Hi all.
I'm not sure if this is a bug or whether I've misunderstood something, but there seems to be a problem with the definition of workspace sizes in dsyevd_gpu and dstedx.
In lines 262-267 of dsyevd_gpu.cpp the variable llwrk2 is defined and used as the input for lwork in the call to magma_dstedx on line 316. However, this value is too small unless lwork used to call magma_dsyevd_gpu is larger than 1 + 6n + 2*n^2.
Perhaps there is some reason why the lwork for dstedx has to be smaller than lwork for dsyevd_gpu, but if this is the case then perhaps the documentation can be updated to reflect this.
Dave
