I.-C. K. Chen, J. T. Coffey, and T. N. Mudge. Analysis of branch prediction via data compression. In Proceedings of the 7th Intl. Conference on Architectural Support for Programming Languages and Operating Systems, pages 128-- 137, Oct. 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... The first study that showed dynamic characteristics of loops was presented by Kobayashi [14] A more recent study presented a thorough examination of the dynamic characteristics of loops [5] Since then, branch prediction has drawn a lot of attention and several mechanisms have been proposed [2, 6, 7, 9, 11, 12, 13]. Some alternative mechanisms include the branch address cache [28] trace cache [21] and control flow prediction with a tree like predictor [8] Seznec et al. have proposed predicting several branches in the same cycle to produce larger traces of instructions using multiple block ahead ....

J. T. Co#ey I. K. Chen and T. Mudge. Analysis of branch prediction via data compression. In Proc. of the Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, pages 128--137, Cambridge, MA, October, 1996.

....the prediction are very important, and second, the duration of execution of those phases is important. A classic prediction model that is easily implementable in hardware is a Markov Model. Markov Models have been used in computer architecture to predict both prefetch addresses [13] and branches [8] in the past. The basic idea behind a Markov Model is that the next state of the system is related to the last set of states. The intuition behind this design is that phase information tends to be characterized by many sections of stable behavior interspersed with abrupt phase changes. The key is ....

I.-C. Chen, J. T. Coffey, and T. N. Mudge. Analysis of branch prediction via data compression. In Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, pages 128-- 137, October 1996.

.... a two level table organization similar to the ones used for branch prediction[8] One of the arguments for the choice of this particular organization is its resemblance to finite context predictors, used in text compression that have in general the best known prediction (compression) performance[9, 10]. A variety of implementations of the proposed value predictor are evaluated by comparing prediction performance with the ideal performance reported in [7] A number of issues relevant to context based prediction such as order, aliasing and hash functions are investigated. The performance of the ....

.... organization of successful correlation predictors [8, 17, 18, 19, 20] Recently it was demonstrated theoretically that two level tables, as used for control dependence prediction, resemble the prediction models used for text compression that are known to have the best prediction performance [10]. This is an important observation because it re enforces the approach used for control dependence prediction and also suggest that methods based on compression can be used for data value prediction. Data can take a large range of values, this gives rise to the problem of how to reduce the values ....

I.-C. K. Cheng, J. T. Coffey, and T. N. Mudge, "Analysis of branch prediction via data compression," in Proceedings of the 7th International Conference on Architectural Support for Programming Languages and Operating Systems, October 1996.

....of the branches is dictated by the maximum history length that size can entertain. Some wrong history mispredictions will therefore occur, even though they might be eliminated by a more idealized organization. At the limit, one might consider infinite history or prediction by partial matching [4]. This would not measure wrong history, but rather the intrinsic predictability of a branch. Our approach characterizes the degree to which a particular predictor size produces wrong history mispredictions and a different history type for the same predictor size could remove those mispredictions ....

....predictor outperforms the other two in terms of the whole program. A. 3 Interference Free Prediction and Upper Bounds Recently, researchers have modeled branch prediction using Markovian chains and showed that two level branch prediction is a simplified version of prediction by partial matching [4]. Thus, an open question at the theoretical level is how to define the states in the Markovian chains such that we will have high confidence about the transitions between states (very low high transition prob21 Fig. 13. Dynamic and static distribution of branches preferring local or global ....

I.-C. Chen, J. T. Coffey, and T. N. Mudge. Analysis of branch prediction via data compression. In Proceedings of the Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, pages 128--37, Oct. 1996.

....take advantage of the improving speed of processors versus disk, and notes that the growing disparity between these system elements makes compression close to the processor an appealing feature. Reference [9] and the TinyRISC effort use compression to reduce embedded system code size. References [10] and [11] use compression techniques to increase branchprediction accuracy in microprocessors. Reference [12] describes algorithms and data structures for compressedmemory machines. The primary contributions of this paper are as follows: Motivation and benefits of main memory compression; ....

I.-C. K. Chen, J. T. Coffey, and T. N. Mudge, "Analysis of Branch Prediction via Data Compression," Computer Architecture News 24, 128--137 (October 1996).

.... organization of successful correlation predictors [10, 11, 12, 13] Recently it was demonstrated theoretically that two level tables, as used for control dependence prediction, resemble the prediction models used for text compression that are known to have the best prediction performance [14]. This is an important observation because it re enforces the approach used for control dependence prediction and also suggest that methods based on compression can be used for data value prediction. Data can take a large range of values, this gives rise to the problem of how to reduce the values ....

I.-C. K. Cheng, J. T. Coffey, and T. N. Mudge, "Analysis of branch prediction via data compression," in Proceedings of the 7th International Conference on Architectural Support for Programming Languages and Operating Systems, October 1996.

....literature on branch prediction algorithms is vast. We will not try to summarize the literature here. To the best of our knowledge we are the first ones to identify a formal algorithm for computing the miss rate for branch prediction algorithms. The work that has some similarity with our work is [CCM96]. In [CCM96] authors draw parallel between branch prediction schemes and algorithms for data compression. Based on this analogy they prove optimality results for some specific branch prediction schemes. Due to this similarity we plan to take a closer look at the literature on compression. In an ....

....on branch prediction algorithms is vast. We will not try to summarize the literature here. To the best of our knowledge we are the first ones to identify a formal algorithm for computing the miss rate for branch prediction algorithms. The work that has some similarity with our work is [CCM96] In [CCM96] authors draw parallel between branch prediction schemes and algorithms for data compression. Based on this analogy they prove optimality results for some specific branch prediction schemes. Due to this similarity we plan to take a closer look at the literature on compression. In an abstract sense ....

I.K. Chen, J.T. Coffey, and T.N. Mudge. Analysis of branch prediction via data compression. In Proceedings of the Seventh International Conference on Architec14 tural Support for Programming Languages and Operating Systems (ASPLOS-VII), October 1996.

....used this profile, along with a desired accuracy they wanted to achieve, to select which histories would be used to generate value predictions, and which histories would predict low confidence. This was then used to guide confidence estimation for their value prediction architecture. Chen et al. [5] examined using techniques from data compression to improve the performance of branch prediction. They looked at using Prediction by Partial Matching (PPM) where there are M tables from size 2 to 2 M . Each PPM entry contains a frequency for the number of times the next bit was 0 (not taken) ....

I.-C. Chen, J.T. Coffey, and T.N. Mudge. Analysis of branch prediction via data compression. In Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, October 1996.

....lengths for 2 level predictors performed better for different classes of branches. Although the design space for branch predictors grew tremendously, an understanding of how and why these predictors worked did not. More recently, however, a number of papers, such as [23] 12] 6] 16] and [4], have provided insights into mechanisms for creating more accurate branch predictors. Our goal is to continue this process towards a deeper understanding about the nature of branch behavior and branch predictor performance. In this paper, we introduce a new metric for branch behavior, branch ....

....using some form of either dynamic or static classification. Several proposals have schemes that vary the history length either on a per branch basis or program basis. Also, work has been done to better understand the branch behavior and dynamics of 2 level branch predictors. Chen et al. [4] use techniques established in the field of data compression to form a theoretical basis for branch prediction and make suggestions for improvement. They develop this theoretical basis by demonstrating that current correlated predictors are simplifications of the PPM (Prediction by Partial ....

[Article contains additional citation context not shown here]

I.-C. K. Chen, J. Coffey, and T. Mudge. Analysis of branch prediction via data compression. In Proceedings of the 7th Intl. Conf. on Architectural Support for Programming Languages and Operating Systems (ASPLOS VII), pages 128-- 137, Cambridge, MA, October 1996.

....of times. As a result, the loop terminating branch is mispredicted most of the time. However, since the loop decrements to xlen = len (OPSIZ 8) while (xlen 0) op t ) dstp) 0] cccc; op t ) dstp) 1] cccc; op t ) dstp) 2] cccc; op t ) dstp) 3] cccc; op t ) dstp)[4] = cccc; op t ) dstp) 5] cccc; op t ) dstp) 6] cccc; op t ) dstp) 7] cccc; dstp = 8 OPSIZ; xlen = 1; Fig. 3. BDP removes 79 of the mispredictions of this difficult to predict branch in gcc.gcc. 0, using the loop counter value as an input to branch prediction permits ....

....cache like structure which is used to predict only the difficult, exceptional cases. The replacement policy used by the REP explicitly searches for patterns that lead to correct branch predictions differing from the one made by the backing predictor. The REP derives from partial pattern matching [4, 17], and these structures have appeared in various kinds of predictors [11, 18, 25] most similarly the YAGS branch predictor [5] Referring again to Fig. 5, while the VHT is being accessed, the REP is accessed in parallel with the PC and global branch history (GBH) The value history is used for ....

I-Cheng K. Chen, John T. Coffey, Trevor N. Mudge, "Analysis of Branch Prediction via Data Compression," 7th Intl. Conf. on Arch. Support for Prog. Lang. and Op. Sys., pp. 128137, Oct. 1996.

....in this study, with 4 entry filter, obtains 23.7 misprediction rate with a 512 entry, four way associative dual path hybrid predictor as second stage. Hybrid prediction for conditional branches was first proposed in [McFar93] Recent results can be found in [CHP95] and [ECP96] Chen et al. [CCM96] propose Partial Prefix Matching prediction for conditional branch prediction and show that a PPM predictor performs better than a twolevel predictor for a similar hardware budget. Since a PPM predictor chooses the prediction of the longest pattern for which a prediction is available (choosing ....

I-Cheng K.Chen, John T.Coffey, Trevor N. Mudge. Analysis of Branch Prediction via Data Compression. ASPLOS'96 Proceedings.

....and deallocating nodes, as in the adaptive CWT algorithm. In [140] idealized versions of these algorithms are applied to conditional branch prediction with some success. However, a realistic hardware implementation has not yet been described. Another version of the PPM algorithm is described in [141]. It has been successfully applied to areas such as data compression [138] file prefetching [142] and cache replacement policies [98] The PPM predictor computes probabilities for symbols (as is the case with the CWT and the CWT based PPM algorithm) Its corresponding version for conditional ....

....101 bit pattern in the 3rd order Markov predictor, then it will look for pattern 01 in the 2nd order Markov predictor, etc. The update step of the frequency counts, across different order Markov models in the PPM algorithm, varies. Throughout this thesis, we will use an update exclusion policy [141]. This protocol excludes all predictors with a lower order (i.e. lower than the one that actually makes the prediction) from being updated. Only the predictor that makes the decision, and the predictors with a higher order, are updated. Implementations of the PPM algorithm are discussed in the ....

[Article contains additional citation context not shown here]

I.C.K. Chen, J.T. Coffey, and T.N. Mudge. Analysis of Branch Prediction via Data Compression. In Proceedings of the International Conference on Architectural support for programming languages and operating systems, pages 128--137, October 1996.

....to have a predictor table with infinite capacity so that every unique branch substream defined by an (address, history) pair will have a dedicated predictor. Chen et al. have shown that two level predictors are close to being optimal, provided unlimited resources for implementing the predictors [3]. Real world constraints, of course, do not permit this. Chip die area budgets and access time constraints limit predictor table size, and most tables proposed in the literature are further constrained in that they are direct mapped and without tags. Fixed sized predictor tables lead to a ....

I.-C.K. Chen, J.T. Coffey, and T.N. Mudge. Analysis of branch prediction via data compression. In Proceedings of the 7th International Conference on Architectural Support for Programming Languages and Operating Systems, Oct. 1996.

....[118] determine a path length per branch by profiling each branch in a separate profiling run. Their technique is an instance of profile based classification which is much more fine grained than the classifications used in Section 9.2 (see Section 10.2. 1 for more detail) Cheng, Coffey and Mudge [28] propose Partial Prefix Matching (PPM) prediction, based on a compression technique, for conditional branch prediction. They show that a PPM predictor performs better than a two level predictor for a similar hardware budget. Since a PPM predictor predicts for the longest pattern for which a ....

I-Cheng K.Chen, John T.Coffey, Trevor N. Mudge. Analysis of Branch Prediction via Data Compression. ASPLOS'96 Proceedings.

....we present an overview of the compression techniques that have been proposed so far for the task of branch prediction. We discuss two major compression techniques: the first is the Prediction by Partial Matching (PPM) 26, 27] algorithm that has been proposed for predicting conditional branches in [28] and the second (and most recent) one is the Context Weight Tree (CWT) method [29] which has been presented in [30] as a another potential conditional branch predictor. The primary goal of data compression is to represent an original data sequence with another that has fewer data elements. Modern ....

....generating and deallocating nodes as in the adaptive CWT algorithm. In [30] idealized versions of these algorithms are applied to conditional branch prediction with success. However, a realistic hardware implementation has yet to be shown. Another version of the PPM algorithm is described in [28]. That version has been successfully applied to areas such as data compression [27] file prefetching [33] and cache replacement policies [34] Again, the PPM predictor computes probabilities for symbols (as is the case with the CWT and the CWT based PPM algorithm) Its corresponding version for ....

[Article contains additional citation context not shown here]

I.C.K. Chen, J.T. Coffey, and T.N. Mudge. Analysis of Branch Prediction via Data Compression. In Proceedings of the International Conference on Architectural support for programming languages and operating systems, pages 128--137, October 1996.

.... RSE[6] RSE[7] a) Forming the BVIT index 3 3 3 3 3 3 3 Logical mapping Logical mapping Logical mapping Logical mapping Logical mapping Logical mapping Logical mapping Logical ADD Tree Register set tag 3 3 3 3 3 ADD ADD ADD ADD ADD ADD RSE[3] RSE[4] RSE[5] RSE[6] RSE[7] RSE[8] 3 3 3 3 3 RSE[1] 3 Shadow register map table P1 P2 P3 P4 P6 P7 P8 P5 (b) Forming the register set tag Figure 4. Generating values for ARVI 4.5 Forming the DD chain depth key Tight loops can experience identical paths to a branch on successive iterations. We find it is important to ....

....21, 23, 25, 31] only small incremental improvements have been realized with these approaches. There is still a large number of dynamic branches that are mispredicted, e.g. for go. Current branch predictor designs appear to be reaching the limit relative to the type of input information provided [8]. Related approaches that include additional information into the branch prediction process involve correlating the actual branch register values with the branch outcome [14] using a conventional value predictor. The authors of the study acknowledge that the accuracy of value prediction is low. ....

I.-C. K. Chen, J. T. Coffey, and T. N. Mudge. Analysis of Branch Prediction via Data Compression. 7th International Conference on Architectural Support for Programming Languages and Operating Systems, pages 128--37, October 1996.

No context found.

I.-C. K. Chen, J. T. Coffey, and T. N. Mudge. Analysis of branch prediction via data compression. In Proceedings of the 7th Intl. Conference on Architectural Support for Programming Languages and Operating Systems, pages 128-- 137, Oct. 1996.

No context found.

I-Cheng K. Chen, John T. Co#ey, and Trevor N. Mudge. Analysis of Branch Prediction via Data Compression. In ASPLOS VII, pages 128--137, Cambridge, Massachusetts, USA, October 1996.

No context found.

I.-C. K. Chen, J. T. Co#ey, and T. N. Mudge, "Analysis of branch prediction via data compression," in Architectural Support for Programming Languages and Operating Systems (ASPLOS-VII), pp. 128--137, 1996.

No context found.

I.-C. Chen, J.T. Coffey, and T.N. Mudge. Analysis of branch prediction via data compression. In Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, October 1996.

No context found.

I.-C. Chen, J.T. Co#ey, and T.N. Mudge. Analysis of branch prediction via data compression. In Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, October 1996.

No context found.

I.-C. Chen, J. T. Co#ey, and T. N. Mudge. Analysis of branch prediction via data compression. In Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, pages 128-- 137, October 1996.

No context found.

I-Cheng K. Chen, J. T. Coffey, and Trevor N. Mudge. Analysis of Branch Prediction via Data Compression. In Proc. ASPLOS-VII, October 1996.

No context found.

T. N. M. I-Cheng K. Chen, John T. Coffey, "Analysis of branch prediction via data compression," in Proceedings of the Seventh International Conference on Architectural Support for Programming Languages and Operating Systems, pp. 128--13, SIGOPS, ACM, October 1996.

No context found.

Chen, I-Cheng K., Coey, John T., and Mudge, Trevor N. Analysis of Branch Prediction via Data Compression. ASPLOS VII (October 1996), 128-137.

First 50 documents

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