It is difficult to say if PLASMA and MAGMA are going to merge into one project.
Right now both PLASMA and MAGMA have their strengths and weaknesses.
MAGMA is more capable of accelerating "standard" (LAPACK) algorithms, both linear systems and eigenvalues.
MAGMA is also GPU-centric, i.e., the principle is to use the GPU to the fullest.
PLASMA employed new algorithms, which we refer to as "tile algorithms", which utilize multicore CPUs better.
But this applies to linear systems and not so well to eigenvalues.
And PLASMA is CPU-centric, i.e., the principle is to use the CPUs to the fullest.
We are looking into accelerating PLASMA using a GPU and have very promising results.
So, I am afraid that the answer is that PLASMA and MAGMA are going to be separate project, each having its own niche.