Differential FCM: increasing value prediction accuracy by improving table usage efficiency

Abstract

Value prediction is a relatively new technique to increase the Instruction Level Parallelism (ILP) in future microprocessors. An important problem when designing a value predictor is efficiency: an accurate predictor requires huge prediction tables. This is especially, the case for the finite context method (FCM) predictor the most accurate one. In this paper we show that the prediction accuracy of the FCM can be greatly improved by making the FCM predict strides instead of values. This new predictor is called the differential finite context method (DFCM) predictor The DFCM predictor outperforms a similar FCM predictor by as much as 33%, depending on the prediction table size. If we take the additional storage into account, the difference is still 15% for realistic predictor sizes. We use several metrics to show that the key to this success is reduced aliasing in the level-a table. We also show that the DFCM is superior to hybrid predictors based on FCM and stride predictors, since its prediction accuracy is higher than that of a hybrid one using a perfect meta-predictor

Lossless Compression of Hexahedral Meshes

“…As pointed out in [13], in constant-stride subsequences such as nested loops the single difference in stride due to loop restart usually causes two mispredictions: once when the counter first wraps, and once when the constant stride re-emerges since it has then been replaced by the wraparound stride in the hash table. Similar wraparounds frequently occur in hexahedral meshes, e.g.…”

“…Each row holds the vertices of an element that span the three spatial dimensions. Though the example in Table 1 is overly simplistic, the runs of constant strides v A more general stride-based approach is the differential finite context method (DFCM) [13], which has been used successfully for trace file and floating-point compression [7,12]. The basic DFCM predictor is a hash table that maps a set of recent strides to the current, predicted stride.…”

Leakage Energy Reduction in Value Predictors through Static Decay

“…The differential finite context method value predictor (DFCM), introduced by Goeman et al [5], joins the two previous predictors. DFCM works like FCM (two-level prediction tables), but it stores the differences between the values instead of the values themselves, plus the last value of the instruction.…”

“…Value Prediction (VP) has also been proposed [4] [5][8] [13] as an effective way of improving superscalar processor performance by overcoming data dependences. However, the use of VP structures, despite the speedup provided in superscalar processors (average speedup of 15% as reported in [2]), has not been widely spread, mainly due to complexity-delay issues.…”

Targeted Data Prefetching

“…In order to contain the amount of resources needed to realize the prefetch mechanism, we chose a prediction based scheme. In particular, we chose to use a hybrid stride and differential finite context method predictor [13]. This predictor is shown in Fig.…”