Draw the equivalent of Figure 4.14 for an

88 matrix multiplication with 2x2 tiling

and 44

tiling. Verify that the reduction in global

memory bandwidth is indeed

proportional to the

dimension size of the tiles.

N.o Noi N2,1 M0.0 Mo, Mo.2 Mos M20 MM,2M, Po2 Pos Pro Pu P12 P1,3 P20 P21P22 P23 P3.0 P31P3.2 P3.3 FIGURE 4.14 Tiling Mand N to utilize shared memory