CS 385 Lab 8, Spring 2009

In this lab, you will port your IncrementAll program to CUDA. You can use this lab to troubleshoot your CUDA setup at home, or you can use our server when it comes online. Keep watching this page for infrastructure updates.

Preparation

Be sure to set up your environment variables to link to the CUDA compiler and libraries. On my personal Mac, I added the following two lines to my ~/.bashrc file:

export PATH=/usr/local/cuda/bin:$PATH
export DYLD_LIBRARY_PATH=/usr/local/cuda/lib:$DYLD_LIBRARY_PATH

The default installation paths for CUDA will be different on Linux and Windows systems. If you do try CUDA at home, run some of the sample programs in the SDK (template, reduction, etc.) to make sure these tests pass.

Specification

You will need to write the following code in yourlastnameL8.cu, some of which can be copied from previous labs:

A function called InitArray that takes as input a numerical array of size N and initializes it with random values. You may constrain these values to be within some reasonable range.
A function called PrintArray that takes as input a numerical array of size N and prints every entry in the array and its index. You may comment this function out in your submitted code.
A CUDA kernel that takes as input a pointer to a numerical array of size N and increments elements of the array by 1.
A main function that does the following:
- Declares the array (host and device). Both arrays should be dynamically allocated; the host array can be allocated either with malloc or new, while the device array should be allocated with cudaMalloc.
- Calls InitArray to initialize the host array.
- Calls PrintArray to print the original values of the host array.
- Calls cudaMemcpy to copy the host array to the device.
- Calls the CUDA kernel to increment the elements of the array.
- Calls cudaMemcpy to copy the device array back to the host.
- Calls PrintArray to print the new values of the host array.
- Frees the memory of the arrays. The host array can be deallocated using delete or free, while the device array should be deallocated with cudaFree.

Check that your code works for values of N that are both large and small, as well as values that do not evenly divide the number of thread blocks or threads.

Compiling your code

You should be able to compile your code using nvcc -o outputfile sourcefile.cu -- if you run into problems other than syntax errors, the likely culprit is your environment variables.

Submitting your file

Make sure that your file is correctly named and then copy it to ~srivoire/cs385/submit. Wait 2 minutes, and then check that it was correctly submitted by visiting http://rivoire.cs.sonoma.edu/cs385/lab8sub.txt.