Does anyone know if theres a multiple precision arithmetic library for cuda?
How much precision would you need? What operations are needed? A sample implementation of double-double arithmetic was posted to our registered developer website recently. It implements negation, addition, subtraction, multiplication, division, and square root. double-double uses pairs of double-precision numbers to provide roughly quadruple precision, i.e. about 31 decimal places.
Precision which linux MPFR and GMP offers. Another words cuda software to multiply 1000 digit numbers, but faster than the current MPFR software which uses the cpu’s.
I gather you are looking for an arbitrary precision library for the GPU. I was not readily aware of one, but a quick internet search led me to the gpuprec project: Google Code Archive - Long-term storage for Google Code Project Hosting.. You might want to check that out to see whether it fits your needs.
Yeah. Arbitrary precision. Thanks. I hope that site has it. I’ve read papers from around 2009 that its 10X faster for large large numbers than the cpu method.