Can program profiling support value prediction?

Gabbay, Freddy; Mendelson, Avi

doi:10.1109/micro.1997.645817

Cited by 64 publications

(54 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Fig. 6 shows that our approach results in an increase in execution time less 8. Reduction in miss rates using static and dynamic dictionary than 1% (average) even when a dictionary of 32 entries is used.…”

Section: Evaluation Of the Power-aware Dfvc (Pa-dfvc)mentioning

confidence: 90%

“…One direction concentrates on exploiting the well known phenomenon of locality and especially value locality. Value locality has been initially utilized in the design of value reuse and value prediction mechanisms for superscalar processors [8,16].…”

Section: The Dynamic Behavior Of Frequent Valuesmentioning

confidence: 99%

“…One such simple, yet effective, mechanism is the dictionary (or directory) for frequent values [20,23,25,26]. A frequent-value dictionary stores the program's frequent values (e.g., the 32-bit value "0") and replaces all their occurrences in the cache with the respective dictionary indices (e.g., 8-bit indices for a 256-entry dictionary). However, until now, no mechanism has been proposed to implement a dynamic dictionary for caches [23,24]; in other words, a dictionary whose contents can adapt to the requirements of the running program.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Dynamic Dictionary-Based Data Compression for Level-1 Caches

Κεραμίδας

Aisopos

2006

Architecture of Computing Systems - ARCS 2006

View full text Add to dashboard Cite

Abstract. Data cache compression is actively studied as a venue to make better use of onchip transistors, increase apparent capacity of caches, and hide the long memory latencies. While several techniques have been proposed for L2 compression, L1 compression is an elusive goal. This is due to L1's sensitivity to latency and the inability to create compression schemes that are both fast and adaptable to program behavior, i.e. dynamic. In this paper, we propose the first dynamic dictionary-based compression mechanism for L1 data caches. Our design solves the problem of keeping the compressed contents of the cache and the dictionary entries consistent, using a timekeeping decay technique. A dynamic compression dictionary adapts to program behavior without the need of profiling techniques and/or training phases. We compare our approach to previously proposed static dictionary techniques and we show that we surpass them in terms of power, hit ratio and energy delay product.

show abstract

Section: Evaluation Of the Power-aware Dfvc (Pa-dfvc)mentioning

confidence: 90%

Section: The Dynamic Behavior Of Frequent Valuesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Dynamic Dictionary-Based Data Compression for Level-1 Caches

Κεραμίδας

Aisopos

2006

Architecture of Computing Systems - ARCS 2006

View full text Add to dashboard Cite

show abstract

“…The difficulty in hardware value prediction is in finding out whether an instruction is appropriate for prediction at runtime within acceptable period while keeping the hardware cost reasonable. In order to alleviate the severity of the problem, some work [7], [19] require compiler and/or profiling support such that only values that have high confidence are predicted. Fu et al [4] proposed a software-only value prediction approach that does not require any prediction hardware, and can run on existing microprocessors.…”

Section: Related Workmentioning

confidence: 99%

Exploiting Thread-Level Speculative Parallelism with Software Value Prediction

Li¹,

Yang²,

Du³

et al. 2005

Advances in Computer Systems Architecture

View full text Add to dashboard Cite

Software value prediction (SVP) is an effective and powerful compilation technique helping to expose thread-level speculative parallelism. With the help of value analysis and profiling, the compiler identifies critical and predictable variable values and generates speculatively parallelized programs with appropriate value prediction and misprediction recovery code. In this paper, we examine this technique in detail, describe a complete and versatile SVP framework and its detailed implementation in a thread-level speculative parallelization compiler, and present our evaluation results with Spec2000Int benchmarks. Our results not only confirm quantitatively that value prediction is essential to thread-level speculative parallelism, they also show that the corresponding value prediction can be achieved efficiently and effectively by software. We also present evaluation results of the overhead associated with software value prediction and the importance of different value predictors in speculative parallel loops in Spec2000Int benchmarks.

show abstract

“…There is a large body of work relating to value prediction and speculation, e.g., see [11,12] . This involves using a combination of compiler generated information and runtime information.…”

Section: Related Workmentioning

confidence: 99%

Goal-Directed Value Profiling

Watterson

Debray

2001

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Compilers can exploit knowledge that a variable has a fixed known value at a program point for optimizations such as code specialization and constant folding. Recent work has shown that it is possible to take advantage of such optimizations, and thereby obtain significant performance improvements, even if a variable cannot be statically guaranteed to have a fixed constant value. To do this profitably, however, it is necessary to take into account information about the runtime distribution of values taken on by variables. This information can be obtained though value profiling. Unfortunately, existing approaches to value profiling incur high overheads, primarily because profiling is carried out without consideration for the way in which the resulting information will be used. In this paper, we describe an approach to reduce the cost of value profiling by making the value profiler aware of the utility of the value profiles being gathered. This allows our profiler to avoid wasting resources where the profile can be guaranteed to not be useful for optimization. This results in significant reductions in both the time and space requirements for value profiling. Our approach, implemented in the context of the alto link-time optimizer, is an order of magnitude faster, and uses about 5% of the space, of a straightforward implementation.

show abstract

Can program profiling support value prediction?

Cited by 64 publications

References 9 publications

Dynamic Dictionary-Based Data Compression for Level-1 Caches

Dynamic Dictionary-Based Data Compression for Level-1 Caches

Exploiting Thread-Level Speculative Parallelism with Software Value Prediction

Goal-Directed Value Profiling

Contact Info

Product

Resources

About