The hardware resources for 4x4 SVD are listed in Table 149. (Vivado result)
Engines | BRAM | DSP | Register | LUT | Latency | clock period(ns) |
SVD | 12 | 174 | 57380 | 38076 | 3051 | 3.029 |
The accuracy of SVD implementation has been verified with Lapack dgesvd (QR based SVD) and dgesvj (Jacobi SVD) functions. For a 2545-by-4 matrix, the relative error between our SVD and the two Lapack functions (dgesvd and dgesvj) is about \(1e^{-9}\)
Caution
The profiling resources differ a lot when choosing different chips. Here we use xcu250-figd2104-2L-e, with clock frequency 300MHz and the margin for clock uncertainty is set to 12.5%.