In wgpu 0.7, this was a valid operation, the zero padding was done by hand. However in #872 this was removed as an optimization since buffers are zero padded, however the slice end was taken from buffer.slice which has to be aligned when mapped.