CS698 Selected Topics: GPU Cluster Programming MPI+CUDA

Submit homework on Canvas
See the deadlines posted on canvas. See the homework submission guidelines below.

Do not email your homework. Emailed homework will be discarded.
Homework will not be accepted after the due date. Submit it on time.

Homework 1 - Matrix multiplication with MPI point to point communication in C
Homework 2 - Matrix multiplication with MPI collective communication in C
Homework 3 - Matrix multiplication with MPI one-sided communication in C
Homework 4 - Write an MPI+Cuda program to find dot product in C/C++ and Cuda - must run on any number of machines with any problem size, Fill in the template files.
Homework 5 - Write an MPI+Cuda program to perform tiled matrix multiplication in C/C++ and Cuda - must run on any number of machines with any problem size. Fill in the template files.
Homework 6 - Write an MPI+Cuda program to perform convolution in C/C++ and Cuda. Fill in the template files.
Homework 7 - Write an MPI+Cuda program to perform radix sorting in C, C++ and Cuda
Homework 8 - Write an MPI+Cuda program to perform breadth first search program in C, C++ and Cuda
Homework 9 - Write an MPI+Cuda program convolutional neural networks program in C, C++ and Cuda

Homework submission guidelines: Follow the guidelines below. This example is for homework 1.

After finishing your homework, create a directory called hw1-lastname-firstname. For Jane Doe and homework 1, the directory name is "hw1-doe-jane".
Copy all your hw1 related files to hw1-doe-jane. For HW1, you have two files: template.c and Makefile.
Prepare tgz file: tar czf hw1-doe-jane.tgz hw1-doe-jane
Upload hw1-doe-jane.tgz on Canvas.

Checking homework: After the submission, your your homework will be checked as follows:

I will run a bash script to (1) untar, (2) compile, and (3) execute your code with n=128 to 2048 with np=1-8 on a cluster of four machines. I am using exactly the same machines as the ones I gave you. So if your code compiles and runs on your cluster, it _will_ run on my cluster.
If your code breaks my bash script, you will get zero. If all run successfully, you will get full credit. So make sure you submit a working version with correct directory names and file names. Do adhere to the convention above to make your life as well as my life simpler.
The command I will run is:
mpirun -np {1-8} -hostfile hosts.4 --map-by node template {128-2048}
Make sure again your hw executable name is template.