/*=============================================
*  MapCG: MapReduce for CPU and GPU  ver 0.9
*=============================================*/

TOC:
1. What is MapCG
2. How to install & use
3. Structure of MapCG
4. Bug reporting


1. WHAT IS MAPCG
-----------------

  MapCG is an efficient MapReduce implementation that works on both CPU
  and GPU. With MapCG, one can write a MapReduce program and execute it
  on both CPU and GPU without any modification to the source code. The
  framework is provided as a library with source code. Programmers can
  write code that invokes this library, compile it with proper compiler
  and the resulting executable should be able to run on both CPU and 
  GPU.

  Currently, MapCG is tested on Linux only. Porting to other platforms
  should be easy, because we use OpenMP as the multi-thread programming
  modle. Any compiler that supports OpenMP should be able to compile 
  MapCG and run it on CPU. For GPU, we use CUDA as the language for GPU
  library. As a result, MapCG can only work with CUDA-enabled GPUs.
  Again, the GPU part of MapCG is tested on Linux using NVCC. But it 
  should also work for the Windows version of NVCC.


2. INSTALL & USE
-----------------

  The MapCG library is distributed as source code. No installation is
  required to use it. 

  To use MapCG, the programmer should write several source files 
  specifying the functions needed by the MapReduce framework, such as 
  map() and reduce(). For beginners, we suggest starting from the sample
  programs provided with this distribution. It should be easier to
  modify the sample programs into a new program than to write one from
  scratch.

  Once the program is written, you can compile it using the Makefile.
  A compiler with OpenMP support is required to compile the CPU code.
  The GPU code is always compiled, MapCG requires NVCC to be installed 
  on the machine. If the GPU implementation is not wanted, you can 
  remove the directory and modify the Makefile so that you don't have to
  install NVCC on your machine.

  The Makefile will compile the program and generate an executable. The
  executable can then be executed through command line.

  For the sample programs, there are several command line options to 
  control the execution. One important option is the -exe_mode option.
  exe_mode specifies where the computation should be performed. it 
  accepts 'CPU', 'GPU', and 'CPU_GPU'. For more information on all the
  available options, please refer to HMM/HMMSchedulerSpec.cpp.


3. STRUCTURE
-----------------

  The MapCG framework is composed of two parts: the CPULib, and GPULib.
  The CPULib is the implementation of MapReduce on CPU, and GPULib is
  the files related to GPU MapReduce. There are also files in HMM. They
  act as a wrapper for both the CPU and GPU MapReduce framework. Because


3. BUG REPORT
-----------------
  
  MapCG is a research project. There must be bugs in the code. But as
  a student, I am not able to devote all my time to testing. I would be
  very glad if you can find the bugs for me. If you are not familiar
  with OpenMP or CUDA, you can also send me the test case that would
  reproduce the error. I will try to fix the bug as soon as possible.

  Also, if you have suggestion as to how to improve the framework, 
  please let me know. I would be more than happy if you tell me you 
  are using MapCG in your work. :-)

  You can reach me at 
                chuntao.hong@gmail.com