Investigating the Use of the Intel Xeon Phi for Event Reconstruction

The physics goal of Jefferson Lab is to understand how quarks and gluons form nuclei and it is being upgraded to a higher, 12-GeV beam energy. The new CLAS12 detector in Hall B will collect 5-10 terabytes of data per day and will require considerable computing resources. We are investigating tools, such as the Intel Xeon Phi, to speed up the event reconstruction. The Kalman Filter is one of the methods being studied. It is a linear algebra algorithm that estimates the state of a system by combining existing data and predictions of those measurements. The tools required to apply this technique (i.e. matrix multiplication, matrix inversion) are being written using C$^{++}$ intrinsics for Intel's Xeon Phi Coprocessor, which uses the Many Integrated Cores (MIC) architecture. The Intel MIC is a new high-performance chip that connects to a host machine through the PCIe bus and is built to run highly vectorized and parallelized code making it a well-suited device for applications such as the Kalman Filter. Our tests of the MIC optimized algorithms needed for the filter show significant increases in speed. For example, matrix multiplication of 5x5 matrices on the MIC was able to run up to 69 times faster than the host core.