6/13/14
Working with gpuLucas

New build tested under Ubuntu 14.04 with Cuda 6.5, on a GTX Titan GPU.
I've deleted VC project files; the only build I've tested this on is
the Linux system.

The code is up-to-date with new CUDA 6.5 features like unified memory
commands and CUFFT callbacks to do frequency domain squaring in same
call as forward-FFT.

Dependencies on obsolete CUTIL library have been eliminated.

It continues to use QD to compute the irrational-base convolution weights
in extended precision, to avoid catastrophic cancellation for non-power-of-2
signal-sizes.  See below for address to download qd.

The standard disclaimers apply (BSD license, that sort of thing, see below).
Strictly research code, with no guarantees of any sort.

I've included a subdirectory with some CUFFT convolution sizes and associated
timings, for GTX 480 and TESLA c2050, for size = 2**a * 3**b * 5**c * 7**d.
Stored as .py files so can be read in directly as python lists with no parsing.

(The python scripts won't work with any current code, but they're there.)

Good luck making sense of it.  I'm waiting on another order-of-magnitude
speedup in the hardware before I look at Lucas-Lehmer again, but I'm happy
to incorporate minor changes or see ideas swiped for GIMPS or other group
efforts.

Andrew Thall
Alma College
Summer 2014