FFTLIB — C7^™ Fast Fourier Transform Library

Summary | Features | Kernels | Supported Devices | Setup Instructions | Build Instructions | Related Repos | Licensing | Contributions | Developer Resources

Summary

FFTLIB is TI's optimized Fast Fourier Transform library for the C7^™ DSP architecture. It provides a suite of kernels targeting the C7^™ vector and streaming engines (SE/SA), enabling high-throughput FFT and linear algebra on TI SoCs such as AM62D and AM275.

Each kernel ships in two variants selected at init time via the FFTLIB_FUNCTION_OPT / FFTLIB_FUNCTION_NATC enum in the kernel's InitArgs:

Suffix	Description
_cn	C-native reference (NATC) — portable, functionally correct, used for validation
_ci	C7™ intrinsic (OPT) — hardware-optimized, targets SE/SA streaming engines

Each kernel has a corresponding test driver (_d.c). Some kernels additionally include a Python reference implementation under test/fft_c7x/FFTLIB_idat_gen/FFTLIB_<kernel>/ for test vector generation.

Features

• FFT and IFFT kernels spanning 16-bit integer and 32-bit floating-point data types
• C7^™ streaming engine optimized — all kernels leverage SE/SA hardware for peak throughput
• Batched and single-transform variants — process one or many FFTs in a single kernel call
• Real-to-complex and complex-to-complex input/output support
• Portable reference implementations — _cn variants enable host-side validation and porting
• CMake-based build with presets for PC simulation and on-target deployment

Kernels

FFT — C7^™ SE/SA Kernels (fft_c7x)

Kernel	Description
FFTLIB_fft1d_i16sc_c16sc_o16sc	1-D FFT, 16-bit integer complex input/twiddle/output
FFTLIB_fft1d_i32f_c32fc_o32fc	1-D FFT, 32-bit float real input, complex output
FFTLIB_fft1d_i32fc_c32fc_o32fc	1-D FFT, 32-bit float complex input/twiddle/output
FFTLIB_fft1dBatched_i16sc_c16sc_o16sc	Batched 1-D FFT, 16-bit integer complex
FFTLIB_fft1dBatched_i32f_c32fc_o32fc	Batched 1-D FFT, 32-bit float real-to-complex
FFTLIB_fft1dBatched_i32fc_c32fc_o32fc	Batched 1-D FFT, 32-bit float complex-to-complex
FFTLIB_ifft1d_i32fc_c32fc_o32f	1-D IFFT, 32-bit float complex-to-real output
FFTLIB_ifft1d_i32fc_c32fc_o32fc	1-D IFFT, 32-bit float complex-to-complex output
FFTLIB_ifft1dBatched_i32fc_c32fc_o32f	Batched 1-D IFFT, 32-bit float complex-to-real
FFTLIB_ifft1dBatched_i32fc_c32fc_o32fc	Batched 1-D IFFT, 32-bit float complex-to-complex

FFT — C7^™+MMA DFT Kernels (fft_c7xmma, C7100 and C7120 only; not supported on AM62D C7504 or AM275 C7524)

Kernel	Description
FFTLIB_FFT_fft_ixX_cxX_oxX	Top-level DFT coordinator; dispatches to small, large, or high-radix sub-kernels
FFTLIB_FFT_dftSmall_ixX_cxX_oxX	DFT for small point counts via brute-force matrix multiply
FFTLIB_FFT_dftLarge_ixX_cxX_oxX	DFT for large point counts via brute-force matrix multiply
FFTLIB_FFT_highRadixDecompositions_ixX_cxX_oxX	DFT via high-radix FFT algorithm decompositions

Supported Devices

Device	C7™ Core
AM62D	C7504
AM275	C7524

Setup Instructions

Prerequisites

• Linux host (Ubuntu 20.04 or later recommended)

• CMake ≥ 3.16

• TI C7000 Code Generation Tools (ti-cgt-c7000) — install to ~/ti/

  Download from TI CGT C7000. Default expected path: ~/ti/ti-cgt-c7000_5.0.0.LTS/

  Or set the environment variable before building:

  export CGT7X_ROOT=/path/to/ti-cgt-c7000_<version>

Build Instructions

FFTLIB uses CMake presets to manage build configurations.

Available Presets

Preset	SoC	Platform
release-autotest-am62d-pc	AM62D	PC simulation + tests
release-autotest-am62d-target	AM62D	On-device + tests
release-autotest-am275-pc	AM275	PC simulation + tests
release-autotest-am275-target	AM275	On-device + tests
release-buildlib-am62d-pc	AM62D	PC (library only)
release-buildlib-am62d-target	AM62D	On-device (library only)
release-buildlib-am275-pc	AM275	PC (library only)
release-buildlib-am275-target	AM275	On-device (library only)

Quick Start — PC simulation (AM62D)

# Configure
cmake -S . -B build --preset=release-autotest-am62d-pc

# Build
cmake --build build -- -j$(nproc)

On-Target Build (AM62D)

cmake -S . -B build --preset=release-autotest-am62d-target
cmake --build build -- -j$(nproc)

Library-Only Build

cmake -S . -B build --preset=release-buildlib-am62d-pc
cmake --build build -- -j$(nproc)

Related Repos

• MCU+ SDK — SoC drivers and middleware for AM2x/AM6x devices

Licensing

This repository is licensed under the Apache License, Version 2.0. See LICENSE for the full text.

All source files carry an SPDX Apache-2.0 identifier.

Contributions

This repository is not currently accepting community contributions.

Bug reports and feature requests are welcome via TI E2E Community Forums.

---

Developer Resources

TI E2E™ design support forums