bmpnegative
-----------

A simple program to create the negative of a bitmap image.
Implementations in C, CUDA, HIP and OpenCL.

Building on eX³
---------------

The following explanations assume that this repository is cloned into
$HOME/bmpnegative on eX³, and that you are logged into the default node
(srl-login1). If not, change the commands accordingly.

Currently, the ROCm toolchain can not be loaded via `module`, which makes
compilation a little tedious. The toolchain is installed on n004, so that node
will be used for compilation.

* bmpnegative-serial

  The only thing required to build the serial version is a C compiler.

    $ cd bmpnegative
    $ make serial


* bmpnegative-hip-amd

  To compile the HIP code the ROCm toolchain is required. It is installed on
  n004 (mi100q). Run the following commands to compile for AMD and produce the
  binary `bmpnegative-hip-amd`:

    $ rlogin n004
    $ cd bmpnegative
    $ make hip

* bmpnegative-hip-nvidia

  To compile the HIP code for NVIDIA the CUDA toolchain is needed, as well as
  the ROCm toolchain. This can be loaded using `module`. Run the
  following commands to produce the binary `bmpnegative-hip-nvidia`:

    $ rlogin n004
    $ cd bmpnegative
    $ module load cuda11.0/toolkit/11.0.3
    $ make hip-nvidia


* bmpnegative-opencl

  The OpenCL code is compiled using the ROCm version of OpenCL, i.e., the ROCm
  toolchain is needed. Just like with `bmpnegative-hip`, run the following
  commands to produce the binary `bmpnegative-opencl`:

    $ rlogin n004
    $ cd bmpnegative
    $ make opencl

* bmpnegative-cuda

  To compile the CUDA code, the only thing needed is the CUDA toolchain. To
  compile the code an produce the binary `bmpnegative-cuda`, run the following
  commands:

    $ cd bmpnegative
    $ module load cuda11.0/toolkit/11.0.3
    $ make cuda


Usage
-----

    ./bmpnegative-[serial|cuda|hip-amd|hip-nvidia|opencl] <input.bmp> <output.bmp>