The present invention relates to a GPGPU-based satellite image algorithm parallel optimization method capable of providing an algorithm processing result in quasi real-time. According to the present invention, the GPGPU-based satellite image algorithm parallel optimization method comprises the steps of: forming an NVIDIA VGA device and hardware; installing a normalized difference vegetation index (NAVID) CUDA SDK when the NVIDIA VGA device and the hardware are installed; determining whether a development language of a selected satellite image hydrologic algorithm is C++ language; separating and porting a module to be optimized to a C++ development language when the language, which is determined in the step of determining whether the development language of the satellite image hydrologic algorithm is the C++ language, is not made in the C++ development language; making an algorithm with the C++ development language in the case of a step of starting development of the algorithm itself in the step of determining whether the development language of the satellite image hydrologic algorithm is the C++ language; when the algorithm is completed by development language porting or the development language, removing a mutual dependent code for an algorithm core operation processing portion circulated in a real image, independently separating a shared memory, a shared various and the like, and separating a file input/output portion of a target algorithm; assigning a graphic processing unit (GPU) memory and copying a CPU memory to the GPU memory; performing conversion into a parallel processing kernel code for GPU processing after the copy to the GPU memory; completing parallel processing in the GPU to perform kernel code conversion and copy again the GPU memory to a CPU; and copying the GPU memory to the CPU and inputting a CUDA cancelation command to cancel a CUDA memory.
展开▼
