Simulation is the main tool for computer architects and parallel application developers for developing new architectures and parallel algorithms on many-core machines. Simulating a many-core architecture represent a challenge to software simulators even with parallelization of these SW on multi-cores. Field Programmable Gate Arrays offer an excellent implementation platform due to inherent parallelism. Existing FPGA-based simulators however, are mostly execution-driven which consumes too many FPGA resources. Hence, they still trade-off accuracy with simulation speed as SW simulators do. In this work, an application-level trace-driven FPGA-based many-core simulator is presented. A parameterized Verilog template was developed that can generate any number of simulator tiles. The input trace has an architecturally agnostic format that is directly interpreted by the FPGA-based timing model to re-construct the execution events of the original application with accurate timing. This allows fitting a large number of simulation tiles on a single FPGA without sacrificing simulation speed or accuracy. Experimental results show that the simulator's average accuracy is ~14 percent with simulation speeds ranging from 100's of MIPs to over 2,200 MIPS for a 16-core target architecture. Hence, with accuracy similar to SW simulators, its speed is higher than all other FPGA-based simulators.
1 January 2014 in Canadian Journal of Electrical and Computer Engineering
This paper considers blocking and scheduling for the design and implementation of field-programmable gate array (FPGA)-based floating-point parallel matrix multiplication in the presence of a memory hierarchy. For high performance, on-chip memory holds data that are reused when the computation is divided into blocks, and multiple arithmetic units perform independent operations within each block in parallel. The first contribution of this paper is a detailed analysis of the design space to characterize performance based on the amount of on-chip memory used and the approaches considered for blocking and scheduling of the computation. A comparison is also made to prior work with a unified view. The second contribution is a flexible high-performance implementation for the Altera Stratix IV EP4SGX530C2 FPGA with an interface to external double-data-rate synchronous dynamic RAM (DDR2 SDRAM) memory. Various configuration options support optimization of different objectives, and the resulting configurations have been verified in simulation and in hardware. For double-precision floating-point, a performance of 16 giga-floating-point operations per second (GFLOPS) is achievable with 64 arithmetic units at 160 MHz.
IEEE 802.15.4 is a low‐power, low‐rate MAC/PHY standard that meets most of the stringent requirements of singlehop wireless sensor networks. Sensor networks with nodal populations composed of thousands of devices have been envisioned in conjunction with environmental, vehicular, military applications, and many others. However, such large sensor network deployments necessitate multihop support as well as low power consumption. In the light of the standard′s extremely limited joint support of the two aforementioned attributes, this paper presents two essential contributions. First, a framework is proposed to implement a new IEEE 802.15.4 operating mode, namely, the synchronized peer-to-peer mode. This mode is designed to enable the standard′s low‐power features in peer‐to‐peer multihop‐ready topologies. The second contribution is a distributed GTS (dGTS) management scheme designed to function in the newly devised network mode. This protocol provides reliable contention‐free access in peer‐to‐peer topologies in a completely distributed manner. Assuming optimal routing, our simulation experiments reveal perfect delivery ratios as long as the traffic load does not reach or surpass its saturation threshold. dGTS sustains at least twice the delivery ratio of contention‐based access under suboptimal dynamic routing. Moreover, the dGTS scheme exhibits minimum power consumption by eliminating the retransmissions attributed to contention, which, in turn, reduces the number of transmissions to a minimum.
FPGA technology constitutes an attractive platform for high-performance accelerators of parallel workloads in general-purpose computers. Matrix multiplication is a computationally intensive application that is highly parallelizable. Previous work has typically described custom floating-point components and reported on specific designs or implementations using these components for FPGA-based matrix multiplication. We seek to utilize vendor-supplied or other available floating-point components to explore the system-architecture design space for flexible, high-performance, FPGA-based accelerators. In this paper, we focus on the design of control logic that accommodates the configuration of as many implementation aspects as possible (e.g., scheduling of operations, levels of parallelism, and choice of arithmetic operators) for inclusion in an experimental infrastructure to assess the effects of these parameters on overall system performance.
This paper describes performance optimizations of a transfer controller for an FPGA-based blocked parallel matrix multiplication accelerator. One of the key challenges of the controller is the generation of a sequence of host memory addresses to transfer blocks of matrices between host and on-chip memories. These addresses are not contiguous, thereby introducing complexity for the controller design. This paper first outlines the intended system architecture for this controller. Next, detailed controller specifications are presented for generating host memory addresses. Various pipeline configurations that yield incremental performance improvements are then described. Finally, experimental results are presented, with the best configuration having an operating frequency exceeding 470 MHz on an Altera Stratix III chip. This level of performance is comparable to that of the pipelined floating-point arithmetic units in the complete system.