The matrix-vector multiplication kernel requires the matrix data in a column-major format where the columns are stored contiguously in memory and TP_DIM_A_LEADING = 1.
Setting TP_DIM_A_LEADING to 0, indicates to the graph that the data is in a row-major format and DMA buffer descriptors will be to set the write access of the kernel input so that a transpose can be achieved. This feature is currently only compatible with int32, cint16 and float data types for a single frame per buffer NUM_FRAMES = 1. If multiple frames are required or the data type is int16, cint32 or cfloat, the matrix data must be transposed to a column-major format with the template parameter TP_DIM_A_LEADING being set to 1.