added missing initialization of D buffer in 2mm

[libmld-dev]

The initialization of the D buffer used for the GPU kernel was missing.
The GPU (and also the CPU) version adds values to the D buffer:
D[i * nl + j] += tmp[i * nj + k] * C[k * nl + j];
This means the initial values of D does matter for the computation. The D buffer of the CPU versions get initialized in the init_array function, but the GPUs D buffer initialization was missing.
This has only a very small impact on the result (and performance) and is therefore not detected by the compareResults function.