Graph-partitioning based instruction scheduling for clustered processors

Abstract

This work presents a novel scheme to schedule loops for clustered microarchitectures. The scheme is based on a preliminary cluster assignment phase implemented through graph partitioning techniques followed by a scheduling phase that integrates register allocation and spill code generation. The graph partitioning scheme is shown to be very eflective due to its global view of the whole code while the partition is generated. Results show a significant speedup when compared with previously proposed techniques. For some processor configuration the average speedup for the SPECfi95 is 23% with respect to the published scheme with the best pelfonnance. Besides, the proposed scheme is much faster (between 2-7 times, depending on the configuration).

AGAMOS: A Graph-Based Approach to Modulo Scheduling for Clustered Microarchitectures

“…The technique proposed in this paper combines the benefits of global and local information to overcome the limitations of previous approaches [2]. The main feature of the technique is that the distribution of the instructions among the clusters is performed using a global view of the whole DDG, and at the same time, takes into account the main implications that the partition will have on the scheduling step.…”

Mapping DAG-based applications to multiclusters with background workload

“…Most DAG mapping algorithms can be classified into two categories: list scheduling [5] [7] and graph partitioning [1][4] [16].…”

“…A case study is described to illustrate the misperception caused by regarding a cluster as a single processing node. Suppose three tasks v 1 , v 2 , v 3 are schedulable. Their computational volume is 6, 3 and 6 respectively.…”

“…Each sub-graph is then mapped to a resource for execution. Graph partitioning policies have also been presented in a number of research papers [1][4] [16]. However, these graph partitioning algorithms also assume that the sub-graph can be mapped to a specified resource, and that the tasks in the sub-graph will be run in sequence, which is not necessary the case for the scheduling scenariosconsidered in this paper.…”

Instruction replication for clustered microarchitectures

“…First, the graph is coarsened, that is, a new graph with fewer nodes is built by grouping pairs of nodes of the initial graph into new macro-nodes. To choose the nodes that will be grouped in the new macro-node, we first weight the edges of the graph according to the impact that adding a bus latency to that edge would have on execution time [1]. Next, a maximum weight matching is identified.…”