|Workload type||Lucas-Lehmer test|
CUDALucas is a sophisticated implementation of the IBDWT method for GPU. It has been developed by Yamada as a direct port of the highly optimized sequential MacLucasFFTW, itself a port to FFTW of a program tracing its roots back to original code by Richard Crandall et al.
To improve performance of CUDALucas, you need to run two benchmarks for your GPU. For example, like this:
CUDALucas -cufftbench 1 22680 5 CUDALucas -threadbench 1 22680 5 10
These commands will generate files
DEVICENAME fft.txt and
DEVICENAME threads.txt, enabling CUDALucas to select the parameters for an exponent being tested more efficiently.
To ensure that your program and GPU are working properly, you need to run self tests. For example, like this:
CUDALucas -r 1 CUDALucas -memtest 40 1 CUDALucas 1398269
-r 1 will run a thorough (long) self test. Option
-memtest 40 1 will run a short version of memory test for 1 GB (a few hours test is advised, however). The command
CUDALucas 1398269 will test M35 for primality.