From the doc of vx_khr_nn_1_2_1.pdf, The following equation is implemented:
outputs[j,k,i]=saturate(round(∑l(∑m,n(inputs[j+m,k+n,l]×weights[m,n,l,i]))+biasses[j,k,i]))
But The reference code equation is implemented:
outputs[j,k,i]=saturate(∑l(∑m,n(warp(inputs[j+m,k+n,l]×weights[m,n,l,i]))+biasses[j,k,i]))
From the doc of vx_khr_nn_1_2_1.pdf, The following equation is implemented: outputs[j,k,i]=saturate(round(∑l(∑m,n(inputs[j+m,k+n,l]×weights[m,n,l,i]))+biasses[j,k,i]))
But The reference code equation is implemented: outputs[j,k,i]=saturate(∑l(∑m,n(warp(inputs[j+m,k+n,l]×weights[m,n,l,i]))+biasses[j,k,i]))
Please help to check.