I would like to ask if the vitis BLAS library GEMV supports hls ap_fixed type?
If it does, how do I use it correctly (template args, coding style..?)?
The situation is the gemv giving completely wrong output - not even because of the precision as it was completely different from the correct result (all csim, cosim, and hardware). Change the data type to C++ double and the gemv kernel works properly.
From what I understand, there are two implementations of the dot product method in xf:blas::DotHelper class. One of them use dot_dsp() for data types that are not float or double. That why I think it may work for fixed point as well. but so far, I have failed to make it work.
Hi everyone,
I would like to ask if the vitis BLAS library GEMV supports hls ap_fixed type?
If it does, how do I use it correctly (template args, coding style..?)?
The situation is the gemv giving completely wrong output - not even because of the precision as it was completely different from the correct result (all csim, cosim, and hardware). Change the data type to C++ double and the gemv kernel works properly.
From what I understand, there are two implementations of the dot product method in
xf:blas::DotHelperclass. One of them usedot_dsp()for data types that are not float or double. That why I think it may work for fixed point as well. but so far, I have failed to make it work.Best regards