Buffer Placement and Sizing for High-Performance Dataflow Circuits

Commercial high-level synthesis tools typically produce statically scheduled circuits. Yet, effective C-to-circuit conversion of arbitrary software applications calls for dataflow circuits, as they can handle efficiently variable latencies (e.g., caches), unpredictable memory dependencies, and irregular control flow. Dataflow circuits exhibit an unconventional property: registers (usually referred to as “buffers”) can be placed anywhere in the circuit without changing its semantics, in strong contrast to what happens in traditional datapaths. Yet, although functionally irrelevant, this placement has a significant impact on the circuit’s timing and throughput. In this work, we show how to strategically place buffers into a dataflow circuit to optimize its performance. Our approach extracts a set of choice-free critical loops from arbitrary dataflow circuits and relies on the theory of marked graphs to optimize the buffer placement and sizing. Our performance optimization model supports important high-level synthesis features such as pipelined computational units, units with variable latency and throughput, and if-conversion. We demonstrate the performance benefits of our approach on a set of dataflow circuits obtained from imperative code.

Leveraging Automatic High-Level Synthesis Resource Sharing to Maximize Dynamical Voltage Overscaling with Error Control

Approximate Computing has emerged as an alternative way to further reduce the power consumption of integrated circuits (ICs) by trading off errors at the output with simpler, more efficient logic. So far the main approaches in approximate computing have been to simplify the hardware circuit by pruning the circuit until the maximum error threshold is met. One of the critical issues, though, is the training data used to prune the circuit. The output error can significantly exceed the maximum error if the final workload does not match the training data. Thus, most previous work typically assumes that training data matches with the workload data distribution. In this work, we present a method that dynamically overscales the supply voltage based on different workload distribution at runtime. This allows to adaptively select the supply voltage that leads to the largest power savings while ensuring that the error will never exceed the maximum error threshold. This approach also allows restoring of the original error-free circuit if no matching workload distribution is found. The proposed method also leverages the ability of High-Level Synthesis (HLS) to automatically generate circuits with different properties by setting different synthesis constraints to maximize the available timing slack and, hence, maximize the power savings. Experimental results show that our proposed method works very well, saving on average 47.08% of power as compared to the exact output circuit and 20.25% more than a traditional approximation method.

SVM-based switching filter hardware design for mixed noise reduction in digital images using high-level synthesis tools

Impulse and Gaussian are the two most common types of noise that affect digital images due to imperfections in the imaging process, compression, storage and communication. The conventional filtering approaches, however, reduce the image quality in terms of sharpness and resolution while suppressing the effects of noise. In this work, a machine learning-based filtering structure has been proposed preserves the image quality while effectively removing the noise. Specifically, a support vector machine classifier is employed to detect the type of noise affecting each pixel to select an appropriate filter. The choice of filters includes Median and Bilateral filters of different kernel sizes. The classifier is trained using example images with known noise parameters. The proposed filtering structure has been shown to perform better than the conventional approaches in terms of image quality metrics. Moreover, the design has been implemented as a hardware accelerator on an FPGA device using high-level synthesis tools.

Dependency Graph-based High-level Synthesis for Maximum Instruction Parallelism

Performance optimization is an important goal for High-level Synthesis (HLS). Existing HLS scheduling algorithms are all based on Control and Data Flow Graph (CDFG) and will schedule basic blocks in sequential order. Our study shows that the sequential scheduling order of basic blocks is a big limiting factor for achievable circuit performance. In this article, we propose a Dependency Graph (DG) with two important properties for scheduling. First, DG is a directed acyclic graph. Thus, no loop breaking heuristic is needed for scheduling. Second, DG can be used to identify the exact instruction parallelism. Our experiment shows that DG can lead to 76% instruction parallelism increase over CDFG. Based on DG, we propose a bottom-up scheduling algorithm to achieve much higher instruction parallelism than existing algorithms. Hierarchical state transition graph with guard conditions is proposed for efficient implementation of such high parallelism scheduling. Our experimental results show that our DG-based HLS algorithm can outperform the CDFG-based LegUp and the state-of-the-art industrial tool Vivado HLS by 2.88× and 1.29× on circuit latency, respectively.