Training large-scale neural networks is heavily constrained by GPU memory. In order to circumvent this limitation, gradient checkpointing, or recomputation is a powerful technique. There is active research in this area with methods such as Checkmake or Moccasin. However, both Checkmate and Moccasin rely on mixed integer linear programming or constraint programming, resulting in limited scalability due to their exponentially large search space. This paper proposes a novel algorithm for recomputation (FastSA) based on a simulated annealing heuristic that achieves comparable or even better solutions than state-of-the-art alternatives. FastSA can optimize computational graphs with thousands of nodes within 3 to 30 seconds, several orders of magnitude faster than current solutions. We applied FastSA to PyTorch models and verified its effectiveness through popular large vision and text models, including recent language models with the transformer architecture. The results demonstrate significant memory reductions by 73% with extra 18% computational overheads on average. Our experiments demonstrate the practicality and effectiveness of our recomputation algorithm, further highlighting its potential for wide application in various deep learning domains.

Asynchronous task-based programming models are gaining popularity to address the programmability and performance challenges in high performance computing. One of the main attractions of these models and runtimes is their potential to automatically expose and exploit overlap of computation with communication. However, we find that inefficient interactions between these programming models and the underlying messaging layer (in most cases, MPI) limit the achievable computation-communication overlap and negatively impact the performance of parallel programs. We address this challenge by exposing and exploiting information about MPI internals in a task-based runtime system to make better task-creation and scheduling decisions. In particular, we present two mechanisms for exchanging information between MPI and a task-based runtime, and analyze their trade-offs. Further, we present a detailed evaluation of the proposed mechanisms implemented in MPI and a task-based runtime. We show performance improvements of up to 16.3% and 34.5% for proxy applications with point-to-point and collective communication, respectively.

Asynchronous task-based programming models are gaining popularity to address programmability and performance challenges in high performance computing. One of the main attractions of these models and runtimes is their potential to automatically expose and exploit overlap of computation with communication. However, inefficient interactions between such programming models and the underlying messaging layer (in most cases, MPI) limit the achievable computation-communication overlap and negatively impact the performance of parallel programs. We propose to expose information about MPI internals to a task-based runtime system to make better scheduling decisions. In particular, we show how existing mechanisms used to profile MPI implementations can be used to share information between MPI and a task-based runtime. Further, an evaluation of the proposed method shows performance improvements of up to 30.7% for applications with collective communication.

The growing complexity of multi-core architectures has motivated a wide range of software mechanisms to improve the orchestration of parallel executions. Task parallelism has become a very attractive approach thanks to its programmability, portability and potential for optimizations. However, with the expected increase in core counts, finer-grained tasking will be required to exploit the available parallelism, which will increase the overheads introduced by the runtime system. This work presents Task Dependence Manager (TDM), a hardware/software co-designed mechanism to mitigate runtime system overheads. TDM introduces a hardware unit, denoted Dependence Management Unit (DMU), and minimal ISA extensions that allow the runtime system to offload costly dependence tracking operations to the DMU and to still perform task scheduling in software. With lower hardware cost, TDM outperforms hardware-based solutions and enhances the flexibility, adaptability and composability of the system. Results show that TDM improves performance by 12.3% and reduces EDP by 20.4% on average with respect to a software runtime system. Compared to a runtime system fully implemented in hardware, TDM achieves an average speedup of 4.2% with 7.3x less area requirements and significant EDP reductions. In addition, five different software schedulers are evaluated with TDM, illustrating its flexibility and performance gains.

Managing criticality in task-based programming models opens a wide range of performance and power optimization opportunities in future manycore systems. Criticality aware task schedulers can benefit from these opportunities by scheduling tasks to the most appropriate cores. However, these schedulers may suffer from priority inversion and static binding problems that limit their expected improvements. Based on the observation that task criticality information can be exploited to drive hardware reconfigurations, we propose a Criticality Aware Task Acceleration (CATA) mechanism that dynamically adapts the computational power of a task depending on its criticality. As a result, CATA achieves significant improvements over a baseline static scheduler, reaching average improvements up to 18.4% in execution time and 30.1% in Energy-Delay Product (EDP) on a simulated 32-core system. The cost of reconfiguring hardware by means of a software-only solution rises with the number of cores due to lock contention and reconfiguration overhead. Therefore, novel architectural support is proposed to eliminate these overheads on future manycore systems. This architectural support minimally extends hardware structures already present in current processors, which allows further improvements in performance with negligible overhead. As a consequence, average improvements of up to 20.4% in execution time and 34.0% in EDP are obtained, outperforming state-of-the-art acceleration proposals not aware of task criticality.

In the late years many different interconnection networks have been used with two main tendencies. One is characterized by the use of high-degree routers with long wires while the other uses routers of much smaller degree. The latter rely on two-dimensional mesh and torus topologies with shorter local links. This paper focuses on doubling the degree of common 2D meshes and tori while still preserving an attractive layout for VLSI design. By adding a set of diagonal links in one direction, diagonal networks are obtained. By adding a second set of links, networks of degree eight are built, named king networks. This research presents a comprehensive study of these networks which includes a topological analysis, the proposal of appropriate routing procedures and an empirical evaluation. King networks exhibit a number of attractive characteristics which translate to reduced execution times of parallel applications. For example, the execution times NPB suite are reduced up to a 30 percent. In addition, this work reveals other properties of king networks such as perfect partitioning that deserves further attention for its convenient exploitation in forthcoming high-performance parallel systems.

The increasing number of cores and the anticipated level of heterogeneity in upcoming multicore architectures cause important problems in traditional cache hierarchies. A good way to alleviate these problems is to add scratchpad memories alongside the cache hierarchy, forming a hybrid memory hierarchy. This memory organization has the potential to improve performance and to reduce the power consumption and the on-chip network traffic, but exposing such a complex memory model to the programmer has a very negative impact on the programmability of the architecture. Emerging task-based programming models are a promising alternative to program heterogeneous multicore architectures. In these models the runtime system manages the execution of the tasks on the architecture, allowing them to apply many optimizations in a generic way at the runtime system level. This paper proposes giving the runtime system the responsibility to manage the scratchpad memories of a hybrid memory hierarchy in multicore processors, transparently to the programmer. In the envisioned system, the runtime system takes advantage of the information found in the task dependences to map the inputs and outputs of a task to the scratchpad memory of the core that is going to execute it. In addition, the paper exploits two mechanisms to overlap the data transfers with computation and a locality-aware scheduler to reduce the data motion. In a 32-core multicore architecture, the hybrid memory hierarchy outperforms cache-only hierarchies by up to 16%, reduces on-chip network traffic by up to 31% and saves up to 22% of the consumed power.

Many current VLSI on-chip multiprocessors and systems-on-chip employ point-to-point switched interconnection networks. Rings and 2D-meshes are among the most popular interconnection topologies for these increasingly important onchip networks. Nevertheless, rings cannot scale beyond dozens of nodes and meshes are asymmetric. Two of the key features of square 2D-tori are their scalability and symmetry. As higher scalability is demanded by the increasing number of cores (or specialized units) integrated on a chip and symmetry is critical for high-performance and load balancing, we concentrate on 2D-tori. However, most popular deadlock-free routing mechanisms are based on Dimension Order Routing (DOR) which breaks the torus symmetry when managing adversarial traffic patterns. This paper analyzes this problem and its consequences. After that, it proposes a new deadlock-free fully adaptive minimal routing, denoted as σDOR, that preserves tori symmetry under any load. It uses just two virtual channels to avoid DOR-induced asymmetry, the same as in previous competitive proposals. σDOR exhibits better behavior than any of previous solutions as it allows packets to dynamically adapt to local congestion. Experimental results evidence the superior performance of our mechanism, confirming the negative impact of DOR asymmetry.

The RSA algorithm is the most used public key cipher algorithm. Since it was presented in 1977, designers have proposed several architectures and implementations to improve its performance using different techniques and devices. In all cases, Montgomery's modular multiplication algorithm is the best option when trying to design an efficient RSA. In this paper, we present a high-radix Montgomery's algorithm implementation. The main advantages of this approach are the scalability and modularity, which allow finding the best tradeoff between area and performance depending on the application. The system is able to process different RSA key lengths without needing to reconfigure it.

The use of computational systems to help making the right investment decisions in financial markets is an open research field where multiple efforts have being carried out during the last few years. The ability of improving the assessment process and being faster than the rest of the players is one of the keys for the success on this competitive scenario. This paper explores different options to accelerate the computation of the option pricing problem (supercomputer, FPGA cluster or GPU) using the Montecarlo method to solve the Black-Scholes formula, and presents a quantitative study of their performance and scalability.

The SMILE project attempts to build efficient lowcost clusters based on FPGA boards using their reconfigurability capabilities. A real parallel application of Content-Based Information Retrieval over the SMILE cluster is presented. Using this application the SMILE cluster’s performance is evaluated and compared in terms of time and power consumption with traditional cluster architecture.

The SMILE project accelerates scientific and industrial applications by means of a cluster of low-cost FPGA boards. With this approach the intensive calculation tasks are accelerated using the FPGA logic, while the communication patterns of the applications remains unchanged by using a Message Passing Library over Linux. This paper explains the cluster architecture: the SMILE nodes and the developed high-speed communication network for the FPGA RocketIO interfaces. A SystemC model developed to simulate the cluster is also detailed. In order to show the potential of the SMILE proposal a Content-Based Information Retrieval parallel application has been developed and compared with a HP cluster architecture in terms of response time andpower consumption.