Abstract — Fast changing, increasingly complex, and diverse computing platforms pose central problems in scientific computing: How to achieve, with reasonable effort, portable optimal performance? We present SPIRAL that considers this problem for the performance-critical domain of linear digital signal processing (DSP) transforms. For a specified transform, SPIRAL automatically generates high performance code that is tuned to the given platform. SPIRAL formulates the tuning as an optimization problem, and exploits the domain-specific mathematical structure of transform algorithms to implement a feedback-driven optimizer. Similar to a human expert, for a specified transform, SPIRAL “intelligently ” generates and explores algorithmic and implementation choices to find the best match to the computer’s microarchitecture. The “intelligence” is provided by search and learning techniques that exploit the structure of the algorithm and implementation space to guide the exploration and optimization. SPIRAL generates high performance code for a broad set of DSP transforms including the discrete Fourier transform, other trigonometric transforms, filter transforms, and discrete wavelet transforms. Experimental results show that the code generated by SPIRAL competes with, and sometimes outperforms, the best available human tuned transform library code. Index Terms — library generation, code optimization, adaptation, automatic performance tuning, high performance computing, linear signal transform, discrete Fourier transform, FFT, discrete cosine transform, wavelet, filter, search, learning, genetic and evolutionary algorithm, Markov decision process I.

This paper presents a parameterized soft core generator for the discrete Fourier transform (DFT). Reusable IPs of digital signal processing (DSP) kernels are important time-saving resources in DSP hardware development. Unfortunately, reusable IPs, however optimized, can introduce inefficiencies because they cannot fit the exact requirements of every application context. Given the well-understood and regular computation in DSP kernels, an automatic tool can generate high-quality ready-to-use IPs customized to user-specified cost/performance tradeoffs (beyond basic parameters such as input size and data format). The paper shows that the generated DFT cores can match closely the performance and cost of DFT cores from the Xilinx LogiCore library. Furthermore, the generator can yield DFT cores over a range of different performance/cost tradeoff points that are not available from the library.

We present a domain-specific approach to representing datapaths for hardware implementations of linear signal transform algorithms. We extend the tensor structure for describing linear transform algorithms, adding the ability to explicitly characterize two important dimensions of datapath architecture. This representation allows both algorithm and datapath to be specified within a single formula and gives the designer the ability to easily consider a wide space of possible datapaths at a high level of abstraction. We have constructed a formula manipulation system based on this representation and have written a compiler that can translate a formula into a hardware implementation. This enables an automatic “push button ” compilation flow that produces a register transfer level hardware description from high-level datapath directives and an algorithm (written as a formula). In our experimental results, we demonstrate that this approach yields efficient designs over a large tradeoff space.

HW/SW partitioned implementations promise to offer the best of both worlds—the performance and efficiency of HW and the flexibility of SW. This remains an under-tapped paradigm due to its high design complexity, made worse by inadequate support in current tools and the lack of engineers trained in this design discipline. SPIRAL is a fully automatic design generation framework for the linear DSP transform domain.

This paper shows that there are fast Fourier transform (FFT) algorithms that work, for a fixed number of points, independent of the dimension. Changing the dimension is achieved by relabeling the input and the output and changing the “twiddle factors. ” An important consequence of this result, is that a program designed to compute the 1-dimensional Fourier transform can be easily modified to compute the 2-dimensional and 3-dimensional Fourier transform on the same number of points. 1