Chapter 08 2012 Elsevier Inc All rights reserved

Chapter 08 © 2012 Elsevier, Inc. All rights reserved. 1

Figure 8. 1 A 1 D convolution example, inside elements. © 2012 Elsevier, Inc. All rights reserved. 2

Figure 8. 2 1 D convolution, calculation of P[3]. © 2012 Elsevier, Inc. All rights reserved. 3

Figure 8. 3 1 D convolution boundary condition. © 2012 Elsevier, Inc. All rights reserved. 4

Figure 8. 4 A 2 D convolution example. © 2012 Elsevier, Inc. All rights reserved. 5

Figure 8. 5 A 2 D convolution boundary condition. © 2012 Elsevier, Inc. All rights reserved. 6

Figure 8. 6 A 1 D convolution kernel with boundary condition handling. © 2012 Elsevier, Inc. All rights reserved. 7

Figure 8. 7 A review of the CUDA memory model. © 2012 Elsevier, Inc. All rights reserved. 8

Figure 8. 8 A 1 D convolution kernel using constant memory for M. © 2012 Elsevier, Inc. All rights reserved. 9

Figure 8. 9 A simplified view of the cache hierarchy of modern processors. © 2012 Elsevier, Inc. All rights reserved. 10

Figure 8. 10 A 1 D tiled convolution example. © 2012 Elsevier, Inc. All rights reserved. 11

Figure 8. 11 A tiled 1 D convolution kernel using constant memory for M. © 2012 Elsevier, Inc. All rights reserved. 12

Figure 8. 12 A small example of accessing N elements and ghost elements. © 2012 Elsevier, Inc. All rights reserved. 13

Figure 8. 13 A simpler tiled 1 D convolution kernel using constant memory and general caching. © 2012 Elsevier, Inc. All rights reserved. 14