|
onnx2versal
|
Scalar implementation streaming weights, requires weights stream to be padded from MxCxKxK to MxCx16, KH=KW < 5, requires bias to be shifted, i.e. tbias - tw.reshape(M,-1).sum(1) * X_zero_point, QLinearConvScalarStream<28,32,28,32,1,1,1,1,6,3> total = 776955 (output_window 682564)
#include <qlinearconv.h>
Static Public Member Functions | |
| static void | registerKernelClass () |