Implementation of algorithms with a fine-grained parallelism on GPUs