Does memory coalescing improve performance of cublas

Hi,
Is there extra performance gain for cublas lib by using cudaMallocPitch() to coalesce memory allocation?
Thanks
Bill