The Glimpses Project is being undertaken by Jaswanth Sreeram , Ashwini Bhagwat and Santosh Pande at Georgia Institute of Technology .
We propose an approach for analyzing legacy sequential codebases for the purpose of porting them to the SPEs of the CELL processor. This is a challenging task because of the novel architectural attributes of the processor and the amount of parallelism available. GLIMPSES is a compiler-driven profiling and visualization framework that enables CELL programmers to quickly evaluate a program's dynamic behavior, Memory access patterns, Aliasing etc. and determine its suitability for execution on the SPEs. With this tool, programmers can view call graphs, function characteristics such as stack and heap usage and dynamic memory reference patterns in an interactive visualizer. Glimpses serves as a program comprehension and analysis tool for any programmer.The tool is licensed under an open-source BSD license.
The CELL processor is a high performance, heterogeneous, parallel architecture that is well suited for a diverse range of workloads ranging from scientific applications to digital home entertainment applications. A single CELL processor consists of a PowerPC core called the Power Processing Element (PPE) and 8 Synergistic Processing Elements (SPEs), which are geared towards executing SIMD style code. The PPE and the SPEs are connected to main memory via a high-throughput Element Interconnect Bus (EIB). While the CELL processor is capable of achieving very high throughput, the programmer must be able to thoroughly understand the architectural constraints and program behavior to be able to exploit its true potential. Factors that make porting legacy code to the CELL processor and specifically the SPEs challenging and the program properties that are deemed important for this problem are discussed in Exploiting Parallelism . In Understanding Program Properties we describe the rationale behind extracting each program property as well as the methodology used to extract these important program properties. Visualization Framework describes our visualization framework that makes understanding these properties intuitive and useful