isabella232 / diffcuda

Accelerated bulk diff on GPU

Bulk text diff using NVIDIA CUDA

This is an ongoing effort to accelerate diff-ing of text files in a data parallel fashion. In other words, this project tries to improve on diff-ing many files at once and not increasing the speed of a single diff operation.

The approach chosen is as follows:

• Extract line endings.
• Hash each line.
• Apply Myers algorithm to hashes on the GPU.

Results

This project appeared to be 2x faster than a single-threaded libxdiff run. The profiling results showed that 25% of time is spent on line endings and hashing, which is done in OpenMP multithreaded mode.

License

MIT license.