HP-FFT-HLS

This is in-progress work for HP-FFT

Publication:

FCCM25: HP-FFT: A General High-Performance FFT Generator Using High-Level Synthesis

This repository contains High-Level Synthesis (HLS) implementations and automation scripts for evaluating HP-FFT (High-Performance FFT) designs using various configurations and FFT sizes.

Directory Structure

• common.tcl: Common Vivado HLS settings and scripts used across all configurations.
• env.sh: Environment setup script (modify this to match your local paths).
• run.sh: Main script for executing HLS and RTL synthesis.
• submit.batch: Batch submission script for running on clusters.
• n256/: Contains HLS source code for 256-point FFT.
• n1024/: Contains HLS source code for 1024-point FFT.
• results/: Stores generated synthesis results (both HLS and RTL).

FFT Configurations

Each FFT size directory (n256, n1024) includes subdirectories with different HLS implementations:

• original_C_style
• no_StagePipeline
• UF1
• UF2
• UF4
• UF8
• UF16
• UF32

Each subdirectory corresponds to a specific optimization approach or loop unrolling factor.

Getting Started

Step 1: Environment Setup

Modify paths in env.sh to match your local environment and tool paths.

Step 2: Specify Configuration

In run.sh, specify the desired configuration by editing the CONFIGS variable. For example:

CONFIGS=("1024 Baseline2_no_StagePipeline")

Step 3: Launch Execution

To start synthesis, run:

sbatch submit.batch

If you don't have SRUM cluster, just run ./run.sh directly

Results

After execution, all synthesis reports (both HLS and RTL) will be available under the build/ directory.

We have prepared synthesis reports under results/. You can also view the reports we have prepared, without having to run synthesis yourself.

Important Note:

• The RTL synthesis report counts BRAM usage as 18K per BRAM. However, in paper, BRAM is counted as 36K.