KRINTZ, C., AND CALDER, B. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of the ACM SIGPLAN 2001.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....samplebased profiler did not negatively impact performance. Radhakrishnan et al. 68] used the Ka#e Virtual Machine to measure the performance improvement possible by interpreting, rather than compiling, cold methods for a subset of the SPECjvm98 benchmarks with input size 1. Krintz and Calder [57] use o#ine profiles to generate annotations for Java Bytecode. These annotations are later used to guide selective optimization in a JVM. Plezbert and Cytron [66] propose continuous compilation, as well as several other selective optimization strategies, and simulate their e#ectiveness on C ....

Chandra Krintz and Brad Calder. Using annotations to reduce dynamic optimization time. In SIGPLAN 2001.

.... technique[4] Code annotation for program representations have not been as thoroughly researched as escape analysis, but recently, various annotations have been proposed for enhancing the performance of JIT compilers using Java bytecode and other intermediate representations[3] Krintz et al.[13] annotates bytecode for increasing the performance of programs executed by JVMs. Franz et al. 8] annotate programs syntax trees with escape information and encode them for safe transportation. Both [13] and [8] guarantee the safety of their annotations. Hannan uses a functional approach for the ....

....of JIT compilers using Java bytecode and other intermediate representations[3] Krintz et al. 13] annotates bytecode for increasing the performance of programs executed by JVMs. Franz et al. 8] annotate programs syntax trees with escape information and encode them for safe transportation. Both [13] and [8] guarantee the safety of their annotations. Hannan uses a functional approach for the annotation of escape information types in [10] In contrast to our work, he excludes dynamic class loading and gives no algorithm for the computation of types. 6 Conclusion Static object resolution is ....

C. Krintz and B. Calder. Using annotation to reduce dynamic optimization time. In SIGPLAN Conference on Programming Language Design and Implementation, pages 156--167, 2001.

....analyses to try to determine accurate estimates. The only thing that is required of the JIT compiler, however, is to utilize its knowledge of the number of available registers and its knowledge of how many variables are live at each point, and spill as many variables as necessary. As suggested in [KC01], this may be useful for a more efficient prioritized graph coloring. Spill ranking may be combined with a copy function, to allow the offline safeTSA producer to split live ranges and assign different spill ranking to each live range segment. 3 Escape Analysis Escape analysis is supported by ....

....depicts the resulting SafeTSA representation of class A, where # marks SObject O to be safe and eob to be object bound. 4 Related Work Recently, various annotations have been proposed for enhancing the performance of JIT compilers using Java bytecode and other intermediate representations [ANH00, KC01, BFHS02, Rei01]. The annotations of [KC01] and [BFHS02] will not compromise type safety. Both [ANH00] and [KC01] suggest utilizing annotations for register allocations. Their proposed mechanisms are not dissimilar to our spill rankings. They do not, however, convey information about live ranges as our kill ....

[Article contains additional citation context not shown here]

Chandra Krintz and Brad Calder. Using annotation to reduce dynamic optimization time. In SIGPLAN Conference on Programming Language Design and Implementation, pages 156--167, 2001.

....Runtime Platform (ORP) ORP was designed to support experimentation with different technologies in just in time compilers (JITs) garbage collection (GC) multithreading, and synchronization. Over the past five years, researchers have used ORP to conduct a number of MRTE implementation experiments [15 17, 19 21, 23, 25]. At least three different garbage collectors and eight different JITs have been developed and integrated with ORP. The version of ORP described in this paper is an internal research tool and is not publicly available. Three characteristics of MRTEs provide the key challenges to their ....

Krintz, C. and Calder, B., "Using Annotations to Reduce Dynamic Optimization Time," ACM Conference on Programming Language Design and Implementation, 2001.

....important methods guides insertion into the global priority queue of the OCT. Bytecode annotations of method priorities are inserted into the bytecode as method attributes using a bytecode re writing tool. A detailed study of using annotations to reduce dynamic optimization time can be found in [23]. We currently use a single OCT that synchronously compiles prioritized methods. When uncompiled methods are invoked, they are compiled using the baseline compiler. We may be able to gain additional performance benefits through the use of multiple OCTs once the Jalape no optimizing compiler is ....

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. Technical Report UCSD-CS00-663, University of California, San Diego, November 2000. http://www.ucsd.edu/users/calder/abstracts/UCSD-CSE00-663.html.

No context found.

KRINTZ, C., AND CALDER, B. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of the ACM SIGPLAN 2001.

No context found.

KRINTZ, C., AND CALDER, B. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of the ACM SIGPLAN 2001.

No context found.

C. Krintz and B. Calder. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 156-167, June 2001.

No context found.

Krintz, C., Calder, B.: Using Annotation to Reduce Dynamic Optimization Time. In: Proceedings of the ACM SIGPLAN 2001 Conference on Programming Language Design and Implementation. (1998)

No context found.

Krintz, C., Calder, B.: Using Annotation to Reduce Dynamic Optimization Time. In: Proceedings of the ACM SIGPLAN 2001 Conference on Programming Language Design and Implementation. (1998)

No context found.

C. Krintz and B. Calder. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 156--167, June 2001.

No context found.

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 ConferenceonProgramming Language Design and Implementation [1], pages 156-167. SIGPLAN Notices, 36(5), May 2001.

No context found.

C. Krintz and B. Calder. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 156--167, June 2001.

No context found.

C. Krintz and B. Calder. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 156--167, June 2001.

No context found.

C. Krintz and B. Calder. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of the ACM SIGPLAN 2001.

No context found.

KRINTZ, C., AND CALDER, B. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of the ACM SIGPLAN 2001.

....identifier is not available, the default JikesRVM collection system is used. We specify (possibly multiple) collection systems for a number of heap size ranges. We insert annotations into bytecode programs using an annotation language and a highly compact encoding that we developed in prior work [21]. We discover the best performing collection systems by repeatedly executing the program off line for a number of heap sizes and program inputs and recording the systems that enable minimum program execution times. Since the best performing collection system may depend on the underlying ....

KRINTZ, C., AND CALDER, B. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of the ACM SIGPLAN 2001.

....algorithms. This becomes a problem when code is generated on the fly while a user is waiting for execution to commence. Because the available time is limited, just in time compilation systems often don t make use of the best possible optimization algorithms. Annotation guided optimization [17, 1, 16, 20, 12, 22] o#ers a solution to this problem: analysis is performed o# line and communicated to the compilation system as program annotations. Annotations reduce just in time compilation overhead and enable optimizations that are too time consuming to perform on line. An example of such an analysis is escape ....

.... The first column shows the size of the actual Java class file, the second shows the amount of annotation information one would need to add in order to transport the results of escape analysis using the annotation format of the 16 annotation aware Open Runtime Platform (ORP) 7] described in [17]. The third column gives the size of our CAST format without annotation information encoded, and the last column gives the delta between our ACAST format (including annotations) and CAST. The size overhead for incorporating annotations into ASTs is smaller than that for doing the same for Java ....

[Article contains additional citation context not shown here]

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 Conference on Programming Language Design and Implementation, pages 156--167, Snowbird, Utah, June 20--22, 2001.

....user is waiting for execution to commence, for example, in the context of Java dynamic compilation. A partial solution has been provided by annotation based techniques, enabling optimized execution speeds for dynamically compiled JVM code programs with reduced just in time compilation overhead [33], and communicating analysis information that is too time consuming to collect on line [5, 30, 43, 24, 45] Annotations make costly optimizations feasible in dynamic optimization environments. Analysis results that are time and space consuming to generate can then be employed since the analyses ....

....side is guaranteed not only to be syntactically correct, but also to conform to all the static semantic rules that were part of the encoding; this includes the correctness of the annotations. At the code producer s side, our annotation insertion system is an extension of our prior annotation work [33]. The system is invoked during program encoding and enables both static analysis and execution characteristics about the program to be collected. The annotation insertion process consists of three steps: program compilation, static bytecode analysis, and decompilation. We rst compile the program ....

[Article contains additional citation context not shown here]

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 Conference on Programming Language Design and Implementation [2], pages 156-167. SIGPLAN Notices, 36(5), May 2001.

....performance. Such methods are identi ed in one of two ways. The rst approach is to perform instrumentation and pro ling of methods while the program is executing [7, 2, 5] i.e. on line. The other approach is to perform o line pro ling and to communicate the pro le data via annotation [9, 3, 8, 12]. Both on line and o line pro ling techniques have disadvantages that impact program performance. On line pro ling introduces overhead for instrumentation and measurement sampling. O line pro ling may be imprecise since, in most cases, the inputs used to execute the program di er from those ....

....system, on average, improves the overall performance of the programs studied by 9 to 15 . To implement our system we extended two existing systems: JikesRVM [1] a dynamic and adaptive Java compilation environment and an o line pro ling and annotation system that 3 we developed in prior work [9]. The system that we describe herein uses pro le information made available via program annotation to guide method optimization. All unannotated methods are monitored by the on line pro ling system so that pro le inaccuracies, if any, do not impact program performance. By coupling the use of ....

[Article contains additional citation context not shown here]

C. Krintz and B. Calder. Using Annotation to Reduce Dynamic Optimization Time. In Proceedings of the ACM SIGPLAN 2001.

....algorithms. This becomes a problem when code is generated on the y while a user is waiting for execution to commence. Because the available time is limited, just in time compilation systems often don t make use of the best possible optimization algorithms. Annotation guided optimization [19, 3, 18, 22, 14, 24] o ers a solution to this problem: analysis is performed o line and communicated to the compilation system as program annotations. Annotations reduce just in time compilation overhead and enable optimizations that are too time consuming to perform on line. An example of such an analysis is escape ....

....in bytes. The rst column shows the size of the actual Java class le, the second shows the amount of annotation information one would need to add in order to transport the results of escape analysis using the annotation format of the annotation aware Open Runtime Platform (ORP) 9] described in [19]. The third column gives the size of our CAST format without annotation information encoded, and the last column gives the delta between our ACAST format (including annotations) and CAST. The size overhead for incorporating annotations into ASTs is smaller than that for doing 16 Source ByteCode ....

[Article contains additional citation context not shown here]

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 Conference on Programming Language Design and Implementation [2], pages 156-167. SIGPLAN Notices, 36(5), May 2001.

....delay. The second class of optimizations we present, substantially reduces compilation delay while maintaining optimized execution performance. Two of the techniques we have developed to do this are background compilation [3] in the IBM Jalapeno Virtual Machine and annotation assisted compilation [1] in the Intel Corporation Open Runtime Platform (ORP) Both of these infrastructures (as with others) use two dynamic compilers for program execution. The first is a very fast, non optimizing compiler and the second is a slower, optimizing compiler. The code produced by the latter performs much ....

Chandra Krintz, Brad Calder. Using Annotation to Reduce Dynamic Optimization Time, UC San Diego Technical Report CS

....important methods guides insertion into the global priority queue of the OCT. Bytecode annotations of method priorities are inserted into the bytecode as method attributes using a bytecode re writing tool. A detailed study of using annotations to reduce dynamic optimization time can be found in [23]. We currently use a single OCT that synchronously compiles prioritized methods. When uncompiled methods are invoked, they are compiled using the baseline compiler. We may be able to gain additional performance benefits through the use of multiple OCTs once the Jalapeno optimizing compiler is ....

Krintz C, Calder B. Using annotations to reduce dynamic optimization time. Technical Report UCSD-CS00-663, University of California, San Diego, November 2000. http://www.ucsd.edu/users/calder/abstracts/UCSD-CSE00-663.html.

No context found.

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 Conference on Programming Language Design and Implementation, pages 156--167, Snowbird, Utah, June 20--22, 2001.

No context found.

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 Conference on Programming Language Design and Implmenetation (PLDI), pages 156--167. ACM Press, May 2001.

No context found.

Chandra Krintz and Brad Calder. Using annotations to reduce dynamic optimization time. In Programming Language Design and Implementation, June 2001.

No context found.

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 Conference on Programming Language Design and Implementation, pages 156--167, Snowbird, Utah, June 20--22, 2001.

No context found.

Chandra Krintz and Brad Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN'01 conference on Programming language design and implementation, pages 156--167. ACM Press, 2001.

No context found.

C. Krintz and B. Calder. Using Annotation to Reduce Dynamic Optimization Time. In SIGPLAN Conference on Programming Language Design and Implementation, pages 156--167, 2001.

No context found.

C. Krintz and B. Calder. Using annotation to reduce dynamic optimization time. In SIGPLAN Conference on Programming Language Design and Implementation, 2001.

No context found.

C. Krintz and B. Calder. Using annotations to reduce dynamic optimization time. In Proceedings of the ACM SIGPLAN '01 Conference on Programming Language Design and Implementation, pages 156--167, Snowbird, Utah, June 20--22, 2001.

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