HARP Cluster

<jats:p>Heterogeneous platforms with FPGAs have started to be employed in the High-Performance Computing (HPC) field to improve performance and overall efficiency. These platforms allow the use of specialized hardware to accelerate software applications, but require the software to be adapted in what can be a prolonged and complex process. The main goal of this work is to describe and evaluate mechanisms that can transparently transfer the control flow between CPU and FPGA within the scope of HPC. Combining such a mechanism with transparent software profiling and accelerator configuration could lead to an automatic way of accelerating regular applications. In this work, a mechanism based on the ptrace system call is proposed, and its performance on the Intel Xeon+FPGA platform is evaluated. The feasibility of the proposed approach is demonstrated by a working prototype that performs the transparent control flow transfer of any function call to a matching hardware accelerator. This approach is more general than shared library interposition at the cost of a small time overhead in each accelerator use (about 1.3ms in the prototype implementation).</jats:p>

<jats:p>In this article, we focus on the acceleration of a chemical reaction simulation that relies on a system of stiff ordinary differential equation (ODEs) targeting heterogeneous computing systems with CPUs and field-programmable gate arrays (FPGAs). Specifically, we target an essential kernel of the coupled chemistry aerosol-tracer transport model to the Brazilian developments on the regional atmospheric modeling system (CCATT-BRAMS). We focus on a linear solve step using the QR factorization based on the modified Gram-Schmidt method as the basis of the ODE solver in this application. We target Intel hardware accelerator research program (HARP) architecture with the OpenCL programming environment for these early experiments. Our design exploration reveals a hardware design that is up to 4 times faster than the original iterative Jacobi method used in this solver. Still, even with hardware support, the overall performance of our QR-based hardware is lower than its original software version.</jats:p>

<jats:p> Field-Programmable Gate Arrays (FPGAs) are widely used in the central signal processing design of the Square Kilometer Array (SKA) as hardware accelerators. The frequency domain acceleration search (FDAS) module is an important part of the SKA1-MID pulsar search engine. To develop for a yet to be finalized hardware, for cross-discipline interoperability and to achieve fast prototyping, OpenCL as a high-level FPGA synthesis approaches employed to create the sub-modules of FDAS. The FT convolution and the harmonic-summing plus some other minor sub-modules are elements in the FDAS module that have been well-optimized separately before. In this paper, we explore the design space of combining well-optimized designs, dealing with the ensuing need to trade-off and compromise. Pipeline computing is employed to handle multiple input arrays at high speed. The hardware target is to employ multiple high-end FPGAs to process the combined FDAS module. The results show interesting consequences, where the best individual solutions are not necessarily the best solutions for the speed of a pipeline where FPGA resources and memory bandwidth need to be shared. By proposing multiple buffering techniques to the pipeline, the combined FDAS module can achieve up to 2[Formula: see text] speedup over implementations without pipeline computing. We perform an extensive experimental evaluation on multiple high-end FPGA cards hosted in a workstation and compare to a technology comparable mid-range GPU. </jats:p>

<jats:p>Intel recently introduced the Heterogeneous Architecture Research Platform, HARP. In this platform, the Central Processing Unit and a Field-Programmable Gate Array are connected through a high-bandwidth, low-latency interconnect and both share DRAM memory. For this platform, Open Computing Language (OpenCL), a High-Level Synthesis (HLS) language, is made available. By making use of HLS, a faster design cycle can be achieved compared to programming in a traditional hardware description language. This, however, comes at the cost of having less control over the hardware implementation. We will investigate how OpenCL can be applied to implement a real-time guided image filter on the HARP platform. In the first phase, the performance-critical parameters of the OpenCL programming model are defined using several specialized benchmarks. In a second phase, the guided image filter algorithm is implemented using the insights gained in the first phase. Both a floating-point and a fixed-point implementation were developed for this algorithm, based on a sliding window implementation. This resulted in a maximum floating-point performance of 135 GFLOPS, a maximum fixed-point performance of 430 GOPS and a throughput of HD color images at 74 frames per second.</jats:p>

Heterogeneous computing that exploits simultaneous co-processing with different device types has been shown to be effective at both increasing performance and reducing energy consumption. In this paper, we extend a scheduling framework encapsulated in a high-level C++ template and previously developed for heterogeneous chips comprising CPU and GPU cores, to new high-performance platforms for the data center, which include a cache coherent FPGA fabric and many-core CPU resources. Our goal is to evaluate the suitability of our framework with these new FPGA-based platforms, identifying performance benefits and limitations.We target the state-of-the-art HARP processor that includes 14 high-end Xeon classes tightly coupled to a FPGA device located in the same package. We select eight benchmarks from the high-performance computing domain that have been ported and optimized for this heterogeneous platform. The results show that a dynamic and adaptive scheduler that exploits simultaneous processing among the devices can improve performance up to a factor of 8 × compared to the best alternative solutions that only use the CPU cores or the FPGA fabric. Moreover, our proposal achieves up to 15% and 37% of improvement compared to the best heterogeneous solutions found with a dynamic and static schedulers, respectively.

The advent of FPGA-based hybrid architecture offers the opportunity of customizing memory subsystems to enhance the overall system performance. However, it is not straightforward to design efficient FPGA circuits for emerging FPGAs applications such as in-memory database and graph analytics, which heavily depend on concurrent data structures (CDS'). Highly dynamic behaviors of CDS' have to be orchestrated by synchronization primitives for correct execution. These primitives induce overwhelming memory traffic for synchronizations on FPGAs. This paper proposes a novel method for systematically exploring and exploiting memory-level parallelism (MLP) of CDS by transactional execution on FPGAs. Inspired by the idea that semantics of transactions can be implemented in a more efficient and scalable manner on FPGAs than on CPUs, we propose a transaction-based reconfigurable runtime system for capturing MLP of CDS'. Experiments on linked-list and skip-list show our approach achieves 5.18x and 1.55x throughput improvement on average than lock-based FPGA implementations and optimized CDS algorithms on a state-of-the-art multi-core CPU respectively.