CMPS 297S/396AA: GPU COMPUTING
ASSIGNMENT 3
In this assignment, you will implement a matrix-matrix multiplication kernel that uses shared memory
tiling. Your kernel is expected to work for any set of matrix dimensions so make sure to handle boundary
conditions correctly.
Instructions
1. Place the files provided with this assignment in a single directory. The files are:
main.cu: contains setup and sequential code
kernel.cu: where you will implement your code (you should only modify this file)
common.h: for shared declarations across main.cu and kernel.cu
timer.h: to assist with timing
Makefile: used for compilation
2. Edit kernel.cu where TODO is indicated to implement the following:
Allocate device memory
Copy data from the host to the device
Configure and invoke the CUDA kernel
Copy the results from the device to the host
Free device memory
Perform the computation in the kernel
3. Compile your code by running: make
4. Test your code by running: ./mm-tiled
If you are using the HPC cluster, do not forget to use the submission system. Do not run
on the head node!
For testing on different matrix sizes, you can provide your own values for matrix
dimensions as follows: ./mm-tiled <M> <N> <K>
5. You are also provided with a file called questions.txt which contains questions about the
assignment. Answer the questions in the file after implementing your kernel.
Submission
Submit your modified kernel.cu and questions.txt files via Moodle by the due date. Do not
submit any other files or compressed folders.
