Device-Specific/ GEMM : Removed #pragma unroll for the KL loop

It seems to have a disastrous effect on AMD GPUs. Perhaps this pragma is beneficial on intel, though. We might consider to disable unrolling only if AMD is used. To investigate...