Future value based single assignment program representations and optimizations

An optimizing compiler internal representation fundamentally affects the clarity, efficiency and feasibility of optimization algorithms employed by the compiler. Static Single Assignment (SSA) as a state-of-the-art program representation has great advantages though still can be improved. This dissertation explores the domain of single assignment beyond SSA, and presents two novel program representations: Future Gated Single Assignment (FGSA) and Recursive Future Predicated Form (RFPF). Both FGSA and RFPF embed control flow and data flow information, enabling efficient traversal program information and thus leading to better and simpler optimizations. We introduce future value concept, the designing base of both FGSA and RFPF, which permits a consumer instruction to be encountered before the producer of its source operand(s) in a control flow setting. We show that FGSA is efficiently computable by using a series T1/T2/TR transformation, yielding an expected linear time algorithm for combining together the construction of the pruned single assignment form and live analysis for both reducible and irreducible graphs. As a result, the approach results in an average reduction of 7.7%, with a maximum of 67% in the number of gating functions compared to the pruned SSA form on the SPEC2000 benchmark suite. We present a solid and near optimal framework to perform inverse transformation from single assignment programs. We demonstrate the importance of unrestricted code motion and present RFPF. We develop algorithms which enable instruction movement in acyclic, as well as cyclic regions, and show the ease to perform optimizations such as Partial Redundancy Elimination on RFPF.

Evaluating the usefulness of soil moisture (forecasts) in streamflow prediction for central Texas

The past decade has brought significant advancements in seasonal climate forecasting. However, water resources decision support and management continues to be based almost entirely on historical observations and does not take advantage of climate forecasts. This study builds on previous work that conditioned streamflow ensemble forecasts on observable climate indicators, such as the El Niño-Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) for use in a decision support model for the Highland Lakes multi-reservoir system in central Texas operated by the Lower Colorado River Authority (LCRA). In the current study, seasonal soil moisture is explored as a climate indicator and predictor of annual streamflow for the LCRA region. The main purpose of this study is to evaluate the correlation of fractional soil moisture with streamflow using the 1950-2000 Variable Infiltration Capacity (VIC) Retrospective Land Surface Data Set over the LCRA region. Correlations were determined by examining different annual and seasonal combinations of VIC modeled fractional soil moisture and observed streamflow. The applicability of the VIC Retrospective Land Surface Data Set as a data source for this study is tested along with establishing and analyzing patterns of climatology for the watershed study area using the selected data source (VIC model) and historical data. Correlation results showed potential for the use of soil moisture as a predictor of streamflow over the LCRA region. This was evident by the good correlations found between seasonal soil moisture and seasonal streamflow during coincident seasons as well as between seasonal and annual soil moisture with annual streamflow during coincident years. With the findings of good correlation between seasonal soil moisture from the VIC Retrospective Land Surface Data Set with observed annual streamflow presented in this study, future research would evaluate the application of NOAA Climate Prediction Center (CPC) forecasts of soil moisture in predicting annual streamflow for use in the decision support model for the LCRA.

Using a Prediction and Option Generation Paradigm to Understand Decision Making

In many complex and dynamic domains, the ability to generate and then select the appropriate course of action is based on the decision maker's "reading" of the situation--in other words, their ability to assess the situation and predict how it will evolve over the next few seconds. Current theories regarding option generation during the situation assessment and response phases of decision making offer contrasting views on the cognitive mechanisms that support superior performance. The Recognition-Primed Decision-making model (RPD; Klein, 1989) and Take-The-First heuristic (TTF; Johnson & Raab, 2003) suggest that superior decisions are made by generating few options, and then selecting the first option as the final one. Long-Term Working Memory theory (LTWM; Ericsson & Kintsch, 1995), on the other hand, posits that skilled decision makers construct rich, detailed situation models, and that as a result, skilled performers should have the ability to generate more of the available task-relevant options. The main goal of this dissertation was to use these theories about option generation as a way to further the understanding of how police officers anticipate a perpetrator's actions, and make decisions about how to respond, during dynamic law enforcement situations. An additional goal was to gather information that can be used, in the future, to design training based on the anticipation skills, decision strategies, and processes of experienced officers. Two studies were conducted to achieve these goals. Study 1 identified video-based law enforcement scenarios that could be used to discriminate between experienced and less-experienced police officers, in terms of their ability to anticipate the outcome. The discriminating scenarios were used as the stimuli in Study 2; 23 experienced and 26 less-experienced police officers observed temporally-occluded versions of the scenarios, and then completed assessment and response option-generation tasks. The results provided mixed support for the nature of option generation in these situations. Consistent with RPD and TTF, participants typically selected the first-generated option as their final one, and did so during both the assessment and response phases of decision making. Consistent with LTWM theory, participants--regardless of experience level--generated more task-relevant assessment options than task-irrelevant options. However, an expected interaction between experience level and option-relevance was not observed. Collectively, the two studies provide a deeper understanding of how police officers make decisions in dynamic situations. The methods developed and employed in the studies can be used to investigate anticipation and decision making in other critical domains (e.g., nursing, military). The results are discussed in relation to how they can inform future studies of option-generation performance, and how they could be applied to develop training for law enforcement officers.