J. Fritts. Architecture and Compiler Design Issues in Programmable Media Processors, Ph.D. Thesis, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....They also find that data caches are more effective for reads on MediaBench than SPECint95, while they are less effective for writes. SPECint95 required almost three times more bus bandwidth than MediaBench. MediaBench applications were found to have higher IPC than SPECint95 workloads. Fritts [34] extends the characterization of MediaBench workloads. He finds that nearly 70 of the instructions operate on data sizes of 8 and 16 bits. The average basic block size is found to be small, which leads him to conclude that parallelism in multimedia applications is not within basic blocks. ....

....ratio Figure. 4.3. Branch Statistics (a) for individual multimedia benchmarks and (b) comparison of media applications with other workloads [9] 49 In spite of such a processing, the average basic block size of multimedia applications is over twice that of programs in the SPEC suite. Fritts [34] reports that the average basic block size of multimedia applications in the MediaBench suite is similar to general purpose integer applications. He observes that the average basic block size varies significantly from one media benchmark to the other. Approximately 7 of all branches are ....

[Article contains additional citation context not shown here]

J. Fritts, "Architecture and compiler design issues in programmable media processors," Ph.D. Thesis, Dept. of Electrical Engineering, Princeton University, 2000.

....SIMD nature prohibits parallelism across function boundaries. Consequently, while the SBL method supports both fully parallel and partially parallel loops, it cannot support arbitrary sized tasks like the multiscalar and multithreaded architectures. More on the SBL method can be found in Fritts [11]. NOTE: The final version will discuss the SBL method much more extensively, including listing what is performed at compile time, what is performed at run time, and what additional hardware is necessary for supporting large scale speculative execution. 3 Experiments To examine the ....

Jason Fritts, "Architecture and Compiler Design Issues in Programmable Media Processors," Ph.D. Thesis, Department of Electrical Engineering, Princeton University, 2000.

....(TLS) 10] With each of these, there are mechanisms in either hardware and or software for storing speculative processor state, restoring the old processor state on a misspeculation, and checking for dependence conflicts during execution. More information on these methods can be found in Fritts [11]. 3. Speculative Broadcast Loop We propose the Speculative Broadcast Loop (SBL) method for the speculative execution of parallel loop iterations. This new vector like run time method is a simplified version of the multiscalar [8] and multithreaded [9] 10] methods. It combines SIMD parallelism ....

....Then we shall examine how Multi Level IfConversion is used for scheduling parallel loops for broadcast on the multi cluster architecture. Finally, large scale speculative execution and the resulting hardware modifications are discussed. More detailed information on SBL is available in Fritts [11]. 3.1. Finding Parallel Loops Profiling and register dependence analysis are employed to determine which loops are potential candidates for loop parallelization. An alternative for finding parallel loops would be to use parallel compiler methods such as those discussed in the prior section. ....

[Article contains additional citation context not shown here]

J. Fritts, "Architecture and Compiler Design Issues in Programmable Media Processors," Ph.D. Thesis, Dept. of Electrical Engineering, Princeton Univ., 2000.

No context found.

Jason Fritts, "Architecture and Compiler Design Issues in Programmable Media Processors," Ph.D. Thesis, Department of Electrical Engineering, Princeton University, 2000.

....processors specifically designed to provide efficient media processing across the full range of multimedia applications. The current methods of support for multimedia include: a) application specific hardware, b) multimedia extensions to general purpose processors, and c) media processors [1][2]. Application specific hardware provides excellent performance at low cost, but has limited flexibility and is unable to support evolving and future applications. General purpose processors with multimedia extensions provide the necessary flexibility, but are designed primarily for general purpose ....

....speech processing, and encryption and computer graphics. We augmented MediaBench with two additional video applications, H.263 and MPEG 4, to make it more representative of emerging multimedia applications. More information on MediaBench may be found in Lee et al. 10] 11] and Fritts et al. 12][2]. The compilation and simulation tools for this architecture style evaluation were provided by the IMPACT compiler, produced by Wen mei Hwu s group at the University of Illinois at Urbana Champaign [13] The IMPACT compiler is an aggressive profile based ILP compiler. In addition to the ....

[Article contains additional citation context not shown here]

J. Fritts, "Architecture and Compiler Design Issues in Programmable Media Processors," Ph.D. Thesis, Dept. of Electrical Engr., Princeton Univ., 2000.

No context found.

J. Fritts. Architecture and Compiler Design Issues in Programmable Media Processors, Ph.D. Thesis, 2000.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC