HiKonv: Maximizing the Throughput of Quantized Convolution With Novel Bit-wise Management and Computation

TL;DR

HiKonv introduces a novel approach to maximize convolution throughput in quantized CNNs by leveraging bit-wise management and parallel computation, significantly improving performance on CPU and FPGA hardware.

Contribution

It presents a unified framework and performance models for high-throughput low-bitwidth convolution using existing full-bitwidth units, enabling substantial hardware efficiency gains.

Findings

Single CPU 32-bit unit can perform multiple low-bitwidth convolutions simultaneously.

HiKonv achieves up to 7.6x performance improvement on CPU for 1-D convolution.

On FPGA, HiKonv enhances throughput and DSP efficiency, outperforming previous models.

Abstract

Quantization for CNN has shown significant progress with the intention of reducing the cost of computation and storage with low-bitwidth data representations. There are, however, no systematic studies on how an existing full-bitwidth processing unit, such as ALU in CPUs and DSP in FPGAs, can be better utilized to deliver significantly higher computation throughput for convolution under various quantized bitwidths. In this study, we propose HiKonv, a unified solution that maximizes the throughput of convolution on a given underlying processing unit with low-bitwidth quantized data inputs through novel bit-wise management and parallel computation. We establish theoretical framework and performance models using a full-bitwidth multiplier for highly parallelized low-bitwidth convolution, and demonstrate new breakthroughs for high-performance computing in this critical domain. For example, a single 32-bit processing unit in CPU can deliver 128 binarized convolution operations (multiplications and additions) and 13 4-bit convolution operations with a single multiplication instruction, and a single 27x18 multiplier in the FPGA DSP can deliver 60, 8 or 2 convolution operations with 1, 4 or 8-bit inputs in one clock cycle. We demonstrate the effectiveness of HiKonv on both CPU and FPGA. On CPU, HiKonv outperforms the baseline implementation with 1 to 8-bit inputs and provides up to 7.6x and 1.4x performance improvements for 1-D convolution, and performs 2.74x and 3.19x over the baseline implementation for 4-bit signed and unsigned data inputs for 2-D convolution. On FPGA, HiKonv solution enables a single DSP to process multiple convolutions with a shorter processing latency. For binarized input, each DSP with HiKonv is equivalent up to 76.6 LUTs. Compared to the DAC-SDC 2020 champion model, HiKonv achieves a 2.37x throughput improvement and 2.61x DSP efficiency improvement, respectively.