Scribd
Upload
122 views

Example: 201201014-GPU-AS2: Assignments For GPU Programming Course/ Lab

The document provides instructions for 3 assignments for a GPU programming course. Students must submit optimized CUDA code for each assignment, along with a report following a specific format. Assignment 1 involves adding a constant to array elements on the GPU and profiling for different problem sizes. Assignment 2 is matrix multiplication on the GPU using naive and shared memory implementations. Assignment 3 implements numerical integration using the trapezoidal rule in serial and parallel. Reports must include analysis, observations, and performance curves. Code and presentations will be evaluated for each assignment.

Uploaded by

kiran
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
122 views

Example: 201201014-GPU-AS2: Assignments For GPU Programming Course/ Lab

The document provides instructions for 3 assignments for a GPU programming course. Students must submit optimized CUDA code for each assignment, along with a report following a specific format. Assignment 1 involves adding a constant to array elements on the GPU and profiling for different problem sizes. Assignment 2 is matrix multiplication on the GPU using naive and shared memory implementations. Assignment 3 implements numerical integration using the trapezoidal rule in serial and parallel. Reports must include analysis, observations, and performance curves. Code and presentations will be evaluated for each assignment.

Uploaded by

kiran
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 4

Assignments for GPU Programming Course/ Lab

This document will be updated regularly. All assignments have to be


submitted by a given deadline (some issues in accessing the servers
will be resolved shortly).
Submission of assignments: Create separate folders with the name for
each assignment in your google-drive and share the folder with the id:
[email protected] ([email protected])

Naming the folder (strictly follow this): ID + GPU +


Assignment No.

Example: 201201014-GPU-AS2
Also do the above for your previous assignment folder
(asgn-1) in your drive shared with csdaiict, it will
automatically reflect in csdaiict drive.
All the assignments have to be supplemented with a brief write-up or
ppt with the following details (wherever necessary):
1. Context:
Brief description of the problem.
Complexity of the algorithm (serial).
Possible speedup (theoretical).
Optimization strategy.
Problems faced in parallelization and possible solutions.
2. Hardware details: GPU model, no of cores, device properties. Compute
capability.
3. Input parameters. Output. Make sure results from serial and parallel
are same.
4. Nave implementation description. Possible improvements over nave
implementation.
5. Problem Size vs Time (Serial, parallel) curve. Speedup curve.
Observations and comments about the results.
6. If more than one implementation, curves for all algorithms in the same
plot.
7. Wherever necessary use log scale and auxiliary units.
8. Effect of block dimensions and grid launch on speedup.
9. Proper labeling of graphs.
10.
List of observations and conclusion under each table and
figure.

Assignments include submission of codes [optimized serial and parallel codes


(multiple versions if applicable) with necessary comments inside the code].
Also register at http://courses.daiict.ac.in.
Evaluation will be based on the submission of materials (reports +
code) and presentation in class. Date-wise presentation details will be
updated in the list-of-presentation file at-least one day in advance.

Assignment 1 (17th Aug)

Deadline:
26th August

1. Write a CUDA program that adds a number X to all elements of a one-dimensional array
A.
2. The elements of A and X should be single precision floating-point numbers.
3. Using the necessary timer calls, have your program report the time needed to copy data
from the CPU to the GPU, the time needed to add X to all elements of A in the GPU, and
the time needed to copy the data back from the GPU to the CPU.
4. The elements of A should be initialized with some value (not random). So that
comparison with serial code is possible.
5. Vary the number of elements in A from min of 1Million to the maximum number that can
be supported by single invocation of a GPU kernel in power of two steps, i.e., 1M, 2M,
4M, 16M, etc.
6. For every different array size, have your program print three time measurements: the time
required to copy A from the CPU to the GPU, the time taken by the kernel, and the time
required to copy the data from the GPU to the CPU.
7. The output should be reported in tabular form, like:
Elements(M) ; CPUtoGPU(ms) ; Kernel(ms) ; GPUtoCPU(ms)
Explore the following possibilities for profiling.

cutStartTimer(myTimer)

Events

8. Comment about CGMA ratio in the case of above program.


9. Extend your kernel specifying how many times X should be added to each element. Do
not use multiplication for these additions. Create a loop.
10. For the maximum number of elements that can be supported by a single kernel invocation
have your program print out the three time measurements above as a function of the
number of times X is added. Do so, for a range of 1 through 256 in power of two steps.
Your programs output should look as follows:
XaddedTimes; Elements(M); CPUtoGPU(ms) ; Kernel(ms); GPUtoCPU(ms)

What to submit/ report:

Submit the version that prints both measurements. (i.e., time as a function of element
count and time as a function of the number of additions).

Make sure it compiles and runs correctly.

Presentation of around 10 slides summarizing the above 10 points and other observations.
Supported by necessary curves.

Assignment 2 (26th Aug)

Deadline: September
2

Shared / shared CUDA Matrix Multiply


Write a serial CPU code for matrix multiplication (MM)
Parallelize the serial code (nave implementation of MM) using CUDA.
Modify the nave implementation using shared memory as discussed in the
class. (not tiled implementation, yet to be fully discussed in the class)
Investigate how each of the following factors influences performance of matrix
multiplication in CUDA (for both implementation nave and shared memory ):
(1) The size of matrices to be multiplied

(2) The size of the block computed by each thread block

In addition to 8 points discussed in page-1, also make comments about your observations and
the most optimized implementation.

Assignment 3 (Sep 2)
16

Deadline: September

Numerical Integration using trapezoidal rule as discussed in the class.


1. Serial implementation.
2. Parallel implementation using CUDA.
In addition to 10 points discussed in page-1, also make comments about your observations and
the most optimized implementation in the report.

You might also like