1
edit
Changes
→Assignment 2
#Write cuda kernel to replace 'RCTAlgorithmBackprojection' in dll
#Integrate cuda kernel in to RabbitCTRunner
Using preprocessor directives the RabbitCTRunner was fitted with a token named GPU. If GPU was defined, our kernel along with all the necessary code will be compiled and used by RabbitCTRunner. This allowed us to have a small footprint on the changes applied and also allows us to use the flag to compile back the original RabbitCTRunner.
======Kernel======
<pre>
__global__
void Backprojection(float * gpu_matrix_result, RabbitCtGlobalData * r) // problem_size is r.L
#else
</pre>
=====Metrics=====
Original RabbitCTRunner:
<pre>$ RabbitCTRunner.exe LolaBunny.dll rabbitct_512_v2.rctd nocuda 128</pre>
=== Assignment 3 ===