porting ADDA to CUDA with IA

[Michel GROSS]

  Hello,

  I am working on an AI-assisted CUDA port of ADDA:

  https://github.com/michelgross34/adda-cuda

  The main objectives are:

  - automate the CUDA conversion process and reduce implementation errors;
  - accelerate ADDA on NVIDIA GPUs;
  - keep iterative-solver vectors resident on the GPU, avoiding CPU-GPU transfers during the iterative calculation;
  - provide single-precision versions with strongly reduced GPU-memory usage;
  - enable larger calculations, including preliminary tests on a (360^3) computational grid using an NVIDIA Titan V with
    12 GB of GPU memory;

  - add additional iterative solvers whose code originates from IFDDA.

  A similar AI-assisted CUDA porting effort is being developed for DDSCAT:

  https://github.com/michelgross34/ddscat-cuda

  Both projects aim to combine automatic AI-assisted porting, GPU acceleration, and reduced memory consumption. The implementations are still under development and require further validation against the original CPU codes before
  production use.

Best  

Michel