SAT-based compilation to a non-von neumann processor

Coarse Grained Reconfigurable Arrays (CGRAs) are a promising class of architectures for accelerating applications using a large number of parallel execution units for high throughput. While they are typically good at utilizing many processing elements for a single task with automatic parallelization, all processing elements are required to perform their operations in lock step; this makes applications that involve multiple data streams, multiple tasks, or unpredictable schedules more difficult to program and use their resources inefficiently. Other architectures like Massively Parallel Processor Arrays (MPPAs) are better suited for these applications and excel at executing unrelated tasks simultaneously, but the amount of resources easily utilized for a single task is limited. We are developing a new architecture with the multi-task flexibility of an MPPA and the automatic parallelization of a CGRA. A key to the flexibility of MPPAs is the ability for subtasks to execute independently instead of in lock step with all other subtasks on the array. In this paper, we develop the special network and control circuitry to add support for this execution style in a CGRA with less than 2% area overhead. Additionally, we also describe the CAD tool modifications and application developer guidelines for utilizing the resulting hybrid CGRA/MPPA architecture.

Hardware Trojans (HTs) are malicious hardware components designed to leak confidential information or cause the chip/circuit on which they are integrated to malfunction during operation. When we deploy such hardware platforms for edge computing, FPGA-based implementations of Coarse-Grained Reconfigurable Array (CGRA) are also currently falling victim to HT insertion. However, for CGRA, an evolvable hardware (EHW) platform, which has the ability to dynamically change its configuration and behavioral characteristics based on inputs from the environment, provides us with a new way to mitigate HT attacks. In this regard, we investigate the feasibility of using EHW to mitigate HTs that disrupt normal functionality in CGRA in this paper. When it is determined that HT is inserted into certain processing elements (PEs), the array autonomously reconfigures the circuit structure based on an evolutionary algorithm (EA) to avoid the use of HT-infected (HT-I) PEs. We show that the proposed approach is applicable to: (1) hardware platforms that support coarse-grained reconfiguration; and (2) pure combinatorial circuits. In a simulation environment built in Python, this paper reports experimental results for two target evolutionary circuits and outlines the effectiveness of the proposed method.

Coarse-grain reconfigurable arrays (CGRAs) are emerging accelerators that promise low-power acceleration of compute-intensive loops in applications. The acceleration achieved by CGRA relies on the efficient mapping of the compute-intensive loops by the CGRA compiler, onto the CGRA architecture. The CGRA mapping problem, being NP-complete, is performed in a two-step process, namely, scheduling and mapping. The scheduling algorithm allocates timeslots to the nodes of the data flow graph, and the mapping algorithm maps the scheduled nodes onto the processing elements of the CGRA. On a mapping failure, the initiation interval (II) is increased and a new schedule is obtained for the increased II. Most previous mapping techniques use the iterative modulo scheduling (IMS) algorithm to find a schedule for a given II. Since IMS generates a resource-constrained as-soon-as-possible (ASAP) scheduling, even with increased II, it tends to generate a similar schedule that is not mappable. Therefore, IMS does not explore the schedule space effectively. To address these issues, this article proposes CRIMSON, compute-intensive loop acceleration by randomized IMS and optimized mapping technique that generates random modulo schedules by exploring the schedule space, thereby creating different modulo schedules at a given and increased II. CRIMSON also employs a novel conservative test after scheduling to prune valid schedules that are not mappable. From our study conducted on the top 24 performance-critical loops (run for more than 7% of application time) from MiBench, Rodinia, and Parboil, we found that previous state-of-the-art approaches that use IMS, such as RAMP and GraphMinor could not map five and seven loops, respectively, on a 4×4 CGRA, whereas CRIMSON was able to map them all. For loops mapped by the previous approaches, CRIMSON achieved a comparable II.

This paper presents an evaluation of a Heterogeneous Multicore Architecture (HMA) by implementing Orthogonal Frequency-Division Multiplexing (OFDM) receiver blocks as designs for the test of functionality. OFDM receiver consists of computationally intensive and general-purpose processing tasks that can provide maximum coverage to test and evaluate a massively-parallel as well as a general-purpose platform like the HMA. The blocks of the receiver are primarily designed by crafting template-based Coarse-Grained Reconfigurable Array (CGRA) devices and then arranging them in a sequence over a Network-on-Chip (NoC) structure along with a few RISC cores for complete OFDM processing. The OFDM blocks such as Fast Fourier Transform (FFT) and Time Synchronization are computationally intensive and require parallel processing. The OFDM receiver also contains tasks such as frequency offset estimation which require the processing of Taylor series and CORDIC algorithms that are serial in nature. Such a combination of serial and parallel algorithms can perform a thorough exploration and evaluation of almost all the design features of an HMA. The OFDM implementation has led to scale CGRAs to different dimensions, instantiate Processing Elements (PEs) as multiple arithmetic resources and to establish almost all possible ways of PE interconnections. It further explores time-multiplexed patterns for data placement in the CGRA memories. Nevertheless, the data can also be exchanged among different nodes over NoC structure simultaneously and independently by using direct memory access devices. In this experimental work, the performance of each CGRA, the collective performance of the whole platform and the NoC traffic are recorded in terms of the number of clock cycles and several high-level performance metrics. Today's HMAs are generally over or under resourced for the applications that they are designed for and thus not an optimal choice for the end user. Apart from the interesting comparisons to the other state-of-the-art, our experimental setup has provided important insight and guidelines that the designers can use to implement near-optimal solutions for their target applications.

In recent past, we developed 4×8 and 4×16 processing element (PE) template-based Coarse-Grain Reconfigurable Arrays (CGRAs) and mapped different length and type of Fast Fourier Transform (FFT) algorithms on them. In this paper, we have considered radix-4 and radix-(2, 4) FFT accelerators which were generated from 4 × 8 and 4 × 16 PE CGRA templates respectively. We estimated their power and energy consumption while radix-4 accelerator was processing 64 and 1024 points and radix-(2, 4) accelerator was processing 64 and 128 points of FFT algorithms. The power consumption was estimated by timing simulation of postfit gate-level netlist of both of the accelerators for a Field Programmable Gate Array (FPGA) used as target platform. Based on the measurements, we have compared both of the accelerators for their power and energy consumption.

Ultra-low-power (ULP) devices are becoming pervasive, enabling many emerging sensing applications. Energy-efficiency is paramount in these applications, as efficiency determines device lifetime in battery-powered deployments and performance in energy-harvesting deployments. Unfortunately, existing designs fall short because ASICs' upfront costs are too high and prior ULP architectures are too inefficient or inflexible.We present Snafu, the first framework to flexibly generate ULP coarse-grain reconfigurable arrays (CGRAs). Snafu provides a standard interface for processing elements (PE), making it easy to integrate new types of PEs for new applications. Unlike prior high-performance, high-power CGRAs, Snafu is designed from the ground up to minimize energy consumption while maximizing flexibility. Snafu saves energy by configuring PEs and routers for a single operation to minimize switching activity; by minimizing buffering within the fabric; by implementing a statically routed, bufferless, multi-hop network; and by executing operations in-order to avoid expensive tag-token matching.We further present Snafu-Arch, a complete ULP system that integrates an instantiation of the Snafu fabric alongside a scalar RISC-V core and memory. We implement Snafu in RTL and evaluate it on an industrial sub-28 nm FinFET process across a suite of common sensing benchmarks. Snafu-Arch operates at <1 mW, orders-of-magnitude less power than most prior CGRAs. Snafu-Arch uses 41% less energy and runs 4.4× faster than the prior state-of-the-art general-purpose ULP architecture. Moreover, we conduct three comprehensive case-studies to quantify the cost of programmability in Snafu. We find that Snafu-Arch is close to ASIC designs built in the same technology, using just 2.6× more energy on average.

Scaling the memory hierarchy is a major challenge when we scale the number of cores in a multicore processor. Software Managed Multicore (SMM) architectures come up as one of the promising solutions. In an SMM architecture, there are no caches, and each core has only a local scratchpad memory [Banakar et al. 2002]. As the local memory usually is small, large applications cannot be directly executed on it. Code and data of the task mapped to each core need to be managed between global memory and local memory. This article solves the problem of efficiently managing code on an SMM architecture. The primary requirement of generating efficient code assignments is a correct management cost model. In this article, we address this problem by proposing a cost calculation graph. In addition, we develop two heuristics CMSM (Code Mapping for Software Managed multicores) and CMSM_advanced that result in efficient code management execution on the local scratchpad memory. Experimental results collected after executing applications from the MiBench suite [Guthaus et al. 2001] demonstrate that merely by adopting the correct management cost calculation, even using previous code assignment schemes, we can improve performance by an average of 12%. Combining the correct management cost model and a more optimized code mapping algorithm together, our heuristics can reduce runtime in more than 80% of the cases, and by up to 20% on our set of benchmarks, compared to the state-of-the-art code assignment approach [Jung et al. 2010]. When compared with Instruction-level Parallelism (ILP) results, CMSM_advanced performs an average of 5% worse. We also simulate the benchmarks on a cache-based system, and find that the code management overhead on SMM core with our code management is much less than memory latency of a cache-based system.

Coarse-grained reconfigurable arrays (CGRAs) are programmable logic devices with ALU-style processing elements and datapath interconnect. CGRAs can be realized as custom ASICs or implemented on FPGAs as overlays . A key element of CGRAs is that they are typically software programmable with rapid compile times – an advantage arising from their coarse-grained characteristics, simplifying CAD mapping tasks. We implement two previously published CGRAs as overlays on two commercial FPGAs (Intel and Xilinx), and consider the impact of the underlying FPGA architecture on the CGRA area and performance. We present optimizations for the overlays to take advantage of the FPGA architectural features and show a peak performance improvement of 1.93x, as well as maximum area savings of 31.1% and 48.5% for Intel and Xilinx, respectively, relative to a naive first-cut implementation. We also present a novel technique for a configurable multiplexer implementation, which embeds the select signals into SRAM configuration, saving 35.7% in area. The research is conducted using the open-source CGRA-ME (modeling and exploration) framework [1].

Fast placement and routing by extending coarse-grained reconfigurable arrays with Omega Networks

While parallel architectures based on clusters of Processing Elements (PEs) sharing L1 memory are widespread, there is no consensus on how lean their PE should be. Architecting PEs as vector processors holds the promise to greatly reduce their instruction fetch bandwidth, mitigating the Von Neumann Bottleneck (VNB). However, due to their historical association with supercomputers, classical vector machines include micro-architectural tricks to improve the Instruction Level Parallelism (ILP), which increases their instruction fetch and decode energy overhead. In this paper, we explore for the first time vector processing as an option to build small and efficient PEs for large-scale shared-L1 clusters. We propose Spatz, a compact, modular 32-bit vector processing unit based on the integer embedded subset of the RISC-V Vector Extension version 1.0. A Spatz-based cluster with four Multiply-Accumulate Units (MACUs) needs only 7.9 pJ per 32-bit integer multiply-accumulate operation, 40% less energy than an equivalent cluster built with four Snitch scalar cores. We analyzed Spatz' performance by integrating it within MemPool, a large-scale many-core shared-L1 cluster. The Spatz-based MemPool system achieves up to 285 GOPS when running a 256x256 32-bit integer matrix multiplication, 70% more than the equivalent Snitch-based MemPool system. In terms of energy efficiency, the Spatz-based MemPool system achieves up to 266 GOPS/W when running the same kernel, more than twice the energy efficiency of the Snitch-based MemPool system, which reaches 128 GOPS/W. Those results show the viability of lean vector processors as high-performance and energy-efficient PEs for large-scale clusters with tightly-coupled L1 memory.

Coarse-grained reconfigurable array (CGRA) overlays can improve dataflow kernel throughput by an order of magnitude over Vivado HLS. This is possible with a combination of carefully floorplanned high-frequency (645-768 MHz) design and a scalable, communication-aware compiler. Our 2D torus CGRA architecture supports versatile processing element (PE) functionality and a configurable number of communication channels to match application demands. Compared to recent FPGA overlays like CGRA-ME's ADRES and HyCUBE implementations, our design operates at 1.8-3.4× faster clock frequency, while scaling to an orders-of-magnitude larger array size of 19×69 on Xilinx Alveo U280. Our communication-aware compiler targets HLS objectives such as initiation interval (II) and minimizes communication cost using an integer linear programming (ILP) formulation. Unlike SDC schedulers in FPGA HLS tools, we treat data movement as a first-class citizen by encoding the space and time resources communication network of the overlay in the ILP formulation. Given the same constraints on operational resources as Vivado HLS, we can retain our target II and achieve up to 9.2× higher frequency. Our ILP scheduler outperforms a PathFinder space-time router implementation in quality of result by up to nearly 2×.

In this paper, we propose a software shared virtual memory (SVM) system for heterogeneous multicore accelerator clusters with explicitly managed memory hierarchies. The target cluster consists of a single manager node and many compute nodes. The manager node contains a generalpurpose processor and larger main memory, and each compute node contains a heterogeneous multicore processor and smaller main memory. These nodes are connected with an interconnection network, such as Gigabit Ethernet. The heterogeneous multicore processor in each compute node consists of a general-purpose processor element (GPE) and multiple accelerator processor elements (APEs). The GPE runs an OS and the multiple APEs are dedicated to compute-intensive workloads. The GPE is typically backed by a deep on-chip cache hierarchy and hardware cache coherence. On the other hand, the APEs have small explicitly-addressed local memory instead of caches. This APE local memory is not coherent with the main memory. Different main and local memory units in the accelerator cluster can be viewed as an explicitly managed memory hierarchy: global memory, node local memory, and APE local memory. Since coherence protocols of previous software SVM proposals cannot effectively handle such a memory hierarchy, we propose a new coherence and consistency protocol, called hierarchical centralized release consistency (HCRC). Our software SVM system is built on top of HCRC and software-managed caches. We evaluate the effectiveness and analyze the performance of our software SVM system on a 32-node heterogeneous multicore cluster (a total of 192 APEs).

As computation migrates to the edge, energy efficiency is becoming even more critical in the design of electronic devices. Therefore, new applications for the Internet of Things pose a challenge for designers to find the optimal balance between performance and energy consumption. While Coarse Grained Reconfigurable Architectures (CGRA) have been an alternative to provide the required acceleration, they are far from being the most energy-efficient, since their functional units may be underutilized by some applications. In this work, we introduce a novel approach to handle these CGRA resources, which leverages the variable usage of functional units due to the application program phases. By using our new approach and performing power-gating of under-utilized parts of the CGRA, we demonstrate that it is possible to achieve reductions of up to 52.3% in power dissipation and 75% in energy consumption.

This paper presents an accelerating face detection algorithm using Coarse Grained Reconfigurable Architectures (CGRA). Face detection algorithms usually use several stages of cascaded face detectors and require large amount of feature data, while general processors have small internal memory. Since the latency from external memory is much longer than internal memory, efficient use of memory is important to make face detection faster. In this paper, we do the first-stage of cascaded face detection process for every line using the feature data in the internal memory, and then do next stages for the candidates that pass the first-stage. In addition, by efficient use of software pipelining and vectorization, face detection process can be accelerated. The proposed method was implemented into CGRA@400MHz, and can run at 15.1@800×600 frame per second.

We present a co-design approach to establish redundancy schemes such as Dual Modular Redundancy (DMR) and Triple Modular Redundancy (TMR) to a whole region of a processor array for a class of Coarse-Grained Reconfigurable Arrays (CGRAs). The approach is applied to applications with mixed-criticality properties and experiencing varying Soft Error Rates (SERs) due to environmental reasons, e. g., changing altitude. The core idea is to adapt the degree of fault protection for loop programs executing in parallel on a CGRA to the level of reliability required as well as SER profiles. This is realized through claiming neighbor regions of processing elements for the execution of replicated loop nests. First, at the source code level, a compiler transformation is proposed that realizes these replication schemes in two steps: (1) replicate given parallel loop program two or three times for DMR or TMR, respectively, and (2) add appropriate error handling functions (voting or comparison) in order to detect respectively correct any single errors. Then, using the opportunities of hardware/software co-design, we propose optimized implementations of the error handling functions in software as well as in hardware. Finally, experimental results are given for the analysis of reliability gains for each proposed scheme of array replication in dependence of different SERs.