Application-Specific Arithmetic in High-Level Synthesis Tools

Abstract

This work studies hardware-specific optimization opportunities currently unexploited by high-level synthesis compilers. Some of these optimizations are specializations of floating-point operations that respect the usual semantics of the input program without changing the numerical result. Some other optimizations, locally triggered by the programmer thanks to a pragma, assume a different semantics, where floating-point code is interpreted as the specification of computation with real numbers. The compiler is then in charge to ensure an application-level accuracy constraint expressed in the pragma and has the freedom to use non-standard arithmetic hardware when more efficient. These two classes of optimizations are prototyped in the GeCoS source-to-source compiler and evaluated on the Polybench and EEMBC benchmark suites. Latency is reduced by up to 93%, and resource usage is reduced by up to 58%. CCS Concepts: • Computer systems organization → High-level language architectures; • Hard-ware→ Arithmetic and datapath circuits; Reconfigurable logic applications; Hardware description languages and compilation;

CuFP: An HLS Library for Customized Floating-Point Operators

“…Incorporating CuFP into other projects is straightforward, as users can simply include a header file, ensuring effortless integration and user-friendly implementation. Its platform independence and open-source nature promote transparency and ease of debugging while maintaining performance levels 1 .…”

“…This abstraction allows a focus on system-level functionality, leading to faster design iterations. Moreover, HLS tools automate various optimization processes, reducing manual effort and accelerating development cycles [1].…”

X-Ray Tomography Reconstruction Accelerated on FPGA Through High-Level Synthesis Tools

“…The HLS compilers implement different optimization based on their particular Intermediate Representation (IR), such as pipelining or expressing data-level parallelism. However, they suffer from a lack of support for many other unexploited FPGAspecific optimizations for arithmetic operations [30] for instance. Hence, harnessing FPGAs' full potential via HLS tools requires knowledge of their architecture and a significant effort to adapt the application because OpenCL is not performance-portable from one platform to another.…”

OpenCL FPGA Optimization guided by memory accesses and roofline model analysis applied to tomography acceleration

“…These compilers implement different optimization based on their particular intermediate representation (IR), such as pipelining or expressing data-level parallelism. However, they suffer from a lack of support for many other unexploited FPGA-specific optimizations for arithmetic operations [21] for instance. Hence, harnessing FPGAs' full potential via HLS tools requires knowledge of their architecture and a significant effort to adapt the application because it is not performance-portable.…”