abstract 4..6 HiPAcc xor API OpenCL CUDA xor TextureMemory Linear1D Linear2D Ldg Array2D LocalMemory xor Blocksize bs_32x1 bs_32x2 bs_32x4 bs_32x8 bs_32x16 bs_32x32 bs_64x1 bs_64x2 bs_64x4 bs_64x8 bs_64x16 bs_128x1 bs_128x2 bs_128x4 bs_128x8 bs_256x1 bs_256x2 bs_256x4 bs_512x1 bs_512x2 bs_1024x1 padding -> integer [padding >= 1] [padding <= 17] pixelPerThread -> integer [pixelPerThread >= 1] [pixelPerThread <= 4] Cp : HiPAcc [API] [Blocksize] [padding] [pixelPerThread] [OpenCL => (not Linear1D && not Linear2D && not Ldg)] [((bs_32x32 || bs_64x16 || bs_128x8 || bs_256x4 || bs_512x2) && (pixelPerThread = 3 || pixelPerThread = 4)) => not LocalMemory] [(bs_1024x1 && (pixelPerThread = 2 || pixelPerThread = 3 || pixelPerThread = 4)) => not LocalMemory] [Array2D => padding = 1]