GPGPUに対する理論モデル

This thesis proposes a new theoretical model for GPGPUs(General Purpose on GPUs), which are designed for general purpose computing by using a specialized processing unit to accelerate computation for building and manipulating images. Unlike PRAMs(Parallel Random Access Machines), which are typical theoretical models for parallel computers, Several characteristics of GPGPU, such as bank conflicts of the shared memory access and the coalescing of the global memory are taken into accountin our model. We propose two variants of GPGPU models and verify the feasibility of our models using implementation of some parallel sorting algorithm.修士(工学)法政大学 (Hosei University