---

Structure of a CUDA-C GPU-application

---

• The CUDA C-programming language
  • CUDA C:

    CUDA C = an extension of the C-programming language that enable programmers to write GPU-applications in the C-programming langauge The GPU-functions in a GPU-application will be expressed using C-programming language constructs

  ---

  ---

  ---

  ---

• Structure of a GPU-application
  • A GPU-application contains both:

    Statements in CPU-code Statements in GPU-code

    ---

  • The structure of a GPU-computer application program:

    Explanation:

    The program consists of The main( ) function: The main function is a CPU function (i.e.: executed by the CPU) The program will start executing at the main( ) function The main( ) function can call other functions in the program Other CPU functions (that I list in blue color): A CPU function is a "traditional" function that is executed by the CPU (You have learn to program CPU function in CS255 !!!) GPU functions (that I list in red color): A GPU function is a "vector processing" function that is executed by the GPU (You will learn to program GPU function in CS355 soon !!!)

  ---

  ---

  ---

  ---

• How to specify GPU-functions in CUDA
  • Defining GPU functions in CUDA-C:

    Syntax to define a GPU function: __global__ void-typed C-function definition __device__ void-typed C-function definition

    Explantion:

    When a C-function is preceeded by the keyword __global__ // 2 underscores preceeds and follows the word "global" the defined function is a GPU-function A GPU-function cannot return a value, i.e.: The return type of a GPU-function must be void !!!

    ---

    ---

  • Example:

    __global__ void GPU_function1( ... input-parameters ... ) { ... C-statements (will be executed by the GPU) ... } __device__ void GPU_function2( ... input-parameters ... ) { ... C-statements (will be executed by the GPU) ... }

  ---

  ---

  ---

  ---

• Difference between __global__ and __global__ GPU-functions
  • __global__ GPU-functions:

    A __global__ GPU-function can be called by a host (CPU) function and by another GPU (device) function

    ---

    ---

  • __device__ GPU-functions:

    A __device__ GPU-function can only be called by another GPU (device) function