Building FFTW for Bare-Metal Zynq

PLease Fork and Star if you really use it in your design.
Best regards.

This guide provides step-by-step instructions on how to build FFTW for bare-metal execution on the Xilinx Zynq SoC. There are three main libraries that can be used in zynq:
1- CMSIS-DSP
2- NE10
3- FFTW
Before that, there are other libraries that can be used on Zynq, each with its own pros and cons. For example, CMSIS-DSP supports a maximum FFT size of 4K points, while NE10 only supports FFT sizes that are powers of 2 (2^N). In contrast, FFTW offers high scalability and supports arbitrary FFT sizes.

Comparison: FFTW vs. CMSIS-DSP vs. NE10

Library	Precision	Platform	SIMD Optimized	Fixed-Point Support	Best Use Case
FFTW	Float, Double	General-purpose (Linux/Bare-metal)	✅ Yes (NEON)	❌ No	High-accuracy FFT on embedded systems
CMSIS-DSP	Float, Q15, Q31	Cortex-M/A (Microcontrollers & Zynq)	✅ Yes (NEON)	✅ Yes	Low-power embedded DSP (MCUs, real-time FFT),Less than 4k fft point and many other functions such as FIR and many kind of filtering
NE10	Float, Fixed	ARM Cortex-A (Zynq, mobile)	✅ Yes (NEON)	✅ Yes	High-performance FFT on ARM processors, Only power of 2 FFT point

Which One Should You Use?

• For general-purpose ARM-based FFT (bare-metal or Linux on Zynq) → FFTW
• For microcontrollers or low-power DSP tasks → CMSIS-DSP
• For high-performance ARM Cortex-A applications → NE10

---

Prerequisites for building FFTW

For Ubuntu/Linux Users

Install the required packages:

sudo apt update
sudo apt install -y gcc-arm-none-eabi make autoconf libtool

For Windows Users

If using WSL, install the same packages as above.
If using Windows natively, install the Xilinx Vitis SDK, which includes the ARM cross-compilation toolchain.

---

Step 1: Download FFTW Source Code

wget http://www.fftw.org/fftw-3.3.10.tar.gz
tar -xvzf fftw-3.3.10.tar.gz
cd fftw-3.3.10

---

Step 2: Set Up the Cross-Compiler

For being sure that the compiler have been installed.

sudo apt install --reinstall gcc-arm-none-eabi

---

Step 3: Configure FFTW for ARM Cortex-A9

Run the following command to configure FFTW for bare-metal execution:
**NOTE:**Remember to change --prefix=/home/milad/build to your address.

./configure \
    --host=arm-none-eabi \
    --enable-float \
    --enable-neon \
    --disable-threads \
    --enable-static \
    --disable-shared \
    --prefix=/home/milad/build \
    CFLAGS="-O3 -mcpu=cortex-a9 -mfpu=neon -mfloat-abi=hard -ffast-math -ffreestanding -DFFTW_NO_TIMER" \
    LDFLAGS="-lm --specs=nosys.specs"  

Step 4: Configure FFTW for ARM Cortex-A9

make
make install

**Step 5: Adding FFTW to you project **

Create a folder in your project and name it FFTW (or any name you prefer), then copy the files from the build directory into it, as illustrated in the following figures.

**Step 6: Setting the Linker and adding the library to your compiler **

Add the Library to Your Vitis (SDK) Project Open Xilinx Vitis or SDK and select your project.
Right-click your project → Click Properties.
Navigate to C/C++ Build → Settings.
Under ARM GNU Toolchain → Select Libraries.
Add the following paths: Library search path (-L) as depicted:

**Step 6: Add Library Path **

To make the Xilinx SDK aware of the library path, follow these steps:

Open Xilinx SDK and your project. In the Project Explorer,
right-click on your project and select Properties.

Navigate to C/C++ General > Paths and Symbols.

Under the Libraries tab:

Click Add External Library to add an external library or Add Library to add a library already available in the workspace. You can add the path to the library in the Library Search Path section (under Include paths). As depicted:

Step 7: Test Sample code

After running this code you will face an error: I will explain how to solve it.
Note: Remember to use fftwf since it is 32 bit floating point.

#include <fftw3.h>
#include <stdio.h>

int main() {
    int N = 8;
    fftwf_complex in[N], out[N];
    fftwf_plan p;

    p = fftwf_plan_dft_1d(N, in, out, FFTW_FORWARD, FFTW_ESTIMATE);

    // Set input values
    for (int i = 0; i < N; i++) {
        in[i][0] = i;  // Real part
        in[i][1] = 0;  // Imaginary part
    }

    fftwf_execute(p);
    fftwf_destroy_plan(p);

    // Print output
    for (int i = 0; i < N; i++)
        printf("out[%d] = %f + %fi\n", i, out[i][0], out[i][1]);

    return 0;
} 

Step 8: Solve the problem and enjoy

Creating a dummy function for _gettimeofday at the top of your code and try building it again.

void _gettimeofday(void){}; 

*Step 9: Enjoy

ENJOY !!!!!!!!!!!!!!!
ENJOY !!!!!!!!!!!!!!!
ENJOY !!!!!!!!!!!!!!!
ENJOY !!!!!!!!!!!!!!!

PLease Fork and Star if you really use it in your design. Best regards.

Author

M.Mohmmadi.B
Mr.Kolahiyan
Mr.Barazesh
M.M.Sayadi