The following library provides high precision floating arithmetic in CUDA. The library was developed as part of my Ph.D. research, and unfortunately due to other commitments is no longer maintained.
Chapter 4 of my dissertation provides an overview of the library, some documentation on the APIs and algorithms and some performance data.
Requirements:
- Ubuntu Linux machine with a recent GPU card (Volta+)
- Install GMP
- Install MPFR
/usr/local/cuda/bin/nvcc -DGPU -arch=sm_70 fpa_test.c tests.c fpa.cu -lgmp -lmpfr -o test
./test
Please feel free to email me at [email protected], I'll respond as time permits.