Declare kernel function k1 with input
stream operates on data type int32 that is named
in0 and circular 1d output buffer operates on data type int32 that is named out0.
void k1(input_stream<int32> *in0, output_circular_buffer<int32, adf::extents<OUTPUT_SAMPLE_SIZE>> & out0)
{
auto out0Iter = aie::begin_circular(out0);
for (int ind = 0; ind < OUTPUT_SAMPLE_SIZE; ++ind)
{
*out0Iter++ = readincr(in0);
}
}
Important: When kernels are located on
the same tile in the AI Engine, memory
addressing optimizations become possible. However, it is important to note that some
restrictions might arise due to this optimization. For instance, if K1 and K2 are
situated in the same tile and are connected through an output-to-input buffer, they
must use the same type of addressing. This means they can either opt for linear or
circular addressing, but they cannot have a mixture of both. However, it is worth
noting that this limitation will not be identified during X86 simulation because it
only emulates functional aspects of AI Engine tiles and memory. For more details on X86 simulation models, see Limitations in
AI
Engine Tools and Flows User Guide (UG1076).