Search for: All records

High performance code generation for computationally intensive kernels is a persistent challenge for developers. Given a target architecture and a specific operation, the developer must tune that operation to the lowest-level details of the architecture. This problem is exacerbated by the fact that different architectural targets necessitate different implementations, and even the slightest adjustment to the operation may require large changes in the implementation in order to achieve performance. For performance critical applications this generation is typically performed by hand. However, this level of programming is difficult in terms of the domain knowledge required, and yields coded implementations that increase that challenge of reasoning about the correctness of the problem. Automatic code generation would address these issues. At the very least, by automating the application of the various code transformations needed for performance, this should reduce the issue of correctness, as long as these transformations only lead to correct implementations in the search space. In this paper, we look at a subset of correct implementations of an operation, all valid static schedules of instructions of one particular mix of instructions. We then explore the use of Reinforcement Learning in order to search for the optimal implementation in this subset for the target operation. This work is the first step in automating the exploration of correct implementations using Reinforcement Learning for automatic code generation. 

Abstract Tide gauge water levels are commonly used as a proxy for flood incidence on land. These proxies are useful for projecting how sea‐level rise (SLR) will increase the frequency of coastal flooding. However, tide gauges do not account for land‐based sources of coastal flooding and therefore flood thresholds and the proxies derived from them likely underestimate the current and future frequency of coastal flooding. Here we present a new sensor framework for measuring the incidence of coastal floods that captures both subterranean and land‐based contributions to flooding. The low‐cost, open‐source sensor framework consists of a storm drain water level sensor, roadway camera, and wireless gateway that transmit data in real‐time. During 5 months of deployment in the Town of Beaufort, North Carolina, 24 flood events were recorded. Twenty‐five percent of those events were driven by land‐based sources—rainfall, combined with moderate high tides and reduced capacity in storm drains. Consequently, we find that flood frequency is higher than that suggested by proxies that rely exclusively on tide gauge water levels for determining flood incidence. This finding likely extends to other locations where stormwater networks are at a reduced drainage capacity due to SLR. Our results highlight the benefits of instrumenting stormwater networks directly to capture multiple drivers of coastal flooding. More accurate estimates of the frequency and drivers of floods in low‐lying coastal communities can enable the development of more effective long‐term adaptation strategies.