Here we transfer the data with a bigger tile size on dimension 0 to include zero-padding at the beginning and at the end:
adf::read_access(mtxin.out[0]) = adf::tiling(
{.buffer_dimension = {128, 2},
.tiling_dimension = {160, 1},
.offset = {-16, 0},
.tile_traversal = {
{.dimension = 0, .stride = 128, .wrap = 1},
{.dimension = 1, .stride = 1, .wrap = 2}}});
The tile traversal on dimension 0 is not mandatory. The second read access pattern will generate 2 reads from the share buffer in the following way:
Let’s diplay the output:
Utils/GetTiles.py doc_x86simulator_output/data/Output_22.txt 2Do 160 2 0
This give:
------------------------------------------------------
filename: doc_x86simulator_output/data/Output_22.txt
NCols: 160
NRows: 2
NLayers: 1
------------------------------------------------------
Static Tile Selection
Tile: 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 3 ... 124 125 126 127 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 128 129 130 131 ... 252 253 254 255 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
------------------------------------------------------------