CUDA-fused Fourier ptychography (cuFPM)

This implements MATLAB + CUDA programming for the acceleration of Fourier ptychographic microscopy (FPM) reconstruction. FPM's forward and backward progress are all calculated purely by CUDA in "fullyfusedFPM.cu".

🔔 The codes were built based on MATLAB 2024b and CUDA v12.8, and were tested with a graphic card of NVIDIA RTX 3090 24 GB.

💥 The codes are only available for images with even numbers of pixels due to the implementation of the "fftshift" kernel. [cufftShift]

Acceleration ratio

The codes were tested on a personal desktop

• OS: WINDOWS 11 Pro x64, with 128GB RAM
  
• GPU: NVIDIA RTX 3090 24GB
  
• CPU: 12th Gen Intel(R) Core(TM) i9-12900K 3.2GHz
  

The following image shows the benchmarks. The cuFPM was compared against MATLAB + CPU or MATLAB + GPU. The raw image is of 21gigapixel, which can be obtained from here.

In FPM implementation, a total of 361 images were collected with 2048 by 2048 pixels, 16 bits. The reconstruction upsample rate is 8 so the resolution of the reconstructed image is 16384 by 16384 pixels. [dataset].

The reconstruction duration using the fused FP is 100s on average compared to a conventional MATLAB's gpuArray implementation, which takes about 300s. Acceleration is about 3 folds.

Implementation	Size of input images	Reconstructed size	Batch size	Execution duration (s)
MATLAB	2048 × 2048 × 361	16384 × 16384	26	800 s
MATLAB's gpuArray	2048 × 2048 × 361	16384 × 16384	26	310 s
MATLAB + CUDA (cuFPM)	2048 × 2048 × 361	16384 × 16384	26	100 s
cuFPM-v2	2048 × 2048 × 361	16384 × 16384	26	68 s
cuFPM-v2	2048 × 2048 × 361	16384 × 16384	36	54 s
cuFPM-v2	512 × 512 × 361	4096 × 4096	26	2.5 s
cuFPM-v2	1024 × 1024 × 361	8192 × 8192	26	10.3 s
cuFPM-v2	2048 × 2048 × 93	16384 × 16384	26	21 s

cuFPM-v2 is a pure CUDA C++ implementation of cuFPM.

Requirements

• An NVIDIA GPU; All shown results come from an RTX 3090.
• MATLAB 2024b
• CUDA v12.8
• Visual Studio 2022 Community
• Windows Kits 10.0.26100.0 (higher version may be available but not tested)

To build the codes

for codes:

The cuda codes are designed and implemented based on MATLAB c++ interfaces including "mex.h" and "mxGPUArray.h". The "mex.h" provides basic support for mex and to build the cuda codes. The "mxGPUArray.h" provides support for the array types of MATLAB.

To build the codes "fullyfusedFPM.cu", you will need "mexcuda" to run

mexcuda -lcufft fullyfusedFPM.cu

command in the command line of the MATLAB, at the root of the file "fullyfusedFPM.cu". "-lcufft" is a setting for "mexcuda" that tells the mex to use "cuFFT.h", the CUDA fast Fourier transform library [cuFFT].

The "mexcuda" needs several preconditions. First, to download Visual Studio 2022 community and add "cl.exe" into the environment variables. Then, to download Windows SDK. Third, when the first and second steps were done, the following should be added to the environment variables

C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.42.34433\bin\Hostx64\x64

Create a new system variable named "INCLUDE", and add the following to the list.

C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.42.34433\include;
C:\Program Files (x86)\Windows Kits\10\Include\10.0.26100.0\ucrt;
C:\Program Files (x86)\Windows Kits\10\Include\10.0.26100.0\um;
C:\Program Files (x86)\Windows Kits\10\Include\10.0.26100.0\shared;

Create a new system variable named "LIB", and add the following to the list.

C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.42.34433\lib\x64;
C:\Program Files (x86)\Windows Kits\10\Lib\10.0.26100.0\ucrt\x64;
C:\Program Files (x86)\Windows Kits\10\Lib\10.0.26100.0\um\x64;

When running mexcuda in MATLAB, one should first type

setenv("NVCC_APPEND_FLAGS", '-allow-unsupported-compiler')

in the MATLAB command line so that MATLAB can use the compiler of VS 2022 Community. Otherwise, MATLAB may pop up errors.

for codes_v2:

The codes_v2 folder contains a pure CUDA C++ implementation of FPM reconstruction with MATLAB as an interface for data reading/loading and checking.

• To build the CUDA codes, first, please make sure that all requirements mentioned above are met.
• Then run "build.m" in the "codes_v2/cuFPM/" folder. The MATLAB should create a subfolder named "mex_obj" and finish three compilations.
• The generated "cuFPM_pure.mexw64" is available to be called by MATLAB.

License and Citation

This framework is licensed under the BSD 3-clause license. Please see LICENSE.txt for details.

If you use it in your research, we would appreciate a citation via

@software{CUDA-fused FPM,
	author = {THU Hololab},
	license = {BSD-3-Clause},
	month = {2},
	title = {{CUDA-fused FPM}},
	url = {https://github.com/THUHoloLab/Fused-Fourier-Ptycho},
	version = {1.0},
	year = {2025}
}

Questions?

If you find any questions during the implementation, please feel free to open an issue. Thank you very much 😊!