This paper from Lawrence Berkeley National Laboratory introduces the Roofline model as a way to judge the suitability of your kernel to any given architecture.
| Resource information | Details |
|---|---|
| Paper Title | Roofline: An Insightful Visual Performance Model for Floating-Point Programs and Multicore Architectures |
| Author | Samuel Williams, Andrew Waterman and David Patterson |
| Focus | Interoperable libraries |
The operational intensity introduced in the Roofline Model -- operations per byte of DRAM traffic -- is a simple model that can be used to determine what architectures are the best match for a given computational kernel or, conversely, in what ways to optimize a kernel so it performs better on a given architecture. Operational intensity is not typically provided directly by performance tools but can be estimated from other readily available measurements.
