Introduction

The first practice session is divided into several exercises which are relatively independent; do not remain stuck on a particular one but move on to the next one if needed.

The first 4 exercises are a “warm up”, and the last on is the first part of the Mandelbrot project that you will continue during the next session.

You should learn how to:

check your CUDA device properties;
run some kernels and manage data allocation;
build your code with compatibility in mind;
code some simple kernels.

Then, moving on to the Mandelbrot project, you will:

implement a computation intensive algorithm using CUDA;
start to think about how to compute an histogram in CUDA.

Part 1: Warming up

Device query: obtain important information about your CUDA hardware.
Host code for simple kernels: manage memory allocation and kernel launch.
Compilation and Runtime: build and run code for a particular architecture.
Kernel programming: code some simple kernels.

Each exercise set is stored in a dedicated directory with detailed instructions: please check the instructions.html files.

Part 2: Mandelbrot

As soon as you complete the warm-up exercises, you should move on to the “TP1” directory and start the Mandelbrot project. You are expected to complete:

exercise 1 (code the Escape time algorithm);
exercise 2 (pixel value transformation — map pattern);
and start thinking about exercise 3 (histogram computation — reduce pattern).

How to proceed

You should now be ready to start:

Connect to the GPU-enabled computer assigned to you,
upload your assignment files,
and start working on the TP0/ex1.

Good luck!

GPU Computing

Introduction

Contents

Part 1: Warming up

Part 2: Mandelbrot

How to proceed