1
edit
Changes
→Team Sonic
= Team Sonic =
[[Image:sonicteam.png|right|frameless|widthpx727| ]]
== Members ==
# [mailto:pvaaheeswaran@senecacollege.ca?subject=gpu610 Prasanth Vaaheeswaran]
== About ==
== Required Files ==
#[http://www5.cs.fau.de/research/projects/rabbitct/download/ RabbitCT]#[http://www5.cs.fau.de/fileadmin/Forschung/Software/RabbitCT/download/rabbitct_512-v2.rctd Dataset (Included problem sizes: 128, 256, 512)]
== Progress ==
=== Assignment 1 ===
=====Background=====
In assignment 1 we identified the 'RCTAlgorithmBackprojection' function located in the dll/shared library was taking up virtually 100% of the execution time. Our goal for this assignment was to off-load that logic to the GPU using cuda api, leaving everything else intact.
=====SummaryProcess=====
The work we did can be separated in to three basic steps:
#Compile base software from source (RabbitCTRunner.exe and dll)
Original RabbitCTRunner with volume size 128:
<pre>
C:\Users\pvaaheeswaran\Desktop\cuda rabbit>Rabbit.exe LolaBunny.dll rabbitct_512-v2.rctd original 128
RabbitCT runner http://www.rabbitct.com/
Info: using 4 buffer subsets with 240 projections each.
Running ... this may take some time.
(\_/)
(='.'=)
(")_(")
--------------------------------------------------------------
Quality of reconstructed volume:
Root Mean Squared Error: 38914.3 HU
Mean Squared Error: 1.51433e+009 HU^2
Max. Absolute Error: 65535 HU
PSNR: -19.5571 dB
--------------------------------------------------------------
Runtime statistics:
Total: 339.759 s
Average: 353.915 ms
</pre>
CudaRabbit, volume size 128.
<pre>
C:\Users\pvaaheeswaran\Desktop\cuda rabbit>CudaRabbit.exe LolaBunny.dll rabbitct_512-v2.rctd result 128/
RabbitCT runner http://www.rabbitct.com/
Info: using 4 buffer subsets with 240 projections each.
Running ... this may take some time.
(\_/)
(='.'=)
(")_(")
--------------------------------------------------------------
Runtime statistics:
Total: 0.004881 s
Average: 0.00508437 ms
FULL RUNTIME: 0.512 secs
C:\Users\pvaaheeswaran\Desktop\cuda rabbit>
</pre>
=====Graph=====
This graph represents for only 128 volume size.
[[Image:sonicgraph1.png|left|frame]]
=====Summary=====
As you can see, using cuda has improved the performance by ~99.85%, we went from over 5 minutes to under a second. However, this code still needs to be optimized and run on bigger volume sizes. This will be our next goal.
=== Assignment 3 ===
