Portable, Predictable and Partitionable: A Domain Specific approach to Heterogeneous Computing

Add to your list(s) Download to your calendar using vCal

• Gordon Inggs (Imperial College)
• Monday 27 July 2015, 11:30-12:30
• Level 9 Mahanakorn Lab., EE Dept..

If you have a question about this talk, please contact Grigorios Mingas.

Computing is increasingly heterogeneous. Beyond Central Processing Unit (CPU) architectures with varying degrees of parallel compute capability, different architectures such as massively parallel Graphics Processing Units (GPUs) and reconfigurable Field Programmable Gate Arrays (FPGAs) are seeing widespread adoption. However the failure of conventional programming approaches to support portable execution across these platforms, predict the runtime characteristics of execution upon heterogeneous platforms and partition workloads optimally is hindering the realisation of heterogeneous computing’s potential.

By narrowing the scope of expression, these three challenges could be more readily addressed. Application domains provide a natural limit on the scope that does not unduly impair programmers. A domain specific heterogeneous computing methodology enables three features: Portability, Prediction and Partitioning. Portably efficient execution is enabled by a domain specific approach because only a limited subset of domain functions need to be supported across the heterogeneous computing platforms. Predictive models of runtime characteristics or domain metrics are similarly enabled as the structure of the domain functions may be analysed a priori. Finally optimal partitioning is possible because the domain metric models can be used to express partitioning of tasks as an optimisation program that can be solved by either heuristic, global optimisation or Mixed Integer Linear Programming (MILP) approaches.

Using the example of the computational finance domain of derivatives pricing, a domain specific application framework, the Forward Financial Framework, can execute a single pricing task description upon a diverse range of CPU , GPU and FPGA platforms from many different vendors. Not only do these portable implementations exhibit strong parallel scaling, but are competitive with state-of-the-art, expert created implementations of the same problem. Furthermore, the Forward Financial Framework can model the crucial runtime metrics of latency and accuracy for these heterogeneous platforms using a small benchmarking procedure to within 10% of the runtime value of these metrics. Finally, the framework can optimal partition work across heterogeneous platforms, using a MILP framework, that is up to 250 times more efficient than what is achieved by merely allocating tasks according to theoretical platform capability.

This talk is part of the CAS Talks series.

This talk is included in these lists:

• Andrea Picciau's list
• CAS Talks
• Circuits and Systems Group: Internal Seminars
• Level 9 Mahanakorn Lab., EE Dept.

Note that ex-directory lists are not shown.