Team Members: Wu, Xingxing      Zhang, Yi
Title: Improvement of CT Image Reconstruction Speed Using MMX/SSE2

                            Abstract

    The Image Reconstruction is a very important part for the CT
technology.  Its performance directly affects the image quality and the
speed of the scan.  Because of the huge amount of data needed to be
processed, many methods have been explored to increase the speed of CT
image reconstruction for the past decades, such as the design of
specific hardware, parallel algorithm running on the super computes. 
This project focus on the improvement of image reconstruction speed
using MMX/SSE2 technology without special requirements on the hardware. 
In this project, we first identify the critical parts of the algorithms
which are most computationally intensive by using the profiling tool. 
We choose to use SSE2 based on the analysis of data dynamic range and
the additional functions provided by SSE2.  The algorithm is updated to
best fit into the MMX/SSE2 technology.  Many optimization rules and
mapping methods, such as the data alignment, code scheduling, branch
removing, are used in the implementation to explore the potential of the
MMX/SSE2 architecture and gain the most performance improvement.  Our
implementation shows significant speedup over the pure c code
implementation.