Using branch handling hardware to support profile-driven optimization

Conte, Thomas M.; Patel, Burzin A.; Cox, Jeremiah

doi:10.1145/192724.192726

Cited by 30 publications

(12 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Whether the data was gathered through hardware performance counters [Anderson et al 1997], stratified sampling [Sastry et al 2001], or even potentially in fixed ranges [Zilles and Sohi 2001;Zhou et al 2004], the end result is essentially a list of equivalent items and their counts. While there exists some specialized software and hardware systems that attempt to tightly compress particular types of traces [Anderson et al 1997;Conte et al 1996Conte et al , 1994Narayanasamy et al 2003;Sastry et al 2001;Zilles and Sohi 2001], we believe that we are the first to present a general hardware-based methodology for storing profiles in a hierarchical fashion.…”

Section: Profile Treesmentioning

confidence: 99%

“…Several software techniques, such as binary instrumentation [Buck and Hollingsworth 2000;Luk et al 2005;Srivastava et al 2001;Srivastava and Eustace 1994;Bus et al 2004] and sampling [Arnold and Ryder 2001], can be used to generate and analyze this profile information with only a moderate amount of overhead [Arnold and Ryder 2001;Ball and Larus 1996;Calder et al 1997;Chilimbi 2001;Chilimbi and Hirzel 2002;Duesterwald and Bala 2000;Hirzel and Chilimbi 2001;Larus 1999]. Recently, several researchers have proposed various forms of architectural support [Anderson et al 1997;Conte et al 1996Conte et al , 1994Dean et al 1997;Heil and Smith 2000;Narayanasamy et al 2003;Peri et al 1999;Sastry et al 2001;Yang and Gupta 2002;Zilles and Sohi 2001] with the aim of increasing accuracy and further reducing the overhead of software-based techniques. Value profiles can be exploited to perform code specialization [Calder et al 1997], value prediction [Lipasti and Shen 1996;Zhou et al 2003], and value encoding [Yang and Gupta 2002;Yang et al 2000].…”

Section: Profiling With Adaptive Precisionmentioning

confidence: 99%

“…Many researchers have examined the next steps that hardware-assisted profiling should take. Proposed schemes range from those which use existing hardware on the processor to gather information, which is later processed by a software program [Anderson et al 1997;Conte et al 1994;Peri et al 1999], to programmable profiling coprocessors [Zilles and Sohi 2001]. Conte et al [1996] uses profile buffers to collect and analyze information.…”

Section: Hardware-assisted Profilingmentioning

confidence: 99%

See 2 more Smart Citations

Formulating and implementing profiling over adaptive ranges

Mysore

Agrawal

Neuber

et al. 2008

ACM Trans. Archit. Code Optim.

View full text Add to dashboard Cite

Modern computer systems are called on to deal with billions of events every second, whether they are executed instructions, accessed memory locations, or forwarded packets. This presents a serious challenge to those who seek to quantify, analyze, or optimize such systems, because important trends and behaviors may easily be lost in a sea of data. We present range-adaptive profiling (RAP) as a new and general-purpose profiling method capable of hierarchically efficiently classifying streams of data in hardware. Through the use of RAP, events in an input stream are dynamically classified into increasingly precise categories, based on the frequency with which they occur. The more important a class, or range of events, the more precisely it is quantified. Despite the dynamic nature of our technique, we build upon tight theoretic bounds covering both worst-case error, as well as the required memory. In the limit, it is known that error and the memory bounds can be independent of the stream size and grow only linearly with the level of precision desired. Significantly, we expose the critical constants in these algorithms and through careful engineering, algorithm redesign, and use of heuristics, we show how a high-performance profile system can be implemented for range-adaptive profiling. RAP can be used on various profiles, such as PCs, load values, and memory addresses, and has a broad range of uses, from hot-region profiling to quantifying cache miss value locality. We propose two methods of implementation of RAP, one in software and the other with specialized hardware, for which we also describe our prototype FPGA implementation. We show that with just 8KB of memory, range profiles can be gathered with an average accuracy of 98%.

show abstract

Section: Profile Treesmentioning

confidence: 99%

Section: Profiling With Adaptive Precisionmentioning

confidence: 99%

Section: Hardware-assisted Profilingmentioning

confidence: 99%

See 1 more Smart Citation

Formulating and implementing profiling over adaptive ranges

Mysore

Agrawal

Neuber

et al. 2008

ACM Trans. Archit. Code Optim.

View full text Add to dashboard Cite

show abstract

“…In [9], a technique is described on how to estimate trivially a program's edge execution frequencies by periodically reading the contents of BTB. In [10] a hardware called Profile Buffer is proposed, which counts the number of times a branch is taken and not taken.…”

Section: A Hardware-support Profilingmentioning

confidence: 99%

A Runtime Profile Method for Dynamic Binary Translation Using Hardware-Support Technique

Guan

Yang

Liu

et al. 2009

2009 First International Conference on Information Science and Engineering

View full text Add to dashboard Cite

Profile data is valuable for identifying program hotspots and guiding optimizations. Traditional software profiling techniques incur significant overhead and are not suitable for DBT (Dynamic Binary Translation) systems. Hardware can support profile collection through either counters or timer interrupts that permit collection of statistical samples via software. Most hardware-support profiling systems can only achieve either high profile accuracy or low overhead. In this paper, we propose a novel profile approach on DBT using hardware support technique to achieve rapidly and accurately collecting profile information with minimal runtime overhead. This approach makes use of instrumentation code and a set of profiling hardware which supports operations of updating counters. It is believed that such a software-hardware collaborative approach will serve to provide a strong foundation for optimizing DBT systems.

show abstract

“…This profile operation is somewhat more complex than standard basic block or instruction count profiles, since a simulation of all the alternative component predictors must be performed. However, this information could easily be gathered using on-line performance tools [5,4] that provide stochastic instruction sampling, or, as done for this paper, using binary instrumentation [15].…”

Section: Static Hybrid Predictorsmentioning

confidence: 99%

Static methods in hybrid branch prediction

Grunwald¹,

Lindsay²,

Zorn³

Proceedings. 1998 International Conference on Parallel Architectures and Compilation Techniques (Cat. No.98EX192)

View full text Add to dashboard Cite

Hybrid branch predictors combine the predictions of multiple single-level or two-level branch predictors. The prediction-combining hardware -the "meta-predictor" -may itself be large, complex and slow. We show that the combination function is better performed statically, using prediction hints in the branch instructions. The hints are set by profiling or static analysis. Although the meta-predictor is static, the actual predictions remain dynamic, so there is little risk of worst-case performance. An important advantage of our approach is that a branch site only causes interference within a single component predictor, reducing capacity demands. We argue that our proposal is implementable, and that it addresses the scaling issues currently facing hardware designers.We show that the static hybrid method we propose is more effective than existing techniques based on dynamic selection, and requires less hardware. For example, one result shows a conventional 4096-bit dynamic selection mechanism getting a 4.7% average miss rate, while our static approach gets 3.6%. These results are obtained with the Instruction Benchmark Suite (IBS), a realistic whole-system benchmark, and the SPECint95 suite, using realistic hardware sizes. All the results we present are based on a crossvalidation methodology, in which the profile data used for static selection are based on training inputs that are entirely different from the inputs used to evaluate the performance of the technique.

show abstract

Using branch handling hardware to support profile-driven optimization

Cited by 30 publications

References 17 publications

Formulating and implementing profiling over adaptive ranges

Formulating and implementing profiling over adaptive ranges

A Runtime Profile Method for Dynamic Binary Translation Using Hardware-Support Technique

Static methods in hybrid branch prediction

Contact Info

Product

Resources

About