作者: Masahito Yamane , Tsaifa Yu , Saritha Dwarakapuram
DOI:
关键词:
摘要: Systems and methods consistent with the invention relate to performing faster motion estimation through efficient use of General Purpose Graphic Processing Unit (GPGPU) as compute co-processor in a multi-processor architecture. Integer pel fractional algorithms for large block sizes may be performed on GPU, while smaller is central processing unit (CPU). In embodiments described herein, GPU-based integer are using kernels which designed so that multiple thread blocks can run concurrently multiprocessor.