T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, 22(3):490--505, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... data in C programs [105] Since they apply these transformations by hand, Truong also describes plans for automating the data layout techniques based upon profile information [104] Kistler and Franz evaluate a profile based optimization that reorders members in objects to improve spatial locality [59]. Franz and Kistler also propose physically splitting frequently and infrequently accessed members of objects to improve cache performance [38] Other data transformations apply specifically to heap allocated data in a garbage collected environment. Moon describes a mostly depth first copying ....

Thomas Kistler and Michael Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, 22(3):490--505, May 2000.

....cache (cache hits) to those that are fetched from main memory (cache misses) remain the same, with half the memory size, without compromising the application execution time. Traditionally, data reorganization has been used to improve the execution time of applications for a fixed target processor [3, 9, 13, 19, 28]. To a large extent, previous work in this area has been a semi automated process, and in the context of pointer based programs ubiquitous to the C programming language used extensively in the embedded systems domain it has been mainly restricted to memory that is statically allocated. Our ....

....our modeling of the processor core are contained in a technical report [14] 4.2 The Compilation Environment Benchmarks from the DIS, OLDEN and SPEC2000 suites were selected for detailed analysis. The OLDEN benchmarks provide a common frame of reference with previous work on data reorganization [9, 13, 28]. The others provide insight into larger programs. The benchmarks were executed using large input sets, whereas profile information was gathered using much smaller workloads (e.g. a trace size of few million memory instructions sampled along program hot spots) Table 5 summarizes the benchmarks ....

[Article contains additional citation context not shown here]

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, 22(3):490--505, May 2000.

....we look at the impact of better memory system performance on execution cycles and processor performance. 3.1 Benchmarks and Methodology We performed detailed simulations of four Olden, two DIS and one SPEC2000 benchmark. The Olden benchmarks provide a common frame of reference with previous work[8, 18, 32]. The other benchmarks provide insight into large programs. The application characteristics are shown in Table 2. We simulate the benchmarks using both input data sets listed in Table 2. The profile information used to determine which data types to remap was obtained using much smaller data sets. ....

....that for the faster Pentium III processor, despite a factor of 8 reduction in the second level cache size, we attain 20 improvement compared to the UltraSparc. 4 Related Work Previous work has attacked the processor memory gap by computation reordering to increase spatial and temporal locality[5, 8, 7, 32, 18, 22]. Most recently, Crummey et al. 22] explore a coordinated data and computation reordering 10 Performance 3 0 0 0 0 157 158 159 are 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 ART DM Field Health PerimeterT SP T eeAdd Normalized 188 tion Time 8 eedup Pentium III Pentium II UltraSpar ....

[Article contains additional citation context not shown here]

T. Kistler, and M. Franz. "Automated data-member layout of heap objects to improve memory-hierarchy performance ". In ACM Transactions on Programming Languages and Systems, Volume 22, No. 3, pages 490-505, May 2000.

No context found.

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, 22(3):490--505, 2000.

No context found.

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, 22(3):490--505, 2000.

No context found.

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, 22(3):490--505, 2000.

....representation: ffl Higher representation levels allow the program to be portable to a wider range of targets. ffl Similarly, higher representation levels make it easier to optimize the program for specific target characteristics. This may be critical for good performance on modern architectures [24, 4]. ffl Higher representation levels make it easier for the consumer to prove or disprove the type safety of the program. ffl Correspondingly, higher representation levels require more work on the part of the consumer, increasing the time required (and the energy required in embedded ....

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, May 2000.

....thereby becomes immune to changes in underlying technology. Execution Eciency. Mobile code should run as eciently (in terms of speed and memory consumption) as possible on the target platform. It may be important on modern architectures to optimize the code for speci c target characteristics [16,3]. Execution Latency. Mobile code should also be able to start executing on the target platform with minimal delay, implying ecient translation of the transportable format into an executable on the target machine. Compactness. Programs should be transmitted in a format that is as compact as ....

Kistler, T. and M. Franz, Automated data-member layout of heap objects to improve memory-hierarchy performance, ACM Transactions on Programming Languages and Systems (2000).

....representation, we need to generate native code from it prior to execution . Ho wever, this usually happens at run time while the user is waiting, and must be fast. It is also very common to use mobile code in a framework that uses dynamic optimization to incrementally improve code quality[5]. We can improve both start up time and quality of generated native code by shipping relevant annotations along with the mobile code. Typically, these annotations are the results of some static analysis, and help the consumer to generate better code faster. The goal is to shift the load of ....

....of the program. Since our mobile code format contains all the information provided by the programmer at the source language level, the runtime system at the code consumer site can readily use this information to provide optimizations and services based on source language guarantees. Kistler[5] uses the availability of the AST to make dynamic re compilation at runtime feasible. Furthermore, distributing code in source languageequivalent form provides the runtime system with the choice of a platform tailored intermediate representation. For example, it is possible to use an existing ....

T. Kistler and M. Franz; "Automated data-member layout of heap objects to improve memory-hierarchy performance"; in ACM Transactions on Programming Languages and Systems, May 2000.

....of representation: Higher representation levels allow the program to be portable to a wider range of targets. Similarly, higher representation levels make it easier to optimize the program for specific target characteristics. This may be critical for good performance on modern architectures [24, 4]. Higher representation levels make it easier for the consumer to prove or disprove the type safety of the program. Correspondingly, higher representation levels require more work on the part of the consumer, increasing the time required (and the energy required in embedded applications) ....

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, May 2000.

....representation: Higher representation levels allow the program to be portable to a wider range of targets. Similarly, higher representation levels make it easier to optimize the program for specific target characteristics. This may be critical for good performance on modern architectures [24, 4]. Higher representation levels make it easier for the consumer to prove or disprove the type safety of the program. Correspondingly, higher representation levels require more work on the part of the consumer, increasing the time required (and the energy required in embedded applications) ....

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, May 2000.

No context found.

T. Kistler and M. Franz, "Automated data-member layout of heap objects to improve memory-hierarchy performance," ACM Transactions on Programming Languages and Systems, vol. 22, no. 3, pp. 490--505, 2000.

No context found.

T. Kistler and M. Franz. Automated data-member layout of heap objects to improve memory-hierarchy performance. ACM Transactions on Programming Languages and Systems, 22(3):490--505, May 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