Information Technology
Method Predicts Porting Speedup
WARF: P140165US01
Inventors: Karthikeyan Sankaralingam, Newsha Ardalani, Xiaojin (Jerry) Zhu
The Wisconsin Alumni Research Foundation (WARF) is seeking commercial partners interested in developing a method of estimating the amount of speedup that will be obtained from porting a program between two different processor units, for example, between a CPU and a GPU.
Overview
High-performance computers may employ two processor systems having very different architectures, such as a CPU (computer processing unit) and a GPU (graphic processing unit). The ability of a GPU to handle not only graphics tasks but also general tasks in parallel has led to “heterogeneous processing” in which the GPU takes over tasks normally performed by the CPU. Some programs speed up when moved (“ported”) from the CPU to a GPU.
However, porting a program requires substantial restructuring of the software and data organization. Code optimization of such ported programs can be very time consuming and require specialized tools and expertise. Given these costs, it would be advantageous to know if the amount of speedup will justify the effort. Currently, this cannot be known until after the fact.
However, porting a program requires substantial restructuring of the software and data organization. Code optimization of such ported programs can be very time consuming and require specialized tools and expertise. Given these costs, it would be advantageous to know if the amount of speedup will justify the effort. Currently, this cannot be known until after the fact.
The Invention
UW–Madison researchers have developed a method of assessing the benefits of porting a program before the effort and cost of porting are undertaken. In other words, the amount of speedup can be predicted in advance.
In the new method, a computer measures multiple quantifiable execution properties of a given program. These properties describe how the program executes on a first processor system (e.g., a CPU). Next, the properties are applied to a model that estimates the change in speed that will occur if executed on a second processor system (e.g., a GPU).
In the new method, a computer measures multiple quantifiable execution properties of a given program. These properties describe how the program executes on a first processor system (e.g., a CPU). Next, the properties are applied to a model that estimates the change in speed that will occur if executed on a second processor system (e.g., a GPU).
Applications
- Performance-predicting software
Key Benefits
- Assesses porting costs ahead of time
- Saves time and resources
- Does not require detailed instrumentation or specialized knowledge
- Works with a wide variety of different processors and programs
Stage of Development
Promising simulations have been performed using industry accepted benchmarks and studies on three different GPU platforms show the method works and produces highly accurate predictions.
Additional Information
For More Information About the Inventors
Related Technologies
For current licensing status, please contact Jeanine Burmania at [javascript protected email address] or 608-960-9846