high performance computing on graphics processing units: hgpu.org

When L-systems are applied to large and detailed 3D objects, the inherent serial geometric interpretation limits the speed of image generation. To accelerate the interpreting procedure, a Graphic Processing Unit (GPU) based method utilizing Compute Unified Device Architecture (CUDA) is proposed in this paper. The focused approach involves two phases: first is a sequential scan on the state string generated from the derivation of L-system running on CPU; the second is the parallel computing on GPU with CUDA. The symbols in the state string are interpreted as turtle commands and graphic primitives whose turtle states depend on the matrix multiplication operation in sequential scan. Then with positions and directions covered in turtle states, the lines (cylinders) are generated and graphic primitives are transformed into turtle coordinate system using thousands of parallel threads in computing phase. Compared with other methods, the proposed method is more efficient.