Copyright 2012 Takatoshi Nakayama.

This file is part of the CUMP Library.

The CUMP Library is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at your
option) any later version.

The CUMP Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
License for more details.

You should have received a copy of the GNU Lesser General Public License
along with the CUMP Library.  If not, see https://www.gnu.org/licenses/.






			THE CUMP LIBRARY


CUMP [1] is a library for arbitrary precision arithmetic on CUDA [2],
operating on floating point numbers.  It is based on the GNU MP library
(GMP) [3], and its functions have a GMP-like regular interface.

CUMP is designed to be faster on NVIDIA GF100/GF110 GPUs than the GARPREC
library [4].  The performance advantage is achieved by using 64-bit fullwords
as the basic arithmetic type, by using a register blocking technique, and by
using "little endian" word order format.

CUMP is free software and may be freely copied on the terms contained in the
files COPYING.LIB and COPYING (most of CUMP is under the former, some under
the latter).



			OVERVIEW OF CUMP


CUMP can be a substitute of GMP if arbitrary-precision floating-point
arithmetic is necessary in CUDA.  There are functions for host codes and
for device codes in CUMP.

Functions for HOST codes:
  "Host code" means a code operating on CPUs.  These functions are C language
  functions.  Most of them belongs to the cumpf and cumpf_array class
  equivalent to the mpf class in GMP.  Some which communicate with GMP's
  `mpf_t' variables belong to the mpf class.

Functions for DEVICE codes:
  "Device code" means a code operating on GPUs.  These functions are CUDA
  language functions which have the __device__ qualifier.  Since all of them
  is in cump namespace (C++), in GPU kernels with "using namespace cump;",
  it is possible to use CUMP functions which have the same name with GMP.

If you want to use CUMP instead of GMP in your CUDA applications, write CUDA
codes roughly (don't consider cudaMemcpy), assuming that GMP operates on CUDA
at first.  In your host codes, declare your variables as `cumpf_t' instead of
`mpf_t' and `cumpf_array_t' instead of `mpf_t[]', and call the functions in
the cumpf or cumpf_array class instead of the functions in the mpf class.
In your device codes, declare parameters of your GPU kernels as `cump::mpf_t'
instead of `mpf_t' and `cump::mpf_array_t' instead of `mpf_t[]', and write
"using namespace cump;" to the beginning of your GPU kernels.

CUMP has a lack of functions and a lot of limitations compared with GMP.
For more information on how to use CUMP, please refer to the documentation.
(Sorry, the documentation is under writing. Please wait for a moment.)



			REPORTING BUGS

If you find a bug in the library, please tell me about it.
Also welcome your questions and suggestions.



			REFERENCE


1. T. Nakayama and D. Takahashi: Implementation of Multiple-
   Precision Floating-Point Arithmetic Library for GPU Computing, Proc. 23rd
   IASTED International Conference on Parallel and Distributed Computing and
   Systems (PDCS 2011), pp. 343--349 (2011).

2. NVIDIA Corporation: CUDA, https://www.nvidia.com/object/cuda_home_new.html

3. T. Granlund: GMP: The GNU Multiple Precision Arithmetic Library,
   https://gmplib.org/

4. M. Lu, B. He and Q. Luo: Supporting Extended Precision on Graphics
   Processors, Proc. Sixth International Workshop on Data Management on New
   Hardware (DaMoN 2010), pp. 19--26 (2010)