chaithyagr / gpunufft

gpuNUFFT - An Open-Source GPU Library for 3D Gridding with Direct Matlab Interface

Hi there 👋

I am Chaithya G R, a Doctral student working with Philippe CIUCIU at Neurospin, CEA Paris-Saclay. I work with Inria in the MIND team and Neurospin in the METRIC team

The topic of my PhD is deep learning for joint optimizing k-space sampling trajectory and MRI reconstruction, particularly non-Cartesian MRI. This made me interested in all things connected to:

• convex non-smooth optimization
• MRI and healthcare, particularly trajectory design, sequence development and reconstructions
• deep learning

I like to write TensorFlow, PyTorch codes not only to write neural networks but also to build other related functionalities. I particularly love to build onto algorithms to drastically speed up computations through GPU implementations or parallel computations, algorithmic development etc. Open source is definitely my motto.

gpunufft's Issues

Fix issues for fMRI

• Make sure to allow pinned memory pointer provided through python working, Check if the memory pointed is pinned memory already, in which case we don't pin and allocate all over again!
• Density compensation division on GPU.

Fix issues for CUDA 12

The textures are deprecated and fixing issues for them is useful for moving to higher CUDA versions

Write Cuda_array_interface

Setup the array interface for the input and output, allowing to setup the codes to work directly without moving to and fro from GPU