What's the input x for cudnn rnn?

The “input x” for cudnn rnn has a size of seqLength * inputSize * miniBatch sizeof(float).
Do anyone know the structure of x?
Is it batch_0, batch_1, batch_2, … batch_(seqLength - 1)? Each batch has a size of inputSize
miniBatch.

x : [step, batch, input_size] tensor

In cudnn_RNNsample.cu,it define dimA like:

dimA[0] = miniBatch;//batchsize
dimA[1] = inputSize;//input_dim
dimA[2] = 1;

I think it’s really strange.