile-time,
identifying one or more operation requests in an application source code;
preparing an intermediate representation for the operation requests;
at run-time, while the application is executing,
dynamically preparing a set of compute kernels for the intermediate representation, by:
compiling one or more compute kernels for a previously uncompiled first portion of the intermediate representation for execution on the first processing element;
searching a set of previously compiled compute kernels to determine whether a second portion of the intermediate representation has been previously compiled; and
selecting one or more previously compiled compute kernels for execution on the second processing element, the selected compute kernels having been compiled for execution on the second processing element; and
dynamically executing the set of compute kernels on the first and second processing elements,
wherein the first and second processing elements are selected from a group consisting of: single-core central processing units, multi-core central processing units, graphics processing units, single-core co-processors and multi-core co-processors.
6. The method of claim 5 , wherein the operation requests in the application source code correspond to an application program interface function call.
7. The method of claim 5 , wherein the operation requests in the application source code correspond to a code segment written in a specific programming language.
8. The method of claim 5 , further comprising, at compile-time, applying loop analysis to the intermediate representation.
9. The method of claim 5 , further comprising writing results of the execution into shared system memory accessible to the first and second processing elements and the application.
10. A parallel-processing computer system, comprising:
memory;
a first and a second processing element, each having a different instruction set architecture; and
at least one program stored in the memory and executed by the multiple processing elements, the at least one program comprising instructions for:
at run-time, while the application is executing,
receiving one or more operation requests from an application;
preparing an intermediate representation for the operation requests;
dynamically preparing a set of compute kernels for the intermediate representation, by:
compiling one or more compute kernels for a previously uncompiled first portion of the intermediate representation for execution on the first processing element;
searching a set of previously compiled compute kernels to determine whether a second portion of the intermediate representation has been previously compiled; and
selecting one or more previously compiled compute kernels for execution on the second processing element, the selected compute kernels having been compiled for execution on the second processing element; and dynamically executing the set of compute kernels on the first and second processing elements,
wherein the first and second processing elements are selected from a group consisting of: single-core central processing units, multi-core central processing units, graphics processing units, single-core co-processors and multi-core co-processors.
11. The computer system of claim 10 , wherein the operation requests correspond to an application program interface function call in the application.
12. The computer system of claim 10 , wherein the operation requests correspond to a code segment written in a specific programming language in the application.
13. The system of claim 10 , further comprising writing results of the execution into shared system memory accessible to the first and second processing elements and the application.
14. A non-transitory computer readable storage medium storing one or more programs configured to be executed by a parallel-processing computer system with a first and a second processing element, each having a different instruction set architecture, the one or more programs comprising instructions for:
at run-time, while the application is executing,
receiving one or more operation requests from an application;
preparing an intermediate representation for the operation requests;
dynamically preparing a set of compute kernels for the intermediate representation, by:
compiling one or more compute kernels for a previously uncompiled first portion of the intermediate representation for execution on the first processing element;
searching a set of previously compiled compute kernels to determine whether a second portion of the intermediate representation has been previously compiled; and
selecting one or more previously compiled compute kernels for execution on the second processing element, the selected compute kernels having been compiled for execution on the second processing element; and
dynamically executing the set of compute kernels on the first and second processing elements of the parallel-processing computer system,
wherein the first and second processing elements are selected from a group consisting of: single-core central processing units, multi-core central processing units, graphics processing units, single-core co-processors and multi-core co-processors.
15. The non-transitory computer readable storage medium of claim 14 , further comprising writing results of the execution into shared system memory accessible to the first and second processing elements and the application.