1.1 V9.1.85 1.2 a) "GeForce GTX 1050 Ti", 6.1 b) (6) Multiprocessors, (128) CUDA Cores/MP c) 4040 MBytes d) 65536, 49152 bytes 2.3 squareArray: Used 5 registers, 328 bytes cmem[0] addArrays: Used 8 registers, 348 bytes cmem[0] 3.3 without memory operations: GPU - 1.85ms CPU - 49.15ms with memory operations: GPU - 3.34ms CPU - 50.72ms 3.4 (8,4,3) 1.5416 ms (8,8,3) 1.5615 ms (32,8,3) 1.84985 ms (32,4,3) 1.8749 ms (8,4,3) seems to be optimal 5.5 sigma runtime 1 16.40ms 3 123.16ms 5 323.29ms 7 622.526ms 6.4 memory runtime global 123.16ms shared 82.12ms texture 444.73ms 9.6 dt - large: the gradients and the image disappear N - large: the image gets an average value 9.7 the result for linear diffusion and gaussian convolution are the same 9.8 a) the resulting image is still blurred, however less than in gaussian case, edges are preserved b) the resulting image is not blurred, edges are preserved 12.2 reduce0 0.355ms cublasSasum 0.00013ms 13.3 global 0.3766ms shared 0.4127ms implementation did not improve the performance