Support for matrix exp

[kolia]

using CuArrays
n = 4
A = randn(n, n) |> cu
exp(A)

ERROR: ReadOnlyMemoryError()                         
Stacktrace:                                          
 [1] exp!(::CuArray{Float32,2}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v0.7/LinearAlgebra/src/lapack.jl:209                                                                            
 [2] exp(::CuArray{Float32,2}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v0.7/LinearAlgebra/src/dense.jl:508                                                                              
 [3] top-level scope at none:0                       

I get the same error without a stack trace for A \ cu(randn(n))

(This is using CuArrays#vc/0.7 (a55b12a) depending on GPUArrays#master, since CuArrays#master doesn't seem to support indexing A[i,j] that exp(m::AbstractMatrix) uses)

How much work would it take to support LinearAlgebra functions like these? If reasonable for a beginner like myself, any pointers to where to start?

Thanks!

[kshyatt]

Coming back here to update that exp(A) will throw a scalar indexing error but the ldiv example does in fact work nowadays! 🎉 . For anyone who wants to support exp, we could probably do so for at least a subset of matrix types (symmetric, diagonal, and co) using the existing cusolverDN wrappers.