Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


dissertation.pdf (1.14 MB)

ETD Abstract Container

Abstract Header

Compiler Techniques for Transformation Verification, Energy Efficiency and Cache Modeling

Bao, Wenlei
http://rave.ohiolink.edu/etdc/view?acc_num=osu1524073563586939

Abstract Details

2018, Doctor of Philosophy, Ohio State University, Computer Science and Engineering.
Performance has been the focus of computer systems for decades, from past Moore law to current parallel computers. Compiler optimizations are used to improve performance by generating code to utilize hardware (e.g. cache)component efficiently. However, modern systems such as large scale system require not only performance but also resilience and energy efficiency. Increasing concern of system resilience and energy efficiency has been shown in both industry and academia. Errors within applications, especially those escape from detection and resulting in silent data corruption, are extremely problematic. Thus, in order to improve the resilience of applications, error detection and vulnerability characterization techniques are an important step towards fault tolerant applications. Compiler transformations, which restructure programs to improve performance by leveraging data locality and parallelism, are often complex and possibly involve bugs that leads to errors in transformed programs. Thus it is essential to guarantee the correctness, however, current approaches suffers from various problems such as transformations supported or space complexity etc. This dissertation presents a novel approach that performs dynamic verification by inserting lightweight checker codes to detect errors of transformations. The errors are exposed by the execution of checker-inserted transformed program if exist. Energy efficiency is of increasingly importance in scenarios ranging from battery-operated devices to data centers striving for lower energy costs. Dynamic voltage and frequency scaling (DVFS) adapts CPU power consumption by modifying processor frequency to improve energy efficiency. Typical DVFS approaches involve default strategies such as reacting to the CPU runtime load to adapt frequency, which have inherent limitations because of processor-specific and application-specific effects. This dissertation developed a novel compile-time characterization to select frequency and number of CPU cores to use, which providing significant additional benefits over the runtime approach. Cache memory, as one of the most fundamental components of modern processors, has a significant impact on the performance of current computer systems. Compiler optimizations on efficient use of cache to reduce data movement, are often based on very approximate cost models due to the lack of precise modeling of hierarchical cache. The challenge of accurately modeling cache misses has made trace-based simulation the current method of choice. This dissertation takes a fundamentally different approach for polyhedral programs, developed a closed-form solution for modeling of misses of set-associative cache by leveraging the power of polyhedral analysis. This solution can enable program transformation choice at compile time to optimize cache misses. In sum, the dissertation makes contributions to advance compiler technology to achieve program transformation verification, to reduce energy costs, and to effectively modeling cache behaviors.
Ponnuswamy Sadayappan (Advisor)
Gagan Agrawal (Committee Member)
Radu Teodorescu (Committee Member)
Louis-Noel Pouchet (Committee Member)
Sriram Krishnamoorthy (Committee Member)
187 p.
Computer Science
compiler optimization; polyhedral compilation; program verification; energy optimization; DVFS; cache modeling; vulnerability analysis

Recommended Citations

Citations

  • Bao, W. (2018). Compiler Techniques for Transformation Verification, Energy Efficiency and Cache Modeling [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1524073563586939

    APA Style (7th edition)

  • Bao, Wenlei. Compiler Techniques for Transformation Verification, Energy Efficiency and Cache Modeling. 2018. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1524073563586939.

    MLA Style (8th edition)

  • Bao, Wenlei. "Compiler Techniques for Transformation Verification, Energy Efficiency and Cache Modeling." Doctoral dissertation, Ohio State University, 2018. http://rave.ohiolink.edu/etdc/view?acc_num=osu1524073563586939

    Chicago Manual of Style (17th edition)

osu1524073563586939
519
© 2018, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.