Performance Impact Applying Compression Format to Sparse Matrix on Kernel Polynomial Method Using GPU (ACM)
Kernel Polynomial Method (KPM) is an efficient method used for simulations of crystal lattice system in research field of condensed matter physics and chemistry. KPM involves matrix operations such as matrix-vector multiplication in which the storage format of the matrix has a great impact not only on the performance of KPM but also the memory consumption. This paper proposes an implementation of the KPM on the recent graphics processing units (GPU) where the CRS format is applied to the matrix. This paper also illustrates performance evaluation of the implementation of GPU and that of CPU. We also compare performances among the cases with/without the CRS format in KPM. The evaluation shows that the GPU-based implementation achieves several times better performance than the CPU-based one.
Paper available at ACM.