Closed-form Fibonacci on GPU

Using Swift and Metal this project implements Closed-form Fibonacci on a GPU

Requirements

This project uses apple/swift-arguments-parser

How to use

After building in Xcode, simply run in the command line. For one result at index use:

  Closed-form\ Fibonacci\ on\ GPU <index>

To do a range set a start:

  Closed-form\ Fibonacci\ on\ GPU --start <start_index> <index>

To run on the CPU instead include the CPU flag:

  Closed-form\ Fibonacci\ on\ GPU --cpu --start <start_index> <index>

Limitations

The CPU calculations (using the --cpu flag) are only accurate in the range (0, 70].

The GPU calculations are only accurate in the range (0, 33].

Improvements

This project could be improved by implementing a arbitrary-precision arithmetic. This would require a new data structure on both the Swift and MSL sides. And it would need a way to accurately pass the data between the CPU and GPU.

Notes

If the Metal compiler uses Fast Math (the -ffast-math flag—which it does by default) GPU calculations will not work.