|
onnx2versal
|
Scalar implementation for chunking 32-bit stream input into two separate streams, requires 2*OVERLAP <= OUT_W, (INP_W-OUT_W) % FIRST_STRIDE == 0 if OVERLAP > 0, requires OUT_W*LCNT - OVERLAP*(LCNT-1) <= INP_W, if OVERLAP <= 0, SplitTwo32bitStreams<f,10,64,22,1>::filter2 total = 676 SplitTwo32bitStreams<f,10,64,31,-1>::filter2 total = 812.
#include <split.h>
Public Member Functions | |
| void | filter (input_stream< TT > *in, output_stream< TT > *restrict out0, output_stream< TT > *restrict out1) |
Static Public Member Functions | |
| static void | registerKernelClass () |