Search

While industry continues to develop SIMD vector ISAs by providing new instructions and wider data-paths, modern SIMD architectures still rely on the programmer or compiler to transform code to vector form only when it is safe. Limitations in the power of a compiler's memory alias analysis and the presence of infrequent memory data dependences mean that whole regions of code cannot be safely vectorised without risking changing the semantics of the application, restricting the available performance.We present a new SIMD architecture to address this issue, which relies on speculation to identify and catch memory- dependence violations that occur during vector execution. Once identified, only those SIMD lanes that have used erroneous data are replayed; other lanes, both older and younger, keep the results of their latest execution. We use the compiler to mark loops with possible cross-iteration dependences and safely vectorise them by executing on our architecture, termed selective-replay vectorisation (SRV). Evaluating on a range of general-purpose and HPC benchmarks gives an average loop speedup of 2.9 ×, and up to 5.3 × in the best case, over already-vectorised code. This leads to a whole-program speedup of up to 1.19 × (average 1.06 ×) over already-vectorised applications.

This article describes the ARM Scalable Vector Extension (SVE). Several goals guided the design of the architecture. First was the need to extend the vector processing capability associated with the ARM AArch64 execution state to better address the computational requirements in domains such as high-performance computing, data analytics, computer vision, and machine learning. Second was the desire to introduce an extension that can scale across multiple implementations, both now and into the future, allowing CPU designers to choose the vector length most suitable for their power, performance, and area targets. Finally, the architecture should avoid imposing a software development cost as the vector length changes and where possible reduce it by improving the reach of compiler auto-vectorization technologies. SVE achieves these goals. It allows implementations to choose a vector register length between 128 and 2,048 bits. It supports a vector-length agnostic programming model that lets code run and scale automatically across all vector lengths without recompilation. Finally, it introduces several innovative features that begin to overcome some of the traditional barriers to autovectorization., 8 pages, 8 figures, IEEE Micro paper

Non-uniform cache architecture (NUCA) aims to limit the wire-delay problem typical of large on-chip last level caches: by partitioning a large cache into several banks, with the latency of each one depending on its physical location and by employing a scalable on-chip network to interconnect the banks with the cache controller, the average access latency can be reduced with respect to a traditional cache. The addition of a migration mechanism to move the most frequently accessed data towards the cache controller (D-NUCA) further improves the average access latency. In this work we propose a last-level cache design, based on the D-NUCA scheme, which is able to significantly limit its static power consumption by dynamically adapting to the needs of the running application: the way adaptable D-NUCA cache. This design leads to a fast and power-efficient memory hierarchy with an average reduction by 31.2% in energy-delay product (EDP) with respect to a traditional D-NUCA. We propose and discuss a methodology for tuning the intrinsic parameters of our design and investigate the adoption of the way adaptable D-NUCA scheme as a shared L2 cache in a chip multiprocessor (CMP) system (24% reduction of EDP).

Non-uniform cache architectures (NUCAs) are a novel design paradigm for large last-level on-chip caches, which have been introduced to deliver low access latencies in wire-delay-dominated environments. Their structure is partitioned into sub-banks and the resulting access latency is a function of the physical position of the requested data. Typically, NUCA caches employ a switched network, made up of links and routers with buffered queues, to connect the different sub-banks and the cache controller, and the characteristics of the network elements may affect the performance of the entire system. This work analyses how different parameters for the network routers, namely cut-through latency and buffering capacity, affect the overall performance of NUCA-based systems for the single processor case, assuming a reference NUCA organisation proposed in literature. The entire analysis is performed utilising a cycle-accurate execution-driven simulator of the entire system and real workloads. The results indicate that the sensitivity of the system to the cut-through latency is very high, thus limiting the effectiveness of the NUCA solution, and that modest buffering capacity is sufficient to achieve a good performance level. As a consequence, in this work we propose an alternative clustered NUCA organisation that limits the average number of hops experienced by cache accesses. This organisation is better performing and scales better as the cut-through latency increases, thus simplifying the implementation of routers, and it is also more effective than another latency reduction solution proposed in literature (hybrid network).

Non uniform cache architectures (NUCA) are a novel design paradigm for large last-level on-chip caches that has been introduced to deliver low access latencies in wire delay dominated environments. Their structure is partitioned into sub-banks and the resulting access latency is a function of the physical position of the requested data. Typically, to connect the different sub-banks and the cache controller, NUCA caches employ a switched network, made up of links and routers with buffered queues; the characteristics of such switched network may affect the performance of the entire system. This work analyzes how different parameters for the routers, namely cut-through latency and buffering capacity, affect the overall performance of NUCA based systems for the single processor case, assuming a reference organization proposed in literature. The results indicate that the sensitivity of the system to the cut-through latency is very high and that limited buffering capacity is sufficient to achieve a good performance level. As a consequence, we propose an alternative NUCA organization that limits the average number of hops experienced by cache accesses. This organization is better performing in most of the cases and scales better as the cut-through latency increases, thus simplifying the implementation of routers.

D-NUCA caches are cache memories that, thanks to banked organization, broadcast search and promotion/demotion mechanism, are able to tolerate the increasing wire delay effects introduced by technology scaling. As a consequence, they will outperform conventional caches (UCA, Uniform CacheArchitectures) in future generation cores. Due to the promotion/ demotion mechanism, we observed that the distribution of hits across the ways of a D-NUCA cache varies across applications as well as across different execution phases within a single application. In this work, we show how such a behavior can be leveraged to improve the D-NUCA power efficiency as well as to decrease its access latency.In particular, we propose: 1) A new micro-architectural technique to reduce the static power consumption of a D-NUCA cache by dynamically adapting the number of active (i.e. powered-on) ways to the need of the running application; our evaluation shows that a strong reduction of the average number of active ways (37.1%) is achievable, without significantly affecting the IPC (-2.83%), leading to a resultant reduction of the Energy Delay Product (EDP) of 30.9%. 2) A strategy to estimate the characteristic parameters of the proposed technique. 3) An evaluation of the effectiveness of the proposed technique in the multicore environment.

NUCA caches are large L2 on-chip cache memories characterized by multi-bank partitioning and designed to hide wire delay effects. They exhibit high hit rates while keeping access latency low. Proposed designs for such caches are Static NUCA, in which data are statically allocated to the cache banks, and Dynamic NUCA, in which data may reside in different banks, and a migration mechanism is introduced to better tolerate wire delay effects. The two architectures permit to achieve different performances by acting on architectural parameters and data management policies, at the cost of different balances between static and dynamic power consumption and energy dissipation. In this work, we propose preliminary results of the characterization of such balances, by presenting an evaluation of performance and energy consumption of conventional UCAs, and Static and Dynamic NUCA caches. All the considered caches architectures are equal sized and they are supposed to be used in an aggressive high frequency system running some applications from the SPEC CPU2000 and the NAS Parallel Benchmarks suites. The experimental results obtained indicate that, although the migration of data contributes to increase the dynamic energy consumption in Dynamic NUCA caches, the higher IPC achieved permits to save static energy, which dominates the power/energy balance in all the considered architectures. As a consequence, such results would designate NUCA caches as the most performing and energy saving architectures. Besides, according to the obtained results, future power improvements for NUCA caches should concentrate on static energy, while, for the dynamic energy, the on-chip network is the most critical element. Migration of data is acceptable, since it has a positive impact on performance, and the increased dynamic energy is overwhelmed by the static energy savings resulting from the shorter execution time. In order to give a general validity to such statements, we need to explore more design space points for each architecture (by varying the running clock rate and other design parameters) and to evaluate them considering a larger set of benchmarks.