A Strategy for Automatic Performance Tuning of Stencil Computations on GPUs

Table 5

Breakdown by data loading technique (AMD).

Data loading technique	Speedup			Best count
	Dims	Opts	Hybrid	Dims	Opts	Hybrid
Global without vectorization	0.95	1.02	1.01	4	12	11
Global with vectorization	0.87	0.85	0.92	6	4	17
Image	0.56	0.58	0.58	0	0	0
Local	0.93	1.18	1.20	1	26	27