F. Jesús Sánchez

Mitosis compiler: an infrastructure for speculative threading based on pre-computation slices

Speculative parallelization can provide significant sources of additional thread-level parallelism, especially for irregular applications that are hard to parallelize by conventional approaches. In this paper, we present the Mitosis compiler, which partitions applications into speculative threads, with special emphasis on applications for which conventional parallelizing approaches fail.The management of inter-thread data dependences is crucial for the performance of the system. The Mitosis framework uses a pure software approach to predict/compute the threads input values. This software approach is based on the use of pre-computation slices (p-slices), which are built by the Mitosis compiler and added at the beginning of the speculative thread. P-slices must compute thread input values accurately but they do not need to guarantee correctness, since the underlying architecture can detect and recover from misspeculations. This allows the compiler to use aggressive/unsafe optimizations to significantly reduce their overhead. The most important optimizations included in the Mitosis compiler and presented in this paper are branch pruning, memory and register dependence speculation, and early thread squashing.Performance evaluation of Mitosis compiler/architecture shows an average speedup of 2.2.

Modulo scheduling for a fully-distributed clustered VLIW architecture

Clustering is an approach that many microprocessors are adopting in recent times in order to mitigate the increasing penalties of wire delays. We propose a novel clustered VLIW architecture which has all its resources partitioned among clusters, including the cache memory. A modulo scheduling scheme for this architecture is also proposed. This algorithm takes into account both register and memory inter-cluster communications so that the final schedule results in a cluster assignment that favors cluster locality in cache references and register accesses. It has been evaluated for both 2- and 4-cluster configurations and for differing numbers and latencies of inter-cluster buses. The proposed algorithm produces schedules with very low communication requirements and outperforms previous cluster-oriented schedulers.

Graph-partitioning based instruction scheduling for clustered processors

This paper 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 effective 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 SPECfp95 is 23% with respect to the published scheme with the best performance. Besides, the proposed scheme is much faster (between 2-7 times, depending on the configuration).

A Unified Modulo Scheduling and Register Allocation Technique for Clustered Processors

This work presents a modulo scheduling framework for clustered ILP processors that integrates the cluster assignment, instruction scheduling and register allocation steps in a single phase. This unified approach is more effective than traditional approaches based on sequentially performing some (or all) of the three steps, since it allows optimizing the global code generation problem instead of searching for optimal solutions to each individual step. Besides, it avoids the iterative nature of traditional approaches, which require repeated applications of the three steps until a valid solution is found. The proposed framework includes a mechanism to insert spill code on-the-fly and heuristics to evaluate the quality of partial schedules considering simultaneously inter-cluster communications, memory pressure and register pressure. Transformations that allow trading pressure on a type of resource for another resource are also included. We show that the proposed technique outperforms previously proposed techniques. For instance, the average speed-up for the SPECfp95 is 36% for a 4-cluster configuration.

Instruction scheduling for clustered VLIW architectures

Clustered VLIW organizations are nowadays a common trend in the design of embedded/DSP processors. In this work we propose a novel modulo scheduling approach for such architectures. The proposed technique performs the cluster assignment and the instruction scheduling in a single pass, which is more effective than doing first the assignment and latter the scheduling. We also show that loop unrolling significantly enhances the performance of the proposed scheduler, especially when the communication channel among clusters is the main performance bottleneck. By selectively unrolling some loops, we can obtain the best performance with the minimum increase in code size. Performance evaluation for the SPECfp95 shows that the clustered architecture achieves about the same IPC (Instructions Per Cycle) as a unified architecture with the same resources. Moreover, when the cycle time is taken into account, a 4-cluster configuration is 3.6 times faster than the unified architecture.

The effectiveness of loop unrolling for modulo scheduling in clustered VLIW architectures

Clustered organizations are becoming a common trend in the design of VLIW architectures. In this work we propose a novel modulo scheduling approach for such architectures. The proposed technique performs the cluster assignment and the instruction scheduling in a single pass, which is shown to be more effective than doing first the assignment and later the scheduling. We also show that loop unrolling significantly enhances the performance of the proposed scheduler especially when the communication channel among clusters is the main performance bottleneck. By selectively unrolling some loops, we can obtain the best performance with the minimum increase in code size. Performance evaluation for the SPECfp95 shows that the clustered architecture achieves about the same IPC (Instructions Per Cycle) as a unified architecture with the same resources. Moreover when the cycle time is taken into account, a 4-cluster configurations is 3.6 times faster than the unified architecture.

A locality sensitive multi-module cache with explicit management

Cache memories are often inefficiently managed, which results in significant memory penalties. An important reason for this poor performance is the homogeneous management of all memory references, even though different memory references may exhibit a very different locality. In this work, we present a novel data cache architecture composed of different modules, each module exploiting a particular type of locality. The information of which module each fetched data is placed on is passed to the hardware by means of a hint encoded in the memory instructions. This hint is set based on a locality analysis that can be performed by the compiler or using profiling data. The proposed data cache organization exhibits a high-performance for numerical codes, even with low capacity, A SKI3 cache has a miss ratio that is about 0.4 times the miss ratio of an SKI3 direct-mapped cache and it is very close to that of a 64KB fullyassociative cache, when data prefetching is not considered. Furthermore, the locality analysis allows for a selective one-block lookahead prefetch scheme, just for those references that exhibit spatial locality. This prefetch firther reduces the miss ratio to one half of that of a 64KB fully-associative cache, with a negligible increase in memory traffic compared with the non-prefetching scheme, due to the locality-driven selective approach. Although the proposed data cache organization is oriented towards numerical codes, which usually suffer more memory penalties than non-numerical ones, for the latest ones, a very simple profiling analysis results in a performance very close to a conventional cache, in spite of its lower capacity.

Exploiting pseudo-schedules to guide data dependence graph partitioning

This paper presents a new modulo scheduling algorithm for clustered microarchitectures. The main feature of the proposed scheme is that the assignment of instructions to clusters is done by means of graph partitioning algorithms that are guided by a pseudo-scheduler. This pseudo-scheduler is a simplified version of the full instruction scheduler and estimates key constraints that would be encountered in the final schedule. The final scheduling process is bi-directional and includes on-the-fly spill code generation. The proposed scheme is evaluated against previous scheduling approaches using the SPECfp95 benchmark suite. Our modeling results show that better schedules are obtained for most programs across a range of different architectures. For a 4-cluster VLIW architecture with 32 registers and a 2-cycle inter-cluster communication delay we obtain an average speedup of 38.5%.

Effective instruction scheduling techniques for an interleaved cache clustered VLIW processor

Clustering is a common technique to overcome the wire delay problem incurred by the evolution of technology. Fully-distributed architectures, where the register file, the functional units and the data cache are partitioned, are particularly effective to deal with these constraints and besides they are very scalable. In this paper effective instruction scheduling techniques for a clustered VLIW processor with a word-interleaved cache are proposed Such scheduling techniques rely on: (i) loop unrolling and variable alignment to increase the percentage of local accesses, (ii) a latency assignment process to schedule memory operations with an appropriate latency and (iii) different heuristics to assign instructions to clusters. In particular, the number of local accesses is increased by more than 25% if these techniques are used and the ratio of stall time over compute time is small. Next, the main source of remote accesses and stall time is investigated. Stall time is mainly due to remote hits, and Attraction Buffers are used to increase local accesses and reduce stall time. Stall time is reduced by 29% and 34% depending on the scheduling heuristic. IPC results for a word-interleaved cache clustered VLIW processor are similar to those of the multiVLIW (a cache-coherent clustered processor with a more complex hardware design), and are 10% and 5% better (depending on the scheduling heuristic) than the IPC for a clustered processor with a unified cache.

Flexible compiler-managed L0 buffers for clustered VLIW processors

Wire delays are a major concern for current and forthcoming processors. One approach to attack this problem is to divide the processor into semi-independent units referred to as clusters. A cluster usually consists of a local register file and a subset of the functional units, while the data cache remains centralized. However, as technology evolves, the latency of such a centralized cache increase leading to an important performance impact. In this paper, we propose to include flexible low-latency buffers in each cluster in order to reduce the performance impact of higher cache latencies. The reduced number of entries in each buffer permits the design of flexible ways to map data from L1 to these buffers. The proposed L0 buffers are managed by the compiler, which is responsible to decide which memory instructions make us of them. Effective instruction scheduling techniques are proposed to generate code that exploits these buffers. Results for the Mediabench benchmark suite show that the performance of a clustered VLIW processor with a unified L1 data cache is improved by 16% when such buffers are used. In addition, the proposed architecture also shows significant advantages over both MultiVLIW processors and clustered processors with a word-interleaved cache, two state-of-the-art designs with a distributed L1 data cache.